Universal, with honest boundaries.
Current development source: 49f2d152. The 0.12.0 changelog remains unreleased. Source inventory, delivered contracts and observed execution are different evidence.
96Canonical skills
69Agent role definitions
33Hook scripts
Consolidation removes overlapping entrypoints while retaining useful methods, saved work and provider contracts. Counts are not a quality score or a target of eighty. .claude/decisions/0034-native-workflow-consolidation.md · docs/migrations/2026-09-25-native-workflows.md
One selected work map and verified profile, content-bound review/QA, the shared Review Method and optional MARS panels. Missing mandatory evidence or independent review stays open. skills/review/references/method.md · skills/review/references/evidence.md
Documents: Word, PowerPoint, workbook, PDF and Visio retain staged boundaries. Removed format acceptance was not passed or waived. The PDF writer remains; no Lintel PDF reader verifies output text, pages or rendering. .claude/decisions/0033-remove-document-format-acceptance-and-pdf-reader.md
96 canonical skills
Current source metadata, not acceptance claims. Selected starter wrappers load canonical workflows, not the entire library.
| Skill / layer | Declared purpose | Current source |
|---|---|---|
| adr-new foundation | Use when a non-trivial decision needs recording to bootstrap a new Architecture Decision Record from the template, asking context-gathering questions to fill Status, Context, Decision, and Consequences. Reach for it whenever you've made or are about to make an architectural choice worth a durable record. | skills/adr-new/SKILL.md |
| analyze foundation | Use to check that PLAN and BUILD still match the approved DEFINE design — run when a plan was revised or a build deviated, to catch drift between what was approved and what shipped. Checks the DEFINE-PLAN-BUILD legs and authority alignment, writes a severity-classified report, and is re-runnable read-only at any cycle point. | skills/analyze/SKILL.md |
| audit foundation | Read the unified Lintel audit trail — surface .claude/runtime/audit/ (repo events) and ~/.lintel/audit/ (operator events) <category>.jsonl records with optional category / kind / since-days filters. Read-only. | skills/audit/SKILL.md |
| brief-forge foundation | Use when a workflow explicitly hands work across a boundary — spawning a subagent, transitioning a phase, passing to a cold executor, or taking operator input — to build a structured envelope and run the active pack's evaluators on it. | skills/brief-forge/SKILL.md |
| build foundation | Use to execute an approved plan in bounded work packages, preserving short task IDs and acceptance evidence while reviewing each package for spec compliance and quality. | skills/build/SKILL.md |
| capture foundation | Use after SHIP, at the end of a task, to make what was learned durable — updates lessons, drafts an ADR for any non-trivial decision, appends the evolution log, and reaffirms the cold-executor handoff trio against build evidence. The cycle's last step; carries continuity to the next session. | skills/capture/SKILL.md |
| catalog foundation | Use to discover Lintel skills and agents by name, purpose, category, voice or declared client support, or regenerate the committed skill catalog after frontmatter changes. | skills/catalog/SKILL.md |
| clean foundation | Use for an honest context-weight readout and a pause/fresh-session/resume path when the current session feels heavy. | skills/clean/SKILL.md |
| cli-fingerprint foundation | Use to identify the current CLI, desktop, IDE or cloud surface and inspect its actual tools without inferring capability from a vendor name. | skills/cli-fingerprint/SKILL.md |
| code-freeze foundation | Use to add, list or lift advisory session freeze paths; preserves project policy and never grants or removes host write permission. | skills/code-freeze/SKILL.md |
| code-review foundation | Use before landing a change to review just the diff — focused on the changed code only, lighter than a full engineering review. Reach for it when you have uncommitted or unmerged changes and want a correctness and quality pass before they land. | skills/code-review/SKILL.md |
| compliance-gate foundation | Compliance-gate aggregator — runs all gates the active pack declares (compliance.hooks) as ONE green/red verdict. Embarrassment protection for compliance (6.10). | skills/compliance-gate/SKILL.md |
| context-budget foundation | Show observed context capacity and usage where available, clearly labeled input estimates otherwise; --watch compares available observations with local warning thresholds. | skills/context-budget/SKILL.md |
| context-cool foundation | Exclude explicitly selected files from future context reads without claiming to remove already-sent conversation content. | skills/context-cool/SKILL.md |
| context-warm foundation | Use to load bounded files, topic-related sources, ADRs or prior sessions with safe selection and honest input-size estimates before reading. | skills/context-warm/SKILL.md |
| context-warm-customer foundation | Load customer-engagement repo state into context — their infrastructure-as-code, their CLAUDE.md, their ADRs, recent commits. | skills/context-warm-customer/SKILL.md |
| context-warm-from-url foundation | Retrieve bounded external reference text only through explicit host policy and redirect checks, preserving source provenance and trust boundaries. | skills/context-warm-from-url/SKILL.md |
| cross-check foundation | Use for an independent second opinion on a diff, plan, code or hypothesis through an actually available authorized reviewer. Retains actor identity, disagreements and shared content-bound evidence without requiring a particular vendor. | skills/cross-check/SKILL.md |
| cycle foundation | Use to run a real multi-step task through the full SENSE-to-CAPTURE pipeline, or a chosen subset of phases. Supports mode presets, hopping in at any phase, and a cost-estimate gate before BUILD. The default entry point for substantial work. | skills/cycle/SKILL.md |
| da foundation | Use for data-architecture depth — schema design, migrations, sharding and partitioning, query-pattern audits, retention policy, and analytics readiness. Reach for it when the work turns on the data model or storage. Runs full, loop, or single-capability, dispatching to the data agents and scoring against a rubric. | skills/da/SKILL.md |
| define foundation | Use to turn a rough idea or scoped request into a source-grounded design with outcomes, constraints, alternatives, risks and acceptance. Preserve selected work and prior approval; ask only unresolved material decisions. Strategy questions are optional. | skills/define/SKILL.md |
| design-dna foundation | Curated design knowledge + retrieval — BM25 search over 84 UI styles, 161 WCAG-audited palettes, 161 product reasoning rules, 73 font pairings, 99 UX guidelines and 16 per-stack rule files; composes complete design-system recommendations; resolves the active design profile (default anthropic-default); validates rendered output against the profile. Use before ANY visual decision — choosing style/palette/fonts, building pages or components, reviewing UI quality. Use proactively when a task changes how anything looks, feels, moves, or is interacted with. | skills/design-dna/SKILL.md |
| dh foundation | Use for devops and hosting depth — deployment plans, rollback strategy, observability specs, SLI/SLO budgets, capacity headroom, cost projection, and on-call playbooks. Reach for it when the work turns on how the system runs in production. Runs full, loop, or single-capability, dispatching to the ops agents and scoring against a rubric. | skills/dh/SKILL.md |
| diagnose foundation | Use when something is broken and you don't yet know why — drives a hypothesis-led investigation that builds a minimum repro, eliminates variables, and isolates the root cause. Reach for it for "why is this failing?" before attempting a fix. | skills/diagnose/SKILL.md |
| discover foundation | Use after DEFINE, before PLAN, to gather context before planning — maps the codebase, surfaces relevant ADRs and lessons, and identifies reusable patterns, agents, and skills for the chosen wedge. Read-only; produces the grounding PLAN needs so it does not reinvent or contradict prior decisions. | skills/discover/SKILL.md |
| doctor foundation | Use to diagnose source, target, profile and installed-file integrity while keeping actual host activation explicitly unverified. | skills/doctor/SKILL.md |
| eval foundation | Run the active pack's voice TEST against its voice CORPUS — per-cell accuracy → CALIBRATION.md. | skills/eval/SKILL.md |
| fix foundation | Use for a known bug with a clear fix path that needs to ship now — runs the abbreviated SENSE, BUILD, REVIEW, SHIP path and skips design, discovery, and planning. The hotfix shortcut; reach for it when the diagnosis is already done and only the fix remains. | skills/fix/SKILL.md |
| frontend-design foundation | Use to direct frontend design, advise on a design-system choice or compare bounded variants while preserving the shared design contract and actual rendering/review evidence. | skills/frontend-design/SKILL.md |
| frontend-design-review foundation | Use to review built UI or produced frontend designs with actual route/viewport evidence, six canonical advisory dimensions and the existing required-control contract. | skills/frontend-design-review/SKILL.md |
| frontend-motion foundation | Frontend design-director sub-skill — picks motion-language (GSAP/Lenis/Theatre/Rive/Motion-One) + scroll-trigger-config + smooth-scroll-config + key-animations-spec. Solo-invokable. | skills/frontend-motion/SKILL.md |
| frontend-shader foundation | Frontend design-director sub-skill — picks shader library (Paper Shaders / OGL / react-three-fiber / Lygia) + visual thesis + GLSL snippet references + perf-budget. Solo-invokable. | skills/frontend-shader/SKILL.md |
| frontend-style-extract foundation | Pattern-level extraction sister to generate-style-learn. Reads artifacts (URLs, screenshots, .tsx files) → extracts layout-grammar + motion-language + interaction-patterns + component-library-fingerprint + shader-thesis → writes ~/.lintel/brand/design-patterns/<name>/. Solo-invokable. | skills/frontend-style-extract/SKILL.md |
| frontend-typography foundation | Frontend design-director sub-skill — picks font-family-stacks + variable-axes-config + size-scale + line-heights + font-loading-strategy from brief. Solo-invokable. | skills/frontend-typography/SKILL.md |
| full-engineering-pass foundation | Use for a customer engagement or major release that needs the whole engineering picture at once — composes all five domain modules (architecture, data, security, devops, testing) in dependency order. One invocation produces architecture decisions, a data model, a security posture, an ops plan, and quality validation together. | skills/full-engineering-pass/SKILL.md |
| generate foundation | Use to produce a finished document or deliverable from a brief — drives outline, writing, design, and QA through to a built artifact in a chosen format such as slides, Word, web, or PDF. Reach for it when the ask is to generate a polished document rather than write code. | skills/generate/SKILL.md |
| generate-app foundation | Full-app scaffold-orchestrator. Reads frontend-design-spec.json + generates vite-react/next-app/svelte-kit project skeleton with motion/shader/typography wired up. Sister to generate-web — same family (rendering-engine), larger scope. | skills/generate-app/SKILL.md |
| generate-design foundation | Produce design-spec.json (per-format layout-mappings + palette + fonts + asset placements) from content.md. Shared content-pipeline sub-skill, solo-invokable. | skills/generate-design/SKILL.md |
| generate-docs foundation | Use to generate source-grounded reference documentation, customer guides or onboarding tutorials while retaining API behavior, examples, provenance and limitations. | skills/generate-docs/SKILL.md |
| generate-outline foundation | Produce outline.md (structured presentation/document skeleton) from a brief. Shared content-pipeline sub-skill, solo-invokable. | skills/generate-outline/SKILL.md |
| generate-pdf foundation | Produce a PDF through an available converter and accepted browser print operation, preserving source content; Lintel does not read the produced PDF, so its text, pages and visual rendering stay unverified. Staged document boundary: no accepted end-to-end format claim. | skills/generate-pdf/SKILL.md |
| generate-ppt foundation | Produce an editable PowerPoint deck through available native tools or pptxgenjs, retaining source detail in notes and inspecting actual rendered slides. Staged document boundary: no accepted end-to-end format claim. | skills/generate-ppt/SKILL.md |
| generate-qa foundation | Validate generated artifacts (any format) against brand, voice, readability, and structure standards. Auto-fixes where possible. Solo-invokable on any artifact (even non-generate-produced). | skills/generate-qa/SKILL.md |
| generate-style-learn foundation | Analyze .pptx/.docx/web-examples and extract a reusable style palette. v3.5 Phase 3 of the doc-generation-pipeline. | skills/generate-style-learn/SKILL.md |
