UNRELEASED DEVELOPMENT SNAPSHOT · READ-ONLY EXCERPT
skills/build/SKILL.md
This is the cited technical source at local revision 275a354, inspected on 22 September 2026. It differs from the public release. This reference publishes only the cited sections, not an installable release or the development branch.
Line numbers refer to the complete recorded file. Uncited lines are omitted. Full-file SHA-256: b58941999400336a4533b845773cf4598ba2aaf9f0df820f149cde93a4b7430a.
1
---2
name: build3
layer: foundation4
description: Use to implement an approved plan task by task. Trigger after PLAN is approved and a plan.md exists with code to write — each task gets a fresh implementer subagent and two-stage review, spec compliance then quality, with a per-task status protocol.5
color: cyan6
tools: Read, Write, Edit, Bash, Grep, Glob7
voice: internal8
cli_support: [claude-code, codex, copilot]9
necessity: REQUIRED10
gap_if_skipped: "No implementation is produced; the plan's tasks are never executed."11
---13
You are the BUILD skill — Phase 5 of the Lintel cycle.15
## What this skill does… omitted uncited lines …
259
Skill: /li:build (task T<N> of plan.md)260
[/lintel-context]"261
```263
NEVER `git add -A`. NEVER commit broken state. Operator can resume from any WIP via `/li:context-restore`.265
If `checkpoint_push: true`: also push WIP to origin.267
### Step 5 — Build log269
Append to `.claude/runtime/state/build-log.md`:270
```yaml271
task: T<N>272
title: <title>273
start_ref: <HEAD sha at 3a dispatch> # 3b-guard diffs against this274
status: DONE | DONE_WITH_CONCERNS | BLOCKED275
implementer_status: <as returned>276
review_mode: dedicated | inline # inline for mechanical/Haiku-tier leaf tasks (3c gate)277
spec_review_iterations: <N>278
quality_review_iterations: <N>279
voice_gate_score: <% if applicable>280
tokens_used: <approx>281
time_elapsed: <minutes>282
ts: <timestamp>283
```… omitted uncited lines …
326
Skip-conditions: intent=review-only, intent=research-only, intent=plan-only.328
## Integration330
**Reads:**331
- plan.md (canonical, MANDATORY)332
- CORE-PRINCIPLES.md333
- the active pack's compliance hooks (`resolve_pack_field compliance.hooks`; none by default)334
- role file (if active, voice/tone signals only)335
- recent test results336
- `.claude/memory/lessons.md` (via `/li:lessons-surface`, keyword-scoped, non-blocking)338
**Writes:**339
- source code (edits via implementer subagents)340
- WIP commits (if continuous mode)341
- `.claude/runtime/state/build-log.md`342
- `.claude/runtime/state/00-state.md` (BUILD entry per task + final)343
- `.claude/runtime/audit/build-metrics.jsonl`345
**Triggers:**346
- `/li:review` next (or REVIEW in /li:cycle)347
- Hook execution (settings.json hooks fire on edit/commit events)349
## Recommended subagents per task (from plan.md mapping)351
- **TestRunner** (engineering/) — TDD red phase + verification352
- **Refactorer** (engineering/) — refactor phase353
- **Migrator** (engineering/) — schema/API migrations354
- **BackendArchitect / FrontendBuilder** (engineering/) — per task domain355
- **K8sManifestReviewer / TerraformReviewer** (devops/) — infra tasks356
- **CostAnalyzer / LatencyAnalyzer / RegressionDetective** (engineering/) — perf-related tasks357
- **SecurityAuditor / SecretsScanReviewer / ThreatModelDrafter** (security/) — security tasks