| generate-visio foundation | ⚠ TEMPLATE ONLY — Slot for Visio diagram generation (architecture sketches, network topologies, process flows). Content not curated. AI generates fresh at invocation per L-001. Staged document boundary: no accepted end-to-end format claim. | skills/generate-visio/SKILL.md |
| generate-web foundation | Use to render a static HTML mockup, a profile-aware single-file page or a Next.js scaffold from a brief or the existing design/content contracts. | skills/generate-web/SKILL.md |
| generate-word foundation | Produce an editable Word document through available native tools or a declared library, preserving source detail and reporting actual inspection evidence. Staged document boundary: no accepted end-to-end format claim. | skills/generate-word/SKILL.md |
| generate-write foundation | Produce content.md (slide/section bodies + bullets + titles) and speaker-notes.md from outline.md. Applies voice corpus. Shared content-pipeline sub-skill, solo-invokable. | skills/generate-write/SKILL.md |
| generate-xlsx foundation | Produce an editable, source-backed workbook through available native tools, verifying formulas, actual recalculation, persisted caches and honest inspection limits. Staged document boundary: no accepted end-to-end format claim. | skills/generate-xlsx/SKILL.md |
| handoff-size-check foundation | Use before handoff to estimate the selected work-map artifacts and actual warming inputs against reported host headroom, keeping unknown capacity and advisory limits explicit. | skills/handoff-size-check/SKILL.md |
| hooks-status foundation | Reader for hooks.jsonl — per-hook observed records, override patterns and hooks with no observed record in a window, read through the structured event reader. Absence stays unobserved. | skills/hooks-status/SKILL.md |
| inspect foundation | Use to inspect a selected plan or repository through engineering, design and developer-experience lenses. Preserve original work and short-leaf acceptance, report concrete findings, and use shared content-bound evidence without treating scores as clearance. | skills/inspect/SKILL.md |
| instruction-parity-check foundation | Use to verify shared session protocol equality and client-entry links without overwriting project prose or confusing similarity with authority. | skills/instruction-parity-check/SKILL.md |
| jobs foundation | Use to see and steer in-flight Lintel jobs — list what's open, continue, replan, abort, or branch a job. The single source of truth for "what's open right now"; reach for it when you've lost track of running work or need to redirect it. | skills/jobs/SKILL.md |
| lessons-add foundation | Use after a correction, insight, or recurring pattern worth remembering to record it as a lesson the next session will read at startup. Reach for it whenever the operator corrects you or you discover something that should prevent the same mistake recurring. | skills/lessons-add/SKILL.md |
| lessons-promote foundation | Promote one ID-managed project lesson into an explicitly named Lintel work tree's scaffolding baseline, with recorded provenance and no implicit branch, commit or push. | skills/lessons-promote/SKILL.md |
| lessons-surface foundation | Use before or during a task for keyword-ranked lessons, the complete index or an exact lesson ID, without changing the memory store. | skills/lessons-surface/SKILL.md |
| maintenance foundation | Use for on-demand storage/context guidance, scoped path diagnostics and labeled usage estimates without treating partial logs as health or claiming model compaction. | skills/maintenance/SKILL.md |
| mars foundation | Use when a problem, plan, spec, implementation or review deserves a deliberate multi-model adversarial second look — runs a small, bounded panel of the latest distinct models (independent first pass, at most one challenge round), preserves dissent and never grants release clearance. | skills/mars/SKILL.md |
| migrations foundation | Read the installed-source migration catalog against the selected target, preserving overdue, unknown and historical recovery states. | skills/migrations/SKILL.md |
| orientator foundation | Phase 3 v4.0 — lightweight routing agent invoked at SENSE. Reads operator prompt + active pack's navigation policy, recommends workflow with confidence + reasoning. Mechanical-first with LLM escalation. | skills/orientator/SKILL.md |
| pack-create foundation | Use to create a blank, inherited, or cloned Lintel pack and validate it before activation. | skills/pack-create/SKILL.md |
| pack-list foundation | List configured-store, repository and installed-source packs with resolver precedence, validation results and the actual effective profile. | skills/pack-list/SKILL.md |
| pack-switch foundation | Use to explicitly switch the effective pack through the structured profile lifecycle, preserving required policy, configured paths and generation-bound recovery. | skills/pack-switch/SKILL.md |
| pack-validate foundation | Validate a pack before activation or after editing its manifest. Checks effective required fields and inheritance with the shared resolver, then reports schema and version compatibility limits. | skills/pack-validate/SKILL.md |
| pause foundation | Use before the context window fills up or before clearing the session to save the current state to a checkpoint file. Reach for it when a session is getting heavy and you want to preserve where you are so a later session can pick up exactly here. | skills/pause/SKILL.md |
| perf-mode foundation | Advise on bounded working sets, context observations and checkpoint strategy for heavy phases; never changes model capacity. | skills/perf-mode/SKILL.md |
| perfbench foundation | Measure performance — runtime, memory, cold-start — and detect regressions vs baseline. | skills/perfbench/SKILL.md |
| plan foundation | Use after DISCOVER, or standalone with an approved design, to produce the cold-executor trio (plan.md + spec.md + prompt.md). Keep short, verifiable leaves and group connected work into bounded packages for implementation and review, with operator approval before BUILD. | skills/plan/SKILL.md |
| profile-switch foundation | Inspect host install state and guide explicitly supported activation or owned snapshot recovery, without inventing plugin controls. | skills/profile-switch/SKILL.md |
| research foundation | Use when the operator wants to understand a domain, codebase or option space before committing to implementation. Runs the research-dive cycle — SENSE, DEFINE and DISCOVER — and stops with sourced findings, never BUILD or SHIP. | skills/research/SKILL.md |
| resume foundation | Use when returning in a fresh session to resume selected committed work, a cycle or a job, or read an owned checkpoint with --from without replacing current work or policy. | skills/resume/SKILL.md |
| review foundation | Use after BUILD, before SHIP, to adversarially review what was built — checks spec compliance, code quality, the active pack's compliance gates, and cross-artifact consistency. A P1 finding blocks SHIP until resolved. | skills/review/SKILL.md |
| role foundation | Use to take on or change a working role — activate one for a lightweight lens, turn it off, swap mid-session, apply its lens to an artifact, or load its full definition on demand. Reach for it when work would benefit from a specific role's perspective; one role is active at a time. | skills/role/SKILL.md |
| role-new foundation | Scaffold a new role file from template via guided interview — IDENTITY, COLD KNOWLEDGE, DECISION CRITERIA, OUTCOME LENS per phase. Evolve an existing role with --update <id> (targeted, sensitivity-aware edits). | skills/role-new/SKILL.md |
| roles-list foundation | List all available roles (public + private, if accessible). Shows id, display name, scope, sensitivity, last-updated. | skills/roles-list/SKILL.md |
| safe-install foundation | Protect explicitly owned installation files with verified snapshots and conflict-preserving restore; announce recovery paths and refuse unowned rollback. | skills/safe-install/SKILL.md |
| sc foundation | Use for security and compliance depth — threat models, auth flows, secret management, dependency-security audits, compliance evidence, and incident runbooks. Reach for it when a change has a security or regulatory surface. Runs full, loop, or single-capability, dispatching to the security agents and scoring against a rubric. | skills/sc/SKILL.md |
| scaffold foundation | Use to initialize or inspect repository foundations through the owned scaffold helper, preserving existing instructions, memory and explicit profile boundaries. | skills/scaffold/SKILL.md |
| scaffold-internal-tool foundation | Create a working internal CLI, service, dashboard or script using the common owned foundation initializer and the project's chosen toolchain. | skills/scaffold-internal-tool/SKILL.md |
| scaffold-mvp foundation | Initialize an MVP around its real users and first working journey, sharing the owned foundation helper and explicit policy, evaluation and deployment boundaries. | skills/scaffold-mvp/SKILL.md |
| scope foundation | Use after SENSE, before DEFINE, when a request's size is ambiguous — turns a raw ask into a sized, disambiguated scope via the scale-estimator, asks one clarifying question only when the work could be small or large, overrides a confidently-wrong route, and emits scope.md. Light and read-only; stays silent on clearly small work. | skills/scope/SKILL.md |
| sense foundation | Use at the very start of a task to read the situation before deciding how to work — detects operator intent, the active pack's compliance mode, the active role, any 00-state left by a prior session, and recommends a workflow mode. The cycle's first step, before SCOPE; a lightweight read-only diagnostic with no gates. | skills/sense/SKILL.md |
| ship foundation | Use after REVIEW passes, when reviewed work is ready to land, to open a PR, deploy, or hand off to a customer — runs the active pack's final compliance hard-stops, voice gates on customer-facing artifacts, and pack-configured CI/deploy validation. The cycle's last gate before code leaves the repo. | skills/ship/SKILL.md |
| skill-new foundation | Use to turn an authorized recurring task or pattern into a new Lintel skill draft using the existing scaffold and frontmatter contract. | skills/skill-new/SKILL.md |
| skill-router foundation | Semantic skill router — given free-text user intent, suggests top 3 matching Lintel skills with rationale. | skills/skill-router/SKILL.md |
| spec-kit foundation | Use when a repository has GitHub Spec Kit artifacts and needs Lintel planning, build-card execution, review or session continuity. Reuses the existing constitution, spec, plan and tasks without creating competing specifications. | skills/spec-kit/SKILL.md |
| status foundation | Use to quickly check where you are in flight — shows what's open right now, an alias for listing jobs. The fast "what was I doing?" check at session start or any time you need orientation. | skills/status/SKILL.md |
| swarm foundation | Use when an approved plan has multiple dependency-independent work domains and the operator wants coordinated multi-agent execution with explicit ownership, attributable isolation, durable evidence, and honest serial fallback. | skills/swarm/SKILL.md |
| ta foundation | Use for technical-architecture depth — service boundaries, API contracts, dependency graphs, scaling plans, complexity audits, and quality attributes. Reach for it when a design needs architectural rigor beyond what PLAN gives. Runs full, loop, or single-capability, dispatching to the architecture agents and scoring against a rubric. | skills/ta/SKILL.md |
| tq foundation | Use for testing and QA-strategy depth — test-pyramid review, coverage audits, contract-test design, regression suites, flaky-test quarantine, perf budgets, and chaos plans. Reach for it when test strategy needs deliberate design rather than ad-hoc tests. Runs full, loop, or single-capability, dispatching to the test agents and scoring against a rubric. | skills/tq/SKILL.md |
| uniformity foundation | Read-only uniformity-contract dashboard — runs the Gate-M3 floor shape-test and points at the regenerable coverage matrix. Reports per-kind D14 necessity adoption. Never writes. | skills/uniformity/SKILL.md |
| usage-log foundation | Append-only usage log for skill/agent invocations — manual writer (one audit_log line) plus reader reports. One log, no per-skill duplicates (L-001). Solo-invokable for reports. | skills/usage-log/SKILL.md |
| verify foundation | Use to check whether a change works without changing it. Runs applicable tests or document checks, reports failures and preserves bound QA evidence. Repair requires an explicit --repair request within the authorized scope. | skills/verify/SKILL.md |
| web-session foundation | Use to browse, extract pages, open an owned browser or validate user-managed sign-in through an actual provider with explicit URL admission and lifecycle evidence. | skills/web-session/SKILL.md |
| welcome foundation | Use on first run to choose a useful task, inspect the actual client surface and take a proportionate plan, build, review or resume path. | skills/welcome/SKILL.md |
69 specialist role definitions
Expertise, scope and reporting. Presence is not native delegation. Independent review requires an attributable separate actor/context; role-play is self-review.
| Role / domain | Declared purpose | Current source |
|---|---|---|
| BlogPostDrafter communication | Drafts long-form blog posts in the pack's voice tier — engineering story, customer case, or POV piece. Use when a team has a public-worthy story, a customer has signed off on a case study, or a launch or industry moment needs a stated point of view. | agents/communication/BlogPostDrafter.md |
| EmailCustomerDrafter communication | Drafts customer-facing emails — intro, follow-up, decision-ask, scope-clarification — in the pack's voice tier. Use when a specific email to a customer needs careful framing, a difficult conversation needs tact, or a multi-recipient note must land for both exec and technical readers. | agents/communication/EmailCustomerDrafter.md |
| LinkedInPostDrafter communication | Drafts LinkedIn posts in three lengths — short hook, mid-tail story, long-tail POV — in the pack's voice tier. Use when an engineering moment, consented customer story, or industry POV is worth a public post. | agents/communication/LinkedInPostDrafter.md |
| SlideNarrationCritic communication | Critiques slide narration scripts for voice consistency, pacing, audience alignment, and recovery-line presence. Use when a demo script needs a rehearsal critique before a customer meeting, a DemoNarratorJunior draft needs review, or a flat demo needs a diagnosis. Talk track, words-per-minute pacing, speaker notes, recovery lines, deck voiceover. | agents/communication/SlideNarrationCritic.md |
| EUAIActReviewer compliance | Reviews AI systems against EU AI Act requirements — risk tier classification, obligations per tier, conformity assessment. Use proactively when an AI system targets the EU market, a risk-tier classification is needed pre-launch, or a general-purpose AI integration is being designed. | agents/compliance/EUAIActReviewer.md |
| GDPRReviewer compliance | Reviews data processing flows for GDPR compliance — legal basis, data minimization, retention, DSR mechanisms. Use proactively when an EU engagement needs a GDPR check, a new data-processing flow is designed, a cross-border transfer is involved, or DSR readiness needs an audit. | agents/compliance/GDPRReviewer.md |
| SOC2Reviewer compliance | Reviews controls against SOC 2 Trust Service Criteria (Security, Availability, Processing Integrity, Confidentiality, Privacy). Use proactively before a SOC 2 audit for gap analysis, when a new service enters audit scope, or for a periodic controls health-check. | agents/compliance/SOC2Reviewer.md |
| CustomerEmpathyCheck customer | Customer-empathy review of draft comms — does this read like a human cares? Use before a customer email, follow-up, or escalation response lands, especially when the customer is frustrated or vulnerable. | agents/customer/CustomerEmpathyCheck.md |
| DemoNarrativeArc customer | Reviews customer demo scripts against narrative principles — opening, escalation, payoff, close. Use before a demo script goes to the voice gate, or when an existing demo lands flat and the cause feels structural. | agents/customer/DemoNarrativeArc.md |
| DemoNarratorJunior customer | Drafts narration scripts for customer demos — segment-by-segment, audience-aware, with "what to say if X breaks". Use after DemoNarrativeArc has set the arc, or before a rehearsal that needs words to practice. | agents/customer/DemoNarratorJunior.md |
| ExecutiveBriefingDrafter customer | Drafts 1-pager executive briefings for customer C-suite — outcome-focused, in the pack's voice tier. Use before a customer exec meeting that needs a pre-read, or after one that needs a recap memo. | agents/customer/ExecutiveBriefingDrafter.md |
| PostDemoFollowup customer | Use after a customer demo to plan the follow-up — what to send, when, and which expansion paths to open. Post-demo follow-up, follow-up cadence, expansion paths, 30-day plan, demo outcome, deal progression. | agents/customer/PostDemoFollowup.md |
| ProposalDrafter customer | Drafts customer engagement proposals — scope, deliverables, timeline, success criteria — from intake brief. Use after a sales intake brief lands, or before a customer scoping conversation that needs a draft to anchor it. | agents/customer/ProposalDrafter.md |
| RFPResponseDrafter customer | Drafts structured RFP responses — point-by-point coverage of customer requirements with proof points. Use after a customer RFP or RFI arrives, or before submission to sanity-check coverage. | agents/customer/RFPResponseDrafter.md |
| DevOpsToolchain devops | Designs the build, ship, and run infrastructure — CI/CD, containers, Kubernetes, observability, incident runbooks. Use when a repo needs CI/CD set up, a container or manifest designed, an observability strategy chosen, or a deploy approach decided. Keywords — DevOps toolchain, pipeline, Docker, OpenTelemetry, Prometheus, canary, blue-green, SRE. | agents/devops/DevOpsToolchain.md |
| GHActionsReviewer devops | Reviews GitHub Actions workflows for security, performance, and best practices — pinned actions, secrets scope, permissions. Use when a new or changed workflow is up for review, a supply-chain attack vector through an action is suspected, or CI security needs a periodic sweep. | agents/devops/GHActionsReviewer.md |
| K8sManifestReviewer devops | Reviews Kubernetes manifests — resource limits, security contexts, network policies, secrets, ingress. Use when a manifest or Helm chart is up for review, a workload is being containerized, or a cluster's workloads need a resource or security audit. | agents/devops/K8sManifestReviewer.md |
| PerformanceAnalyzer devops | Profiles, identifies bottlenecks, suggests optimizations with data — never guesses. Use after a benchmark flags a regression, when reported slowness needs measurement before any code change, or when an optimization is being weighed and you need to know where time actually goes. | agents/devops/PerformanceAnalyzer.md |
| TerraformReviewer devops | Reviews Terraform configurations — state management, module structure, provider versions, security. Use when a Terraform change is up for review, a new module has been written, or a state migration or refactor is about to run. | agents/devops/TerraformReviewer.md |
| PPTNarrativeArchitect doc-gen | Designs the slide arc and per-slide content goal before a deck is built. Use before /li:generate-ppt runs, or when an existing deck needs a structural critique. | agents/doc-gen/PPTNarrativeArchitect.md |
| WebExperienceCritic doc-gen | Applies a 6-pillar UX and brand critique to generated web output. Use before /li:generate-web runs for structural recommendations, and after it produces output for a scored review. | agents/doc-gen/WebExperienceCritic.md |
| WordTechnicalEditor doc-gen | Reviews Word output for structure, factual accuracy, voice, and variant-specific quality. Use after /li:generate-word produces a doc, or before distributing a transparency-note variant. | agents/doc-gen/WordTechnicalEditor.md |
| AccessibilityChecker engineering | WCAG AA accessibility audit of UI components — contrast, semantics, keyboard, screen reader. | agents/engineering/AccessibilityChecker.md |
| ADRDrafter engineering | Drafts Architecture Decision Records in the repo's format — Status, Context, Decision, Consequences. | agents/engineering/ADRDrafter.md |
| APIDesigner engineering | Designs REST and GraphQL APIs — produces OpenAPI/GraphQL schemas, validates backward compat. Use proactively when a new API surface is being designed, endpoints are being added to an existing API, or a significant change to an existing endpoint needs a backward-compat check before implementation. | agents/engineering/APIDesigner.md |
| Architect engineering | Designs new components before implementation — produces design docs, sequence diagrams, interface definitions. Use proactively when a non-trivial feature is about to be built, an area is being refactored at architectural scale, or a cross-cutting structural decision needs options before code is written. | agents/engineering/Architect.md |
| BackendArchitect engineering | Backend systems architect — API design, microservices, distributed patterns, resilience, observability. | agents/engineering/BackendArchitect.md |
| CapacityPlanner engineering | Capacity modeling + bottleneck identification + cost projection. Produces per-component throughput/latency/resource projections against a scaling target. Spawned by TA module's scaling-plan capability. | agents/engineering/CapacityPlanner.md |
| ChangelogMaintainer engineering | Maintains CHANGELOG.md in Keep-a-Changelog format from git history. | agents/engineering/ChangelogMaintainer.md |
| CodeReviewer engineering | Reviews code changes for correctness, quality, security, and convention adherence. Use proactively when a non-trivial diff is staged for ship, a hot-path or security-sensitive change needs a second opinion, or a cross-cutting refactor spans many files. | agents/engineering/CodeReviewer.md |
| ContextBudgetAdvisor engineering | Suggests phase declarations + warmup patterns for unstructured tasks. | agents/engineering/ContextBudgetAdvisor.md |
| ContractTestArchitect engineering | Consumer-driven contract test design + schema-versioning tests. Designs Pact-style or framework-internal tests covering consumer expectations + provider verification + version compatibility matrix. Spawned by TQ module's contract-test-design capability. | agents/engineering/ContractTestArchitect.md |
| CostAnalyzer engineering | Analyzes cloud-cost data — identifies top spenders, waste patterns, optimization opportunities with $ estimates. | agents/engineering/CostAnalyzer.md |
| DatabaseDesigner engineering | Database schema design, indexes, query optimization, migration safety — Postgres-focused with general principles. Use proactively when a new table or significant schema change is proposed, a slow query smells index-related, or a destructive migration is about to run against live data. | agents/engineering/DatabaseDesigner.md |
| DataPipelineDesigner engineering | Data pipeline architect — batch + streaming, dbt + Airflow + Spark, data warehouse patterns. | agents/engineering/DataPipelineDesigner.md |
| DebugForensics engineering | Hypothesis-driven debugging — designs minimum repro, eliminates variables systematically, finds root cause. Use when /diagnose stalls, a recurring bug keeps almost-fixing, a heisenbug reproduces only under specific load or timing, or a multi-component failure puts the symptom far from the cause. | agents/engineering/DebugForensics.md |
| DeploymentEngineer engineering | Deployment pattern reasoning. Blue-green vs canary vs rolling, traffic-cutover stages, feature-flag rollout strategy. Spawned by DH module's deployment-plan capability. | agents/engineering/DeploymentEngineer.md |
| DocWriter engineering | Generates and updates documentation from code — detects doc/code drift. | agents/engineering/DocWriter.md |
| Explorer engineering | Fast read-only search agent — locates code by pattern, finds symbols, answers "where is X". Open-ended codebase exploration, search, discovery, locate files, grep symbols, find references, unknown-scope investigation, "where is X defined", multi-naming search. Use proactively when a symbol, file, or reference needs locating and the target isn't already known. | agents/engineering/Explorer.md |
| FrontendBuilder engineering | Frontend implementation specialist — React/Vue/Next components, accessibility, performance, design tokens. | agents/engineering/FrontendBuilder.md |
| LatencyAnalyzer engineering | Analyzes latency distributions (p50/p95/p99) — identifies hot paths, suggests optimizations with data. | agents/engineering/LatencyAnalyzer.md |
| MigrationPlanner engineering | Zero-downtime migration planning. Reversibility analysis, lock-acquisition strategy, expand-and-contract patterns, validation queries. Spawned by DA module's migration-plan capability. | agents/engineering/MigrationPlanner.md |
| Migrator engineering | Performs schema, API, and dependency migrations — reversible, idempotent, with pre/post verification. | agents/engineering/Migrator.md |
| ObservabilityArchitect engineering | Observability signals design. Metrics (RED + USE), traces (spans + propagation), logs (structured + retention), SLI definitions tied to measurable signals. Spawned by DH module's observability-spec + sli-slo-spec capabilities. | agents/engineering/ObservabilityArchitect.md |
| PerfBudgetEnforcer engineering | Perf budget definition + regression detection thresholds. Sets per-journey budgets tighter than SLO, designs regression detection (drift %, sample window, alarm fan-out), recommends CI enforcement mode. Spawned by TQ module's perf-budget-spec capability. | agents/engineering/PerfBudgetEnforcer.md |
| Planner engineering | Software architect agent for designing implementation plans — step-by-step plans, file identification, trade-offs. Use proactively when a non-trivial task has multiple plausible approaches, a change touches several files or subsystems, or the operator is stuck on sequencing. | agents/engineering/Planner.md |
| ReadOnly engineering | Read-only research and audit agent — explores codebase, answers questions, never modifies. Codebase exploration, survey, "how is X done", context gathering, audit dry-run, read-only investigation, answer open questions, non-destructive analysis, evidence-cited findings. | agents/engineering/ReadOnly.md |
| Refactorer engineering | Performs structural refactoring that preserves behavior — verified by tests before and after. Use proactively when code is correct but mis-shaped (DRY violation, deep nesting, mixed concerns), ground needs preparing before a feature, or a cross-cutting rename spans several files. | agents/engineering/Refactorer.md |
| RegressionDetective engineering | Investigates regressions — bisects between known-good and known-bad, identifies offending commit, recommends fix. Use proactively when something "used to work", a performance regression shows in CI, a test flips green to red, or behavior changed after a deploy. | agents/engineering/RegressionDetective.md |
| ReleaseEngineer engineering | Orchestrates ship gauntlet — review check, sanity, squash plan, PR open, audit log. | agents/engineering/ReleaseEngineer.md |
| ResearchSynthesizer engineering | Synthesizes findings from multiple research sources — internal docs, code, web — into a structured brief. Synthesize research findings, combine sources, research summary, comparative analysis, options evaluation, state-of-the-art brief, gap analysis, pre-design research, literature review, cited recommendations. | agents/engineering/ResearchSynthesizer.md |
| SanityChecker engineering | Cross-component architecture audit before milestone gates — consistency, dead code, drift. Use proactively before a release tag or version bump, before handoff to another developer, or after a large branch merges, to catch dead code, naming drift, and divergent patterns. | agents/engineering/SanityChecker.md |
| SchemaArchitect engineering | Cross-store schema reasoning. Polyglot persistence patterns, partition-key selection, dimensional modeling for analytics. Spawned by DA module's schema-design + sharding-plan + analytics-readiness capabilities. | agents/engineering/SchemaArchitect.md |
| SystemArchitect engineering | System-of-systems thinking. Produces non-functional requirement specs, identifies cross-system invariants, surfaces emergent properties that single-component analysis misses. Spawned by TA module's quality-attributes + boundary-review capabilities. Use proactively when latency or throughput budgets need specifying, cross-system invariants need surfacing, or emergent properties (end-to-end p99, blast radius, cost-per-request) need pinning before scale. | agents/engineering/SystemArchitect.md |
| TestRunner engineering | Runs test suites and reports failures with root-cause hypotheses. Use proactively when a non-trivial diff needs independent test signal, CI is red and wants a local reproduction, or a final verification is needed before ship. | agents/engineering/TestRunner.md |
| DesignSystemAuditor frontend | Quality-gate agent for the frontend-design-review skill. Runs 6-dimension audit (typography hierarchy + motion coherence + shader perf-budget + accessibility WCAG AA + brand conformance + responsive fidelity). Scored rubric: ≥80=green, 60-79=yellow, <60=red per dimension. | agents/frontend/DesignSystemAuditor.md |
| FrontendArchitect frontend | Design-director agent for the frontend-design orchestrator. Synthesizes typography + motion (+ shader) into frontend-design-spec.json. Picks component-library + layout-grammar + interaction-signature. Does NOT write code (that's FrontendBuilder's role). | agents/frontend/FrontendArchitect.md |
| MotionDirector frontend | Motion-language curator for the frontend-motion sub-skill. Picks GSAP/Lenis/Theatre/Rive/Motion-One based on brief energy-level + target-device. Emits motion.json with scroll-trigger-config + key-animations + perf-budget. | agents/frontend/MotionDirector.md |
| ShaderEngineer frontend | WebGL/GLSL specialist for the frontend-shader sub-skill. Picks Paper Shaders/OGL/r3f/Lygia based on visual-thesis + perf-budget + target-device. Emits shader.json with library + glsl-snippets + GPU-fallback. | agents/frontend/ShaderEngineer.md |
| TypographyCurator frontend | Font + variable-axes specialist for the frontend-typography sub-skill. Picks font-stacks from Google Fonts/Pangram/Velvetyne/Recursive/Fraunces/Future-Fonts. Specs size-scale + line-heights + loading-strategy + licensing-context. | agents/frontend/TypographyCurator.md |
| ComplianceOfficer security | Cross-framework compliance evidence orchestration. Maps technical + procedural controls to SOC2/GDPR/HIPAA/PCI-DSS/FedRAMP/ISO27001 requirements; surfaces gaps + collects evidence pointers. Spawned by SC module's compliance-evidence capability. | agents/security/ComplianceOfficer.md |
| DependencyAuditor security | Audits dependencies for CVEs, outdated versions, license incompatibility, and supply chain risks. Use proactively before a release, after a dependency bump to check what came in transitively, or when a license-compatibility question is raised. | agents/security/DependencyAuditor.md |
| JWTSecurityReviewer security | Reviews JWT token usage — signing algorithm, validation chain, claim handling, expiry logic, key rotation. Use proactively when reviewing JWT-using auth code, an alg-none or weak-algo attack is suspected, or a JWT-related vulnerability shows in a dependency scan. | agents/security/JWTSecurityReviewer.md |
| OAuthFlowReviewer security | Reviews OAuth 2.0 / OIDC flows for correct grant type, PKCE usage, scope minimization, and token handling. Use proactively when reviewing a new auth implementation, an OAuth credential leak is suspected, or a flow is migrating off implicit grant to auth-code with PKCE. | agents/security/OAuthFlowReviewer.md |
| PrivacyBoundaryAudit security | Privacy + data-residency sweep — identifies where data crosses boundaries it shouldn't. | agents/security/PrivacyBoundaryAudit.md |
| SBOMAuditor security | Audits Software Bill of Materials (SBOM) for license compliance, vulnerability exposure, and supply-chain risk. | agents/security/SBOMAuditor.md |
| SecretsScanReviewer security | Reviews secrets-scan output (gitleaks, trufflehog, GitGuardian) and triages findings as true-positive / false-positive / rotated. | agents/security/SecretsScanReviewer.md |
| SecurityAuditor security | Security-focused audit — injection vectors, secret leakage, auth bypass, OWASP top 10, supply chain. Use proactively before shipping security-sensitive code (auth, billing, anything touching user input), when a new endpoint or dependency is added, or post-incident to check for a missed pattern. | agents/security/SecurityAuditor.md |
| ThreatModelDrafter security | Drafts STRIDE-based threat models for a system or feature — produces structured threats + mitigations. Use proactively when a new feature is in design, a security-review milestone approaches, or the question is "what could go wrong?" before code exists. | agents/security/ThreatModelDrafter.md |
Outputs and storage
Storage: durable project knowledge is commit-eligible, not automatically committed. Runtime is local ignored state, not encrypted/access-controlled. Configured home and explicit outputs can be elsewhere. Trusted source LINTEL_SOURCE_ROOT differs from working LINTEL_REPO_ROOT. docs/lifecycle.md
An index of declared output families plus the current Universal contracts. Not every run emits every file; arbitrary generated products do not have a finite filename list. Exact mapped paths and canonical instructions govern selected outputs.
Installation and discovery
| ID | Output family / writer | Scope, update and evidence |
|---|---|---|
| I01 | Native bare installation · explicit target | Python-free Bash/PowerShell owned-file lifecycle. Copies reviewed installation content; preserves unrelated files. Installation integrity and host activation are separate. No wholesale replacement or permission grant. docs/native-installation.md |
| I02 | Operator preferences and packs | LINTEL_HOME/profile.yaml supplies mode/role preferences. Active-pack selection differs from verified effective profile context; required policy never silently falls back.docs/lifecycle.md |
| I03 | Configuration-home directories | Explicit helpers/skills may create audit, session, draft, design, benchmark, role and provenance directories under configured home or caller-selected outputs. Presence is not a producer observation or retention policy. docs/lifecycle.md |
| I04 | Owned recovery store | Explicit --store/LINTEL_RECOVERY_STORE, or reported sibling .lintel-recovery-<full-root-digest>. Receipts and snapshots stay together. Not a timestamped whole-home backup.docs/lifecycle.md |
| I05 | Consumer root CLAUDE.md, AGENTS.md, CORE-PRINCIPLES.md, SESSION-PROTOCOL.md, EVOLUTION.md, EVOLUTION-LOG.md, TEMPLATE-skill.md, TEMPLATE-agent.md | li-scaffold init renders/copies only if absent. K. Existing entry files remain user-owned. Commit is suggested, not performed. bin/li-lifecycle.py; bin/li-lifecycle.py. |
| I06 | .claude/memory/{lessons.md,working-state.md,personas.md,personas-example.md}; .claude/plans/todo.md; .claude/decisions/{README.md,TEMPLATE.md}; .claude/rules/README.md; .claude/SUBAGENT-GUIDE.md; .claude/templates/swarm/{coordination.template.json,charter.template.md,agent-brief.template.md,agent-report.template.md,agent-review.template.md} | Scaffold copy-if-absent, K. No local agent overrides are installed. Templates are distinct from live initiative evidence. bin/li-lifecycle.py. |
| I07 | Migration and pointer repair | Layout marker, memory index, explicit migrated paths and redirects. Migration does not change Git staging; checked file transactions and explicit pointer repair preserve ownership. docs/lifecycle.md |
| I08 | Universal repository adapter | One .github/lintel/ bundle with manifest.json, START.md, canonical source, selected docs and client discovery roots. Hooks excluded. Copilot alone adds its three custom-agent profiles.bin/li-copilot.py |
| I09 | Foundation seeds and protocol | Seeds become project-owned. Managed startup blocks preserve surrounding prose. Native wrappers reference canonical source; updates retain clients and refuse modified managed files. docs/client-adapters.md |
| I10 | Host plugin/extension caches/settings, e.g. ~/.claude/plugins/cache/...; manually configured ~/.claude/settings.json and hook links | H/O, host-owned locations and retention. Use the actual host installation route and explicit ownership records. A native Lintel installation does not certify a host plugin cache or install platform enforcement. docs/native-installation.md; bin/li-update; docs/architecture.md. |
Cycle and durable knowledge
| ID | Output family / writer | Scope, update and evidence |
|---|---|---|
| C01 | SENSE · cycle identity | .claude/runtime/state/00-state.md carries explicit cycle, selected map and verified profile reference. workflow_begin precedes phase writes. Truncated DONE cannot advance. Local ledger is not a second task map.skills/spec-kit/references/work-map.md |
| C02 | SCOPE: .claude/runtime/jobs/<id>/scope.md if job active; otherwise .claude/runtime/state/scope.md | Agent fixed-name current scope, R/O. Explicit LINTEL_JOB_DIR or LINTEL_STATE_DIR determines destination. No version/archive contract. skills/scope/SKILL.md; skills/scope/SKILL.md. |
| C03 | DEFINE: selected original design/plan; for new work an explicit .claude/plans/<initiative>/design.md | Agent; durable repository artifact K. Reconcile authorized sections and preserve original IDs, accepted decisions and linked legacy paths. No timestamp alternate or overwrite merely because a document exists. Record actual review evidence and unresolved decisions. skills/define/SKILL.md. |
| C04 | DISCOVER: .claude/runtime/state/discover-report-<datetime>.md | Agent, R timestamped report, path recorded in ledger. No default promotion to committed docs. skills/discover/SKILL.md; skills/discover/SKILL.md; skills/discover/SKILL.md. |
| C05 | PLAN · executable handoff | .claude/plans/<initiative>/{plan.md,spec.md,prompt.md} or mapped originals. Short verifiable leaves grouped into bounded packages. Plan approval does not grant unrelated external permission.skills/plan/SKILL.md |
| C06 | work.json · original authority | .claude/plans/<initiative>/work.json, linked from todo/working-state. Spec Kit keeps original specs/<feature>/{spec,plan,tasks}.md and constitution. Context view is source-status-only, release_clearance:false.skills/spec-kit/references/work-map.md |
| C07 | PLAN analysis and checkpoints | Planner checkpoint/metrics remain local runtime. Analysis uses exact cycle-bound analyze_report_path, never newest/global latest. Required analysis and policy are rechecked before BUILD.skills/plan/SKILL.md |
| C08 | BUILD · product and evidence | Authorized source/tests/docs, .claude/runtime/state/build-log.md, phase/metric observations. Package completion covers original leaf acceptance; coordinator owns shared state.skills/build/SKILL.md |
| C09 | REVIEW · human and structured | Timestamped review/compliance Markdown remains. V2 context, review, QA and corroboration bind actual selected bytes, acceptance and verified profile; Markdown PASS is insufficient. skills/review/references/evidence.md |
| C10 | SHIP · reviewed result | Commits/PR/release artifacts follow authorized path. Shared review/QA gates reread same context; changed inputs or later rejection invalidate old clearance. Gate success is not publication permission. skills/review/references/evidence.md |
| C11 | CAPTURE · reviewed knowledge and handoff | Project lessons use add/update/supersede/no-op; warranted ADRs start Proposed. Reaffirm the original mapped spec/tasks/plan/prompt against actual evidence. Optional retrospectives and release summaries have selected windows and authorized outputs. Blocked work stays open. Granularity calibration remains dormant, not an automatic capture product. skills/capture/SKILL.md |
| C12 | .claude/memory/{MEMORY.md,working-state.md,personas.md,<topic>.md}; .claude/rules/<rule>.md | Agent continuity / compatible native host memory / operator rules, K. Index stays small; details can move to topic files. Budgets only warn; no automatic consolidation/deletion. lib/memory.sh; scaffolding/01-foundation/SESSION-PROTOCOL.md; .claude/decisions/0006-memory-v2.md. |
| C13 | .claude/agent-memory/<agent-name>/MEMORY.md and host-supported topic files | Compatible Claude agent memory, K by ADR intent. Host-specific; CAPTURE may curate, no daemon. Not a guarantee on every CLI/desktop. .claude/decisions/0012-agent-memory-and-model.md; .claude/decisions/0012-agent-memory-and-model.md. |
| C14 | Architecture decisions and risks | .claude/decisions/NNNN-<slug>.md or mapped project ADR store; preserve IDs/rationale and risk records. Role/template is not authority to alter governance or publish.skills/adr-new/SKILL.md |
Swarm coordination
| ID | Output family / writer | Scope, update and evidence |
|---|---|---|
| S01 | Selected initiative swarm/{coordination.json,charter.md}, swarm/briefs/<card-id>.md; work-map opt-in fields | Coordinator, K after approved map/explicit opt-in. Repo-relative pointer-selected paths can differ from example. Coordination is topology, not a duplicate authoritative task list. docs/concepts/swarming-work.md; skills/swarm/SKILL.md. |
| S02 | swarm/reports/<card-id>.md; swarm/reviews/<card-id>.md; worker-owned source | Worker writes owned scope/report; distinct reviewer writes lane review. K, preserved for cold recovery; no expiry. Coordinator alone owns shared plans/ledgers/reducers/commits/integration. docs/concepts/swarming-work.md; skills/swarm/SKILL.md. |
| S03 | Swarm evidence and isolation | Selected worktrees/isolated patches, attributable snapshot, actual reviewer input, original IDs and integration evidence. Local results may await shared evidence. No scheduler/security sandbox is created. docs/concepts/swarming-work.md |
Runtime and evidence
| ID | Output family / writer | Scope, update and evidence |
|---|---|---|
| R01 | Pause · compatible owned checkpoint | .claude/runtime/sessions/<branch>/<timestamp>-r<repository-key>-<slug>[-label]-context-save.md. Use pause then resume --from <checkpoint>. The saved filename suffix remains compatible. Reserved empty files are not checkpoints; readback verifies nonempty content and ownership.skills/pause/SKILL.md |
| R02 | Context bookkeeping | Budget/readiness reports distinguish observed, estimated and unknown headroom. Ignore records do not physically delete tokens. Preserve work/profile identity through save → restart → restore. skills/context-budget/SKILL.md |
| R03 | Advisory freeze and performance markers | .claude/runtime/state/code-freeze/<session-id>.yaml; optional performance marker under runtime sessions. code-freeze --list/--lift uses the shared validated state grammar. The old home freeze file is read-only legacy input. Metadata does not change host permission or guarantee expiry.skills/code-freeze/SKILL.md |
| R04 | Job root J/<workflow>-YYYYMMDD-HHMM-<hash>/{job.yaml,00-state.md,inputs/,outputs/}; optional scope.md, scratch/ | _jobs.sh helper creates YAML/empty ledger/dirs; steps/status mutate YAML via temporary replacement. R on v5, H legacy, O via LINTEL_JOBS_DIR. Outputs depend on workflow. Job auto-spawn remains opt-in/dormant, not all invocations automatically create jobs. bin/_jobs.sh; bin/_jobs.sh; .claude/decisions/0008-activation-contract.md. |
| R05 | J/_active.md; $H/jobs/_active.md | Helper regenerates local index and current scope's global registry rows. Local R/global H; pointers, not replicated job content. Overrides LINTEL_JOBS_ACTIVE / LINTEL_JOBS_REGISTRY. This baseline uses temp-and-rename without a registry read/modify/write lock. bin/_jobs.sh; bin/_jobs.sh. |
| R06 | Jobs · explicit archive/promotion | Resolved job archive; optional job-end copies selected plan/spec/prompt/ADRs, removes scratch and archives. Lessons only surfaced as candidates. Not registered by default; inspect status/destination/overwrite behavior. hooks/shared/job-end/run.sh |
| R07 | .claude/runtime/audit/<category>.jsonl; $H/audit/<category>.jsonl; explicit LINTEL_AUDIT_DIR/<category>.jsonl | Unified helper appends one record per actual call. Repo events R on v5; pack-lifecycle, pack-resolver, migration, migrations, self-test and usage-* default H. Explicit destination wins. Routing resolved when sourced. No rotation, immutable store, server or blanket instrumentation. bin/_audit.sh; bin/_audit.sh. |
| R08 | Routed envelopes-YYYY-MM-DD.jsonl; brief-forge.jsonl; temporary envelope file | Explicit Brief Forge invocation: append envelope/stats, stdout envelope; recipe uses EXIT trap to remove temp. Not universal middleware. skills/brief-forge/SKILL.md; skills/brief-forge/SKILL.md; docs/architecture.md. |
| R09 | Content-bound review / QA | Selected pure record under .claude/runtime/reviews/; explicit context/request/QA/corroboration outside selected inputs. Latest applicable decision precedes verdict. Legacy history is not v2 clearance.skills/review/references/evidence.md |
| R10 | Bound profile generation | <home>/sessions/profiles/<sha256(repo-path,context-id)>/current-profile.json or explicit context path. Content drift blocks; reason-bearing rebind retains prior generation and requires affected replan/review.lib/profile_context.py |
| R11 | $H/sessions/<id>/.last-bloat-warn; $H/sessions/<id>-design-surfaced; $H/.frontend-design-surface-disabled | Compatible opt-in hooks: overwrite warning marker, append surfaced file paths; read the current disabled marker without legacy migration. H, no expiry. tokens.txt and tool-calls.count are optional inputs, not generated by bloat hook. hooks/shared/context-bloat-warn/run.sh; hooks/shared/context-bloat-warn/run.sh; hooks/shared/frontend-design-surface/run.sh; hooks/shared/frontend-design-surface/run.sh. |
| R12 | Historical .claude/runtime/audit/question-preferences.jsonl, questions.jsonl and dated parity reports | Legacy/dormant files, not guaranteed current outputs. PLAN does not write question-policy history. Current instruction-parity-check returns read-only checker/report results; no prescribed dated Markdown writer or blanket question capture. skills/instruction-parity-check/SKILL.md. |
| R13 | Recovery boundary | Selected original map and evidence. Missing runtime does not erase intent; missing pins/evidence do not become cleared BUILD. Abort/archive needs exact owned paths. skills/resume/SKILL.md |
| R14 | Review Method packets | Example .claude/runtime/review/<review-id>/inputs/ holds the rendered brief and method metadata; records/ holds the request/report. The singular packet directory differs from plural runtime/reviews/ decision records. Identical packet body for single and panel reviewers; immutable selected questions and acceptance.skills/review/SKILL.md |
| R15 | MARS · optional review panel | .claude/runtime/mars/<panel>/inputs/ immutable brief/method/snapshot; records/ mutable panel, per-slot reports, identity evidence and synthesis. Changed inputs or a partial panel stay incomplete. The coordinator dispatches with host tools; helper output remains release_clearance:false.skills/mars/SKILL.md |
Engineering domains
| ID | Output family / writer | Scope, update and evidence |
|---|---|---|
| E01 | TA · domain work | Seven sub-capabilities; actual caller-selected diagrams, models, plans, reports or code. Shared observation root .claude/runtime/state/domains/<operation>/iNNNN/. Agent performs work; records preserve start/result, artifacts and profile/QA. Data verification is not execution/clearance.skills/ta/SKILL.md |
| E02 | SystemArchitect / CapacityPlanner | ta/{nfr-spec,invariants,emergent-properties,capacity-model,bottleneck-mitigations}.md. Cost projection embedded in capacity model. agents/engineering/SystemArchitect.md; agents/engineering/CapacityPlanner.md. |
| E03 | DA · domain work | Seven sub-capabilities; actual caller-selected diagrams, models, plans, reports or code. Shared observation root .claude/runtime/state/domains/<operation>/iNNNN/. Agent performs work; records preserve start/result, artifacts and profile/QA. Data verification is not execution/clearance.skills/da/SKILL.md |
| E04 | MigrationPlanner/Migrator/SchemaArchitect | da/{migration-plan,migration-risks,polyglot-model,partition-strategy,dimensional-model}.md; da/up.sql, da/down.sql, per-step SQL drafts where selected. agents/engineering/MigrationPlanner.md; agents/engineering/SchemaArchitect.md; skills/da/SKILL.md. |
| E05 | SC · domain work | Seven sub-capabilities; actual caller-selected diagrams, models, plans, reports or code. Shared observation root .claude/runtime/state/domains/<operation>/iNNNN/. Agent performs work; records preserve start/result, artifacts and profile/QA. Data verification is not execution/clearance.skills/sc/SKILL.md |
| E06 | DH · domain work | Seven sub-capabilities; actual caller-selected diagrams, models, plans, reports or code. Shared observation root .claude/runtime/state/domains/<operation>/iNNNN/. Agent performs work; records preserve start/result, artifacts and profile/QA. Data verification is not execution/clearance.skills/dh/SKILL.md |
| E07 | TQ · domain work | Seven sub-capabilities; actual caller-selected diagrams, models, plans, reports or code. Shared observation root .claude/runtime/state/domains/<operation>/iNNNN/. Agent performs work; records preserve start/result, artifacts and profile/QA. Data verification is not execution/clearance.skills/tq/SKILL.md |
| E08 | Domain recovery | Explicit start/result and artifact references. Started without result is interrupted. Verify/summary reread expected records/artifacts; all output retains release_clearance:false. skills/full-engineering-pass/references/domain-handoff.md |
| E09 | Full engineering composition | TA → DA/SC → DH → TQ. DA/SC concurrency requires opted-in swarm and attributable isolation. Preserve work/profile/acceptance; no module-local ledger grants clearance. docs/concepts/engineering-modules.md |
| E10 | Domain artifacts lacking complete paths | OpenAPI/Protobuf/SDL contracts, schemas/migrations, diagrams, source/doc edits; SBOM sbom.spdx.json or cyclonedx.json; CHANGELOG.md. K/O project-defined. Changelog updates Unreleased/promotes release entry. No finite universal generated filename list. agents/engineering/APIDesigner.md; agents/engineering/DatabaseDesigner.md; agents/engineering/DocWriter.md; agents/security/SBOMAuditor.md; agents/engineering/ChangelogMaintainer.md. |
Design, browser and deliverables
| ID | Output family / writer | Scope, update and evidence |
|---|---|---|
| G01 | Generate pipeline | Outline, content, conditional speaker notes, design, QA and selected formats. Recipes/files are not format acceptance. Preserve partial runs and limits; no general expiry daemon. skills/generate/SKILL.md |
| G02 | Outline / write / design / QA | Explicit --out/--out-dir and actual produced resources. QA preserves immutable obligations and source fidelity. Illustrative filenames are not renderers. skills/generate-qa/SKILL.md |
| G03 | Word / PowerPoint / web | Retained workflows/output contracts. Word/PPT format acceptance remains staged, not PASS. Web output is assessed against selected stack and actual browser evidence. skills/generate-ppt/SKILL.md |
| G04 | PDF / workbook / Visio · staged | A15.1–A15.4 removed from acceptance. PDF reader removed; HTML preparation/browser print remain. Actual document content/layout needs separately obtained reader/viewer evidence. .claude/decisions/0033-remove-document-format-acceptance-and-pdf-reader.md |
| G05 | generate-app | --out, default $H/generate-runs/<run-id>/app/; Next.js app/{layout.tsx,page.tsx,about/page.tsx,contact/page.tsx,globals.css}, components/{ui,motion,hero}/, lib/{motion.ts,typography.ts}, public/, package.json, next.config.js, tsconfig.json, tailwind.config.ts; stack equivalents for Vite/SvelteKit, optional api/, font guidance public/README-fonts.md and README; H/O. Arbitrary subsequent application files extend the family. Audit generate-app-runs. No collision/retention policy. skills/generate-app/SKILL.md; skills/generate-app/SKILL.md; skills/generate-app/SKILL.md; skills/generate-app/SKILL.md. |
| G06 | generate-style-learn | $H/brand/palettes/<name>.json, <name>-STYLE.md; --out-dir; failure fallback ./palettes/<name>.json; H/O/K. Existing palette blocks without --overwrite; replacement writes with >, no expiry. skills/generate-style-learn/SKILL.md; skills/generate-style-learn/SKILL.md; skills/generate-style-learn/SKILL.md. |
| G07 | Resolved frontend design | Brief, current profile, selected assets/retrieval, project manifest/lockfile and font/library provenance. Brief > profile > corpus for aesthetics; policy/stack constrain. None/CSS motion and no shader are valid. skills/design-dna/references/design-contract.md |
| G08 | Design review | Actual observations at explicit artifact/QA paths. Six dimension scores advisory; mandatory measured contrast or absent evidence blocks. Static-page/React-Vite evidence is not all-renderer acceptance. skills/design-dna/references/design-contract.md |
| G09 | frontend-style-extract | $H/brand/design-patterns/<name>/ or --out: pattern.json, typography.json, motion.json, component-imports.json, optional shader-snippets/<n>.glsl, _raw/raw-extraction.json; H/O. Existing pattern directory refused by default; overwrite records previous_pattern_hash. No expiry. Skill calls _raw ignored, but repository-local _raw is not ignored by this source repo's rules; do not rely on that prose. skills/frontend-style-extract/SKILL.md; skills/frontend-style-extract/SKILL.md; skills/frontend-style-extract/SKILL.md; skills/frontend-style-extract/SKILL.md. |
| G10 | Design persistence / tokens | Selected design-system, page overrides and token products. Verify current design/source fidelity; do not infer acceptance from a legacy MASTER.md or cached profile. skills/design-dna/SKILL.md |
| G11 | Design advice, variants, mockup and built-UI review | frontend-design --mode advice|variants preserves owned design artifacts; variants use explicit --out and rendered HTML plus comparison index. generate-web --mode mockup writes selected single-file HTML. frontend-design-review inspects actual UI and retains design-review.json; no invented spec or implicit global run root.docs/migrations/2026-09-25-native-workflows.md |
| G12 | Browser / scrape artifacts | Owned session, admitted URLs, actual DOM/accessibility/action, inspected screenshot/print observations at selected paths. CDP is single-page; explicit browser executable, Node 22+, Python 3.9+. No dependency installation. skills/web-session/references/browser-operations.md |
| G13 | Web session · user-managed authentication | web-session --mode cookies|open retains the chosen authorized surface, URL admission, owner and cleanup. No cookie/credential transfer, personal-profile scan or new navigation permission. Missing supported authentication stays a manual task.skills/web-session/SKILL.md |
| G14 | Performance and developer-experience inspection | perfbench preserves selected benchmark runs and explicitly replaced baselines. inspect --target repo --lens devex retains observed commands/results and scoped optional benchmark work. Baseline update, fresh clones and any writable checks need their actual authority; no automatic run cleanup.skills/inspect/SKILL.md |
| G15 | Documents / PDF / reports / evaluation | Explicit destinations, source fidelity, actual observations and limits. Reports/evals are not independent acceptance. PDF retains ADR-0033 removed-reader boundary. skills/generate-docs/SKILL.md |
| G16 | scaffold-internal-tool | Selected --path or cwd/name: language/type source, README, CLAUDE, ignore, LICENSE, editorconfig, tests; TS package/tsconfig/vitest; Python pyproject/ruff/pytest; Go go.mod; Rust Cargo/src/main; GitHub ci.yml or azure-pipelines.yml; compliance/data-class.md; K/O, agent creates project and first Git commit. Existing nonempty target asks merge/new path. Variants and further app deliverables cannot be finite filenames. skills/scaffold-internal-tool/SKILL.md; skills/scaffold-internal-tool/SKILL.md. |
| G17 | scaffold-mvp | Selected root README/CLAUDE/LICENSE/ignore/editorconfig/CHANGELOG; src/, tests/{unit,e2e,fixtures}/; compliance/{data-class.md,provenance.yaml} plus pack DRAFTs; conditional eval/suite/{README.md,golden/,adversarial/}, eval/runs/.gitkeep; deploy/targets.yaml; docs/design/INDEX.md, docs/adr/, docs/guides/; .github/workflows/{ci.yml,compliance-check.yml}; K/O, agent instruction, Git init/first commit. Frozen FINAL/golden content is preserved by instructed policy. Existing path requests merge/new path. CI/deploy files are stubs/pre-wiring, not proof of activated enforcement/deployment. skills/scaffold-mvp/SKILL.md; skills/scaffold-mvp/SKILL.md. |
| G18 | QA/build/tool-specific products | Tests, snapshots, coverage, traces/HAR, bundles, source maps, dependency locks/caches, images/fonts and generated domain documents under project/tool-selected roots; K/O/R depending actual tool and ignore policy; not a fixed harness schema. Example Playwright trace is a diagnostic pointer, not a universally emitted Lintel path. Source repo ignores node_modules, lockfiles and dist/root build, but these rules are not universal consumer policy. skills/verify/SKILL.md; .gitignore. |
Knowledge, roles and packs
| ID | Output family / writer | Scope, update and evidence |
|---|---|---|
| X01 | Project lessons · stable IDs | .claude/memory/lessons.md on v5. lessons-add classifies add/update/supersede/no-op, uses locked conditional writes and preserves history. lessons-surface reads selected active matches and full blocks. No operator-global sink is activated; old home lessons JSONL is read-only legacy data.skills/lessons-add/SKILL.md |
| X02 | Explicit Lintel checkout: selected lesson-promotion target | Default promotion writes the conditional target and prints a diff; HEAD, branches and index remain unchanged. Optional --commit --expect-branch requires a clean target/index and the selected existing branch. No automatic branch creation. bin/li-lessons-promote. |
| X03 | Private lesson sync · optional | Explicit configured private source/destination, repository identity and authority. Not default CAPTURE or cloud checkpoint replication; never implicit customer/team publication. bin/li-lessons-sync |
| X04 | ROLE-NEW: explicitly selected draft; configured private-role destination or published role file | Publication uses li-lifecycle role-write. Preserve selected draft path; private destination comes from configuration. Updates require expected SHA-256. No fixed global draft-directory contract. skills/role-new/SKILL.md. |
| X05 | Private role sync · optional | Explicit authorized push/pull to chosen private destination. Role files do not activate host agents or establish effective policy. No assumed cross-machine installation. bin/li-roles-sync |
| X06 | CAPTURE vault sink: <capture.vault_sink_path>/YYYY-MM-DD-<repo>-<slug>.md, <repo>.md hub, 00-index.md; vault-init adds sessions.base | O external or relative-to-repo configured directory, enabled and existing only. Agent writes one session summary, creates hub only if absent, prepends index entry and truncates index list to 15, not old note deletion. Mandatory pre-write sensitive-content scan; failure skips export. Helper vault-init never overwrites existing three files. No vault query/automatic read-back into work context; index preservation is the bounded read exception. skills/capture/SKILL.md; bin/li-vault-init. |
| X07 | Owned packs and explicit profile changes | Selected repo/home packs/<name>/pack.yaml; configured active pointer; versioned effective profile history. Create stages/validates; switch/rebind records a reason and retains the prior generation. Missing required policy never becomes neutral success. No private assets or credentials copied by default.docs/lifecycle.md |
| X08 | Extension pack skeleton | Selected pack manifest, README, client entry files and declared capability folders. li-pack-scaffold preflights existing paths before publication. Scaffold existence is not implementation, installation or remote publication.bin/li-pack-scaffold |
| X09 | New skill · owned draft | skill-new writes a selected owned scaffold, validates frontmatter and can start from a chosen lesson. Existing candidate preference/type values remain compatible. A draft is not native discovery or host activation; those need separate verification.skills/skill-new/SKILL.md |
| X10 | Owned profile / safe-install recovery | Exact selected files, verified snapshot ID and bound result receipt. Restore preflights paths/blobs and preserves unrelated/later user edits. No fictional .disabled control or whole-home rollback. skills/safe-install/SKILL.md |
| X11 | Standing review questions | Source lib/review-questions.json; optional project .claude/review/questions.json. Stable IDs and subject/surface selection define packet coverage. Optional accepted/rejected/escaped observations may propose changes; questions do not automatically rewrite themselves.skills/review/references/method.md |
Maintenance and generators
| ID | Output family / writer | Scope, update and evidence |
|---|---|---|
| M01 | .claude/engineering/compat-audits/<date>-<slug>.md; .claude/engineering/audits/uniformity-matrix.md or custom output | Executable maintainer generators, K/O. Same filename replaces prior contents; compatibility temporary diff removed by trap. Uniformity check/stdout modes do not replace report. bin/li-compat-audit; bin/li-compat-audit; bin/li-compat-audit; bin/li-uniformity; bin/li-uniformity. |
| M02 | <output-base>/docs/wiki/{README,skills,agents,packs,schemas}.md; docs/showcase/lintel-the-harness.html; generated root README support table | Wiki helper defaults source root; --output changes base. K/O regenerates and overwrites generated content; temp dir removed by EXIT trap. Check mode compares generated temp version. bin/li-wiki-gen; bin/li-wiki-gen; bin/li-wiki-gen. |
| M03 | skills/CATALOG.md; synchronized protocol entry blocks | Catalog/instructions generators, K, source-derived reducers. Catalog writes complete replacement, --check read-only. Coordinator owns fan-in reducers during swarm. bin/li-catalog.py; docs/concepts/swarming-work.md; bin/li-copilot.py. |
| M04 | Architecture evolution/structure-impact, audit/review evidence and migration notes under chosen repo engineering/docs roots | Agent-authoring convention, K/O. Cycle names structure-changes/<date>-<slug>.md but points at .claude/engineering/evolution/_TEMPLATE.md; existing artifacts also use engineering/evolution. No single enforced path writer; preserve exact repo convention, don't invent a duplicate store. skills/cycle/SKILL.md; .claude/decisions/0006-memory-v2.md. |
| M05 | Maintenance / bounded retention | Inspect and explicitly recover owned paths. No chat compaction/cleanup daemon. Snapshot retention protects newest five plus last 30 days, locks/incomplete restores; prune needs explicit eligible IDs. Not universal runtime GC. skills/maintenance/SKILL.md |
| M06 | OS temp editor/envelope/diff/wiki files; adjacent .tmp/.bak/.lintel-* atomic-replacement files | Individual helpers use cleanup traps/finally/explicit remove; transient T, not durable deliverables. Cleanup is local to each operation, not system-wide. Failed or interrupted processes can still require inspection. bin/li-lessons-promote; skills/brief-forge/SKILL.md; bin/li-wiki-gen; bin/li-copilot.py; skills/pack-create/SKILL.md. |
Universal additions
| ID | Output family / writer | Scope, update and evidence |
|---|---|---|
| N01 | Required profile selection | .claude/profile-requirements.json or explicit LINTEL_PROFILE_PACK. Invalid mandatory policy blocks. Optional fallback labelled, validated neutral baseline only. Actual organizational enforcement remains unverified in merged acceptance setting..claude/decisions/0029-required-profile-context.md |
| N02 | Immutable review obligations | V2 prepared context owns QA before results. Zero tests, missing mandatory checks, invalid N/A or changed acceptance cannot be averaged into success. Actor strings alone do not authenticate independence. skills/review/references/evidence.md |
| N03 | Owned transaction receipts | Exact source/target/store/operation identity; publish, inspect and recover separate. Partial writes cannot be adopted as successful init. Receipts do not roll back host/remote effects. docs/lifecycle.md |
| N04 | Producer event catalog | 23 categories / 66 kinds in lib/event-catalog.json. li-events summary/records/installer reads explicit input; absence is unobserved. Not a hook bus/scheduler/enforcement verdict.bin/li-events.py |
One Review Method. Optional MARS panel.
One method, one set of standing questions. Single REVIEW and each MARS participant receive byte-identical packet bodies; only their headers differ. Reviewers report without fixing. skills/review/references/method.md
| Step | What the contract requires |
|---|---|
| Intent and acceptance | Explicit selected requirements and original leaf IDs |
| Evidence | Executed E1 or traced E2; patterns/hypotheses stay labelled |
| Coverage | Every selected question checked with evidence, finding, justified N/A or explicit not-checked |
| Decision | P1 fails; P2 requests changes; missing coverage/contradiction is incomplete, never PASS |
| Currentness | Content, attempt, work, profile, required QA and latest applicable independent decision |
| MARS option | At least2 distinct observed models; explicit consent and live host eligibility; blind first pass and at most one challenge round |
A complete checker result is not a passing verdict. MARS preserves dissent; the coordinator adjudicates against source rather than voting. MARS inspection always has release_clearance:false. In REVIEW panel mode, REVIEW records its own decision through the existing v2 evidence path. .claude/decisions/0036-mars-multi-model-review.md · skills/review/references/evidence.md
MARS is optional and distinct from implementation Swarm. Unknown capability suppresses automatic offers; auto mode and silence are not consent. The RM9 Copilot App pilot had four observed identities but only three valid reports: partial, multi_model_verified:false, one synthetic subject, no challenge or live single-review comparison. skills/mars/SKILL.md · .claude/plans/mars/pilot-2026-09-25-rm9.md
Hooks: activation, effects and limits
33 scripts · 9 core registrations · 5 lifecycle event types · 24 optional/manual. Native adapter uses Claude Code protocol. Universal repository kit installs no hooks, even for Claude. Copilot supports hooks but this release ships no translation. hooks/hooks.json · docs/copilot.md
Registration is not a firing test. Plain stdout/exit 0 proves emitted advice, not agent visibility. Stop does not continue work or checkpoint. Bash/Git/scanner dependencies and actual input mapping matter. Optional HOOK.md event declarations do not activate scripts.
| Script / activation | Actual behavior | Limits | Current source |
|---|---|---|---|
| context-bloat-warn Not registered · PreToolUse (any) | Reads available token/call counters and emits pause/fresh-session/resume advice at configured thresholds. | No compaction or automatic checkpoint. Observed counters are required; unavailable telemetry is not measured headroom. | hooks/shared/context-bloat-warn/run.sh hooks/shared/context-bloat-warn/HOOK.md |
| customer-data-block PreToolUse · Bash | Same target-aware Git collector with shared customer-data patterns. Match/scanner/collection/scan failure denies via exit 2; explicit override audited. | Broad regex false positives; override authority not authenticated. No complete exfiltration/platform control. | hooks/shared/customer-data-block/run.sh hooks/shared/customer-data-block/HOOK.md |
| cycle-incomplete-warn Stop | Reads the canonical cycle footer; incomplete cycle prints footer plus go/pause/capture advice and A. (H/cycle-incomplete-warn) | Guarded helper sourcing, no-cycle/completed/error paths silent. Explicitly no block/forced continuation. Does not save a checkpoint or advance phase. Plain stdout delivery still needs native evidence. | hooks/shared/cycle-incomplete-warn/run.sh hooks/shared/cycle-incomplete-warn/HOOK.md |
| cycle-position-inject UserPromptSubmit | Active ledger -> compact position and workflow reminder in additionalContext, A. Recognized cycle command without ledger -> context nudge to initialize state. (H/cycle-position-inject) | Reads state/footer/input; does not write phase ledger or force next task. All intentional paths exit 0. Entry recognizer only colon-style commands. Pack continuity provider at :76 is a deferred comment, not implemented. Gap-entry nudge has no direct hook audit call. | hooks/shared/cycle-position-inject/run.sh hooks/shared/cycle-position-inject/HOOK.md |
| da-migration-irreversible-warn Not registered · PreToolUse (Edit|Write on migration files) | Recognizes migration paths/custom glob; scans existing file for DROP TABLE/COLUMN, TRUNCATE, ALTER COLUMN TYPE; warns only when a destructive pattern and no paired down-file/in-file down marker coexist. A. (H/da-migration-irreversible-warn) | Does not warn on every migration lacking rollback. Does not test rollback correctness, assess documented data loss or execute SQL. New not-yet-written files are skipped. Custom resolver trust caveat. | hooks/shared/da-migration-irreversible-warn/run.sh hooks/shared/da-migration-irreversible-warn/HOOK.md |
| da-retention-violation-warn Not registered · PreToolUse (Edit|Write on data-access code) | Reads runtime da/retention-policy.md; extracts table/window/filter tokens, then checks whole-file table reference without corresponding filter token. Warns/A. (H/da-retention-violation-warn) | No runtime retention enforcement or query analysis; days are parsed but not applied. Parser contradicts example - table_name:90days: first whitespace token is '-' and becomes empty, so common bullets are skipped. No-match grep pipelines under -e can terminate early. Existing-file pre-edit inspection only. | hooks/shared/da-retention-violation-warn/run.sh hooks/shared/da-retention-violation-warn/HOOK.md |
| da-schema-drift-warn Not registered · PreToolUse (Edit|Write on schema-ADR-claimed files) | Like architecture watcher, additionally filters ADR text for schema/migration/data/table/column. Warns/A. (H/da-schema-drift-warn) | Same missing canonical ADR directory/frontmatter/regex limitations. Does not compare actual database schema or run migrations. | hooks/shared/da-schema-drift-warn/run.sh hooks/shared/da-schema-drift-warn/HOOK.md |
| dh-cost-budget-warn Not registered · PreToolUse (Edit|Write on IaC files) | Scans existing IaC file for large Azure SKU-like names, high replica count, premium and always-on markers; warns/A. (H/dh-cost-budget-warn) | No price lookup, cost projection, numeric budget comparison or before/after delta. Presence of a premium string is enough. The SKU regex includes small SKUs; “cost-increasing” is only a heuristic label. Custom resolver trust caveat. | hooks/shared/dh-cost-budget-warn/run.sh hooks/shared/dh-cost-budget-warn/HOOK.md |
| dh-deploy-without-rollback-warn Not registered · PreToolUse (Edit|Write on deploy/IaC files) | Searches runtime dh/rollback-strategy-*.md modified within7days or rollback/revert/undo text in file; if absent warns/A. (H/dh-deploy-without-rollback-warn) | Any recent strategy can satisfy check regardless of service; a token/comment can satisfy in-file marker. No deployment interception, rollback execution or validation of strategy contents. Missing state handled with guarded find. Custom resolver trust caveat. | hooks/shared/dh-deploy-without-rollback-warn/run.sh hooks/shared/dh-deploy-without-rollback-warn/HOOK.md |
| dh-observability-gap-warn Not registered · PreToolUse (Edit|Write on new service paths) | Scans existing file for metric/tracing/logging regex markers; warns only when none found. A. (H/dh-observability-gap-warn) | One marker suffices; no runtime traces, dashboards or coverage verification. Middleware instrumentation may false-positive. New files skipped at pre-edit. Custom resolver trust caveat. | hooks/shared/dh-observability-gap-warn/run.sh hooks/shared/dh-observability-gap-warn/HOOK.md |
| frontend-design-surface Not registered · PreToolUse (Read|Edit|Write on *.tsx|*.jsx|*.svelte|*.vue|*.css|*.scss) | Surfaces the canonical frontend-design workflow for matched UI work. | Optional discovery advice; not design execution, mandatory review or a retired-tool kill switch. | hooks/shared/frontend-design-surface/run.sh hooks/shared/frontend-design-surface/HOOK.md |
| frozen-zone-warn Not registered · PreToolUse (Edit | Write) | Uses the shared code-freeze parser for explicit repo/session identity and validated freeze records; advisory warning and audit on a frozen path. | Optional; no filesystem lock. Current state is repo runtime; old home freeze input is read-only legacy compatibility. | hooks/shared/frozen-zone-warn/run.sh hooks/shared/frozen-zone-warn/HOOK.md |
| job-begin Not registered · PreToolUse (Skill /li:<name> where SKILL.md has workflow_root: true) | Explicit positional helper creates job via _jobs.sh; announces success only with written nonempty ID. | Unregistered; mutates job/index/audit. No automatic workflow call proven. | hooks/shared/job-begin/run.sh hooks/shared/job-begin/HOOK.md |
| job-end Not registered · PostToolUse (Skill status=DONE/ABORTED) OR explicit /li:jobs abort | Explicitly promotes selected plan/spec/prompt/ADRs, removes scratch, archives job. Lessons only surfaced as candidates. | Mutating, not warn-only. Unregistered; review actual status/destination/overwrite behavior. | hooks/shared/job-end/run.sh hooks/shared/job-end/HOOK.md |
| job-stale-warn Not registered · SessionStart (first /li:cycle, /li:resume, /li:sense, OR /li:status of session) | Reads stale jobs with default24h threshold, prints continue/abort/branch options; writes jobs.jsonl, not hooks.jsonl. (H/job-stale-warn; bin/_jobs.sh) | Own-tree jobs helper reads last_touched with date conversion fallback to now. mapfile requires newer Bash than macOS stock3.2. Profile/disable paths use literal $HOME/.lintel despite LINTEL_HOME override elsewhere. No host registration or one-time-per-session mechanism in script. Sourcing jobs helper may create dirs. | hooks/shared/job-stale-warn/run.sh hooks/shared/job-stale-warn/HOOK.md |
| memory-budget-warn PostToolUse · Edit|Write | Checks MEMORY.md line budget and active lesson count via memory_budget_check; defaults 200 lines and 30 active lessons. Warns/A and writes <repo>/.claude/runtime/state/.last-memory-budget-warn. (H/memory-budget-warn; lib/memory.sh, :107) | Requires git repo/layout marker/helper. Rate-limited once per hour when a warning occurs. Never consolidates or edits memory automatically. Reads after any matched edit, not only edits to memory. | hooks/shared/memory-budget-warn/run.sh hooks/shared/memory-budget-warn/HOOK.md |
| no-customer-data-in-message UserPromptSubmit | Submitted prompt regex scan; warning or unavailable-scan observation. | No redaction/rejection or proof prompt was not transmitted. Stdout is not proven model delivery. | hooks/shared/no-customer-data-in-message/run.sh hooks/shared/no-customer-data-in-message/HOOK.md |
| no-customer-data-in-screenshot Not registered · PostToolUse (compatible host adapter after web-session screenshot/DOM capture) | Available DOM text scan; warns or unavailable-scan observation. | No pixel inspection, OCR, automatic movement or quarantine. | hooks/shared/no-customer-data-in-screenshot/run.sh hooks/shared/no-customer-data-in-screenshot/HOOK.md |
| no-direct-main-push PreToolUse · Bash | Regex checks pushes to literal main/master, force push and HEAD aliases; plain warning and A including first 120 command chars. (H/no-direct-main-push) | Does not discover actual default branch or verify authorization; incomplete command syntax coverage. No blocking, no protection outside host-intercepted tool invocation. | hooks/shared/no-direct-main-push/run.sh hooks/shared/no-direct-main-push/HOOK.md |
| no-merge-without-review Not registered · PreToolUse (Bash) | Matched merge calls li-review-read --gate-json with expected context/corroboration when supplied. Invalid/missing content-bound review warns and audits. | Opt-in advisory, not merge block/authorization. No old seven-day log-presence heuristic. | hooks/shared/no-merge-without-review/run.sh hooks/shared/no-merge-without-review/HOOK.md |
| no-production-mutation-without-auth Not registered · PreToolUse (Bash) | Regexes selected kubectl/terraform/gh/az/Postgres commands containing prod-like markers, plus global custom pattern file. Warns/A with 120-char command preview. (H/no-production-mutation-without-auth) | Does not read or validate authorization; no blocking and no live-environment detection. Patterns include some reads as well as mutations and miss other commands. The name is stronger than the implemented guarantee. | hooks/shared/no-production-mutation-without-auth/run.sh hooks/shared/no-production-mutation-without-auth/HOOK.md |
| no-secrets-in-edit PreToolUse · Edit|Write | Proposed payload secret scan; warning/finding or scanner-unavailable observation. | No edit prevention. Unavailable scan is not clean. Actual payload/delivery depends on host. | hooks/shared/no-secrets-in-edit/run.sh hooks/shared/no-secrets-in-edit/HOOK.md |
| sc-auth-bypass-warn Not registered · PreToolUse (Edit|Write on auth-flow files) | Counts skip-auth flags, magic admin/root/test credentials, bypass routes, direct role assignments in existing file; warns/A. (H/sc-auth-bypass-warn) | Initial piped loops cannot populate parent array; implemented fallback counts at61–64 determine findings and have slightly narrower patterns. No control-flow analysis or automatic prevention. Reads before edit. Custom resolver trust caveat. | hooks/shared/sc-auth-bypass-warn/run.sh hooks/shared/sc-auth-bypass-warn/HOOK.md |
| sc-compliance-gap-warn Not registered · PreToolUse (Edit|Write on regulated-data paths) | Looks for payment/health/PII/audit-like paths/customglob; reads framework list with fallback soc2,gdpr; missing or >90day evidence files and verdict:gap lines trigger warning/A. (H/sc-compliance-gap-warn) | Runtime sc/compliance-evidence- | hooks/shared/sc-compliance-gap-warn/run.sh hooks/shared/sc-compliance-gap-warn/HOOK.md |
| sc-threat-coverage-warn Not registered · PreToolUse (Edit|Write on auth/data files) | Finds lexicographically latest runtime sc/threat-model-*.md; warns if none, >90days old, or file path absent from text. A. (H/sc-threat-coverage-warn) | Filename sort is not robust chronology; textual mention is not threat coverage. No threat-model generation/test. Custom resolver trust caveat. | hooks/shared/sc-threat-coverage-warn/run.sh hooks/shared/sc-threat-coverage-warn/HOOK.md |
| secret-scan-block PreToolUse · Bash | Matched literal git commit/push: staged/tracked commit edits or selected outgoing histories. Match or unavailable scanner/collection/scan returns exit 2; override audit-logged. | Heuristic command/pattern coverage; supported literal syntax only. Messages outside diff-content contract. Requires compatible interception; not complete DLP. | hooks/shared/secret-scan-block/run.sh hooks/shared/secret-scan-block/HOOK.md |
| session-digest SessionStart · startup|resume|clear | Effective pack/compliance, mode/role, active recent lessons via shared reader, working context/ADRs, status-grounded cycle footer. Emits startup context and best-effort audit. | Fail-open compact context, not authority replacement. Trusted resolver; no automatic resume/policy clearance. | hooks/shared/session-digest/run.sh hooks/shared/session-digest/HOOK.md |
| ta-arch-drift-warn Not registered · PreToolUse (Edit|Write on ADR-claimed files) | Searches ADR frontmatter decisions: for edited path and emits linked ADR title. A. (H/ta-arch-drift-warn) | Scans legacy .lintel/decisions, docs/decisions, docs/adr only: misses canonical .claude/decisions. Does not parse code or establish architectural violation. Interpolates path as awk regex; plain non-frontmatter ADRs do not satisfy the contract. | hooks/shared/ta-arch-drift-warn/run.sh hooks/shared/ta-arch-drift-warn/HOOK.md |
| ta-complexity-budget-warn Not registered · PreToolUse (Edit|Write) | Runs gocyclo for Go, radon cc for Python, otherwise lizard; compares parsed value with profile/default12 cyclomatic and18 cognitive; cognitive is estimated as1.3× cyclomatic. A. (H/ta-complexity-budget-warn) | Missing analyzer -> silent. External output parsing not a typed/validated result; cognitive complexity is not measured. Reads the current disk file before edit, not proposed content. Profile awk parser can miss normal nested keys by resetting section on child lines. No analyzer installed/run in this research. | hooks/shared/ta-complexity-budget-warn/run.sh hooks/shared/ta-complexity-budget-warn/HOOK.md |
| ta-contract-collision-warn Not registered · PreToolUse (Edit|Write on declared interface files) | Matches default/custom interface globs or exact compact-JSON text in consumer-registry.json, then warns about registered consumers. A. (H/ta-contract-collision-warn) | Reads .claude/runtime/state/ta/consumer-registry.json, legacy fallback. Grep depends on literal JSON formatting and counts matching lines, not parsed consumers. Does not compare schemas or detect an actual collision. Custom resolver trust caveat. | hooks/shared/ta-contract-collision-warn/run.sh hooks/shared/ta-contract-collision-warn/HOOK.md |
| tq-contract-break-warn Not registered · PreToolUse (Edit|Write on provider/contract files) | Checks whether any staged filename matches contract/pact/consumer-test patterns; if not warns/A. (H/tq-contract-break-warn) | Pre-edit stage list can be unrelated/incomplete. No pairing to edited provider, schema-diff check or test execution. Presence of unrelated test file suppresses warning. Custom resolver trust caveat. | hooks/shared/tq-contract-break-warn/run.sh hooks/shared/tq-contract-break-warn/HOOK.md |
| tq-coverage-drop-warn Not registered · PreToolUse (Edit|Write on test/source files) | Reads lexicographically last runtime tq/coverage-audit-*.md modified within1day, extracts heuristic coverage numbers and compares overall to default80/profile; prints critical-path info default100. A. (H/tq-coverage-drop-warn) | No coverage collection, no prior baseline or measured drop. Does not inspect event/path itself; uses prior report. Missing directory/unmatched parsing pipeline can exit early under -e before intended silent branch. Profile parsing caveat. | hooks/shared/tq-coverage-drop-warn/run.sh hooks/shared/tq-coverage-drop-warn/HOOK.md |
| tq-perf-regression-warn Not registered · PreToolUse (Edit|Write on perf-budget paths) | Matching pack path or recent perf-budget record; extracts exact-path journey/p95, validates metadata, warns/audits. | No benchmark or measured regression; does not install CI despite printed CI wording. Invalid metadata errors; host behavior unverified. | hooks/shared/tq-perf-regression-warn/run.sh hooks/shared/tq-perf-regression-warn/HOOK.md |
Producer observations, not enforcement: clean passes/non-recording hooks can write nothing. Absent records mean unobserved. 23 categories / 66 audit kinds are not 66 host events. li-events reports verification:not_performed, enforcement:not_established. lib/event-catalog.json · bin/li-events.py
One method. Exact surfaces.
Four supported families · 13 named surfaces + one manual fallback. 14 registry records: 12 native-format routes and 2 manual records. Four discovery roots; 15 starter workflow wrappers including MARS. Copilot retains three native role profiles. .claude/decisions/0035-four-supported-client-families.md · bin/li-copilot.py
| Evidence | Establishes | Does not establish |
|---|---|---|
| Vendor | Dated documentation and conditions | Current permission or availability |
| Delivered | Files and operation/fallback contract | Native discovery or actual execution |
| Observed | Specific version/revision/scenario and result | Other surfaces or unrun behavior |
Registry observations remain three partial Copilot App records, not complete parity. MARS has a separately recorded limited live pilot; it is not a registry-wide acceptance upgrade. lib/cli-tiers.yaml · .claude/plans/mars/pilot-2026-09-25-rm9.md
| Exact surface | Vendor operations recorded | Delivered route / hooks | Observed execution | Registry |
|---|---|---|---|---|
Claude Code CLIclaude-codecli | skills: documented, plan: documented, delegate: documented, isolate: conditional, browser: conditional, memory: documented, hooks: documented | .claude/skills/li-*/SKILL.mdOptional Claude adapter; separate compatible activation | not_run | lib/cli-tiers.yaml |
Claude Desktop Code localclaude-desktopdesktop | skills: documented, isolate: conditional, browser: conditional | .claude/skills/li-*/SKILL.mdNo hook adapter installed | not_run | lib/cli-tiers.yaml |
GitHub Copilot CLIcopilot-clicli | skills: documented, instructions: documented, plan: documented, delegate: documented, hooks: documented | .github/skills/li-*/SKILL.mdNo hook adapter installed | not_run | lib/cli-tiers.yaml |
GitHub Copilot Appcopilot-appdesktop | skills: documented, plan: documented, delegate: conditional, isolate: documented | .github/skills/li-*/SKILL.mdNo hook adapter installed | Partial question/delegate/isolate; host version and exact revision missing. Not full acceptance. | lib/cli-tiers.yaml |
GitHub Copilot VS Codecopilot-vscodeide | skills: documented, plan: documented, delegate: conditional, browser: conditional, hooks: conditional | .github/skills/li-*/SKILL.mdNo hook adapter installed | not_run | lib/cli-tiers.yaml |
GitHub Copilot cloud agentcopilot-cloudcloud | skills: documented, browser: conditional, hooks: documented, isolate: conditional | .github/skills/li-*/SKILL.mdNo hook adapter installed | not_run | lib/cli-tiers.yaml |
Codex CLIcodex-clicli | skills: documented, delegate: conditional, resume: documented | .agents/skills/li-*/SKILL.mdNo hook adapter installed | not_run | lib/cli-tiers.yaml |
Codex desktopcodex-desktopdesktop | skills: documented, isolate: documented | .agents/skills/li-*/SKILL.mdNo hook adapter installed | not_run | lib/cli-tiers.yaml |
Codex IDE extensioncodex-ideide | skills: documented | .agents/skills/li-*/SKILL.mdNo hook adapter installed | not_run | lib/cli-tiers.yaml |
Codex cloudcodex-cloudcloud | No operation capability established | .github/lintel/START.md · manual handoffNo hook adapter installed | not_run | lib/cli-tiers.yaml |
Cursor CLIcursor-clicli | skills: documented, plan: documented, resume: documented, delegate: conditional | .cursor/skills/li-*/SKILL.mdNo hook adapter installed | not_run | lib/cli-tiers.yaml |
Cursor editorcursor-ideide | skills: documented, question: documented, browser: documented, delegate: conditional | .cursor/skills/li-*/SKILL.mdNo hook adapter installed | not_run | lib/cli-tiers.yaml |
Cursor cloudcursor-cloudcloud | skills: conditional, delegate: conditional | .cursor/skills/li-*/SKILL.mdNo hook adapter installed | not_run | lib/cli-tiers.yaml |
Unidentified host (explicit manual route)othermanual | No operation capability established | .github/lintel/START.md · manual handoffNo hook adapter installed | not_run | lib/cli-tiers.yaml |
The repository adapter selects an exact supported surface, preserves existing managed ownership and bundles one common .github/lintel/ source. It installs no hook bundle, client, setting, credential or platform permission. Removed family IDs have no compatibility adapter. docs/client-adapters.md · docs/migrations/2026-09-25-supported-clients-four-families.md
Resume original work
One map links original spec, plan, task IDs and handoff; selected cycle and verified profile travel with it. Explicit selection wins; multiple active initiatives require choice, never newest-file guessing. skills/spec-kit/references/work-map.md
| State | Next action | Boundary |
|---|---|---|
| Runtime absent | Read committed map/artifacts; reconcile actual checkout | Missing evidence is not verified work |
| STARTING / BLOCKED / NEEDS_CONTEXT | Resolve context, resume same unfinished phase | Truncated DONE cannot skip ahead |
| Profile drift/missing | Explicit policy reconciliation/rebind with reason | Never silently neutralize required policy |
| Checked cards | Inspect current content-bound review/QA | Checkbox is not evidence tests ran |
| No delegation/isolation | Serial work / bounded handoff | Independent review remains open |
Swarm stays opt-in over PLAN/BUILD/REVIEW. Coordinator owns shared state/integration; worker/reviewer owns declared lane. Snapshots bind result; reviewer input exports real material. Final integrated REVIEW remains required. docs/concepts/swarming-work.md · .claude/decisions/0027-first-class-swarming.md
Required policy travels with work
Preferences differ from effective policy. profile.yaml supplies mode/role; repository requirements or LINTEL_PROFILE_PACK select mandatory packs. Required invalid policy blocks affected work. .claude/decisions/0029-required-profile-context.md
| Record | Purpose | Update |
|---|---|---|
| .claude/profile-requirements.json | Required policy input | Optional requirements file. Missing or invalid required pack, or bound-policy drift, blocks affected work. |
| Effective profile reference | schema_version/context_id/generation/digest/name/version | Pass unchanged to work, children, review, recovery |
| current-profile.json | Selected bound context | Reverify digests; same-mtime edits drift |
| Explicit rebind | Retain old generation and reason | Replan/review affected work; no silent refresh |
Policy and approved stack remain constraints. Actual organizational required-policy enforcement was unverified in merged acceptance setting; synthetic fixtures are not customer certification. Host permissions/CI configured separately. lib/profile_context.py · .claude/plans/universal-implementation/reports/final.md
Maintain what you own
Learning supports add/update/supersede/no-op; decisions preserve rationale; checkpoints differ from curated memory. No delete-by-age or conversation compactor enabled. skills/maintenance/SKILL.md
| Operation | Implemented behavior | Limit |
|---|---|---|
| Owned snapshot | Exact selected files, verified bytes and result receipt | Not whole-home/remote rollback |
| Retention plan | Union newest five + last 30 days; protect locks/restores | Not automatic deletion |
| Prune | Explicit eligible owned IDs; refuse unexpected contents | Not general runtime GC |
| Adapter upgrade | Inventory/conflict preflight/owned publication-recovery | Preserves project knowledge |
| Event inspection | summary / records / installer over explicit file | Observation ≠ enforcement |
| Continuity | pause → fresh session → resume --from, same work/profile | No asserted reclaimed context |
bin/li-snapshot.py · docs/lifecycle.md · bin/li-events.py
Trusted source and working target stay separate. Changed receipts, later edits and partial publication remain explicit recovery cases; do not adopt current bytes as successful re-init.
Upgrade and pilot evidence
Pilot one bounded task in the exact surface. Record revision, client, discovered workflow/role, selected profile, permissions, actual acceptance and cold resume. Source files are not runtime certification. shims/universal/ADAPTER.md
| Lane | Positive evidence | Negative case |
|---|---|---|
| Discovery | Actual selected root, method and role | Unavailable discovery stays explicit |
| Policy | Required pack and verified context | Missing pack or drift blocks |
| Review + QA | Current bytes, obligations, coverage and independent corroboration | Later rejection, missing review or mandatory failure blocks |
| MARS | Consented roster, identical packets and host identity evidence | Partial panel or changed inputs stays incomplete |
| Swarm | Disjoint attributable lanes and reviewed integration | No isolation means serial; missing review stays open |
| Recovery | Same map, original IDs, verified profile and unmet action | Missing evidence is not done |
| Net value | Accepted outcomes, rework, defects, recovery and process cost | Report failures and negative value too |
The committed Universal report records109/109 in-scope items and its final22-job CI run. Removed document acceptance was removed, not passed or waived. Those records do not certify all clients, policy settings or production deployments. .claude/plans/universal-implementation/reports/final.md · CHANGELOG.md
Historical demonstrations retain original provenance. Current browser evidence is bounded; PDF has no reader and full document-format acceptance remains unverified. A MARS pilot is not a general accuracy or productivity benchmark. skills/web-session/references/browser-operations.md · .claude/plans/mars/pilot-2026-09-25-rm9.md
No matching rows or sections. Clear the search to restore the full reference.