feat: /build skill — plan-to-implementation orchestrator

[jeff-artemys]

Summary

• Adds /build, a new skill that closes the workflow gap: plan → review → build → ship
• Reads plan artifacts from /plan-ceo-review, /plan-eng-review, and /plan-design-review
• Uses TeamCreate + Agent for multi-agent parallel implementation on larger plans
• Falls back to single-agent sequential mode for small plans (< 5 tasks)
• Includes a Code Standards section establishing the "every line of code is a liability" philosophy — propagated into every spawned agent's prompt
• Registers the template in gen-skill-docs.ts so CI freshness checks pass

How it works

/plan-ceo-review  ─┐
/plan-eng-review  ─┼──▶  /build  ──▶  /ship
/plan-design-review┘

1. Gather — reads CEO plan, eng review, test plan, design doc from ~/.gstack/projects/
2. Decompose — breaks the plan into discrete work units with dependency graph
3. Assemble — spawns a team of agents sized to the plan's parallelism (1-4 agents)
4. Orchestrate — assigns tasks in dependency waves, handles blockers, tracks progress
5. Verify — runs tests, checks coverage against test plan, validates plan conformance
6. Handoff — outputs build summary, persists log, ready for /ship

Code Standards philosophy

The skill establishes and enforces code quality principles on all generated code:

• The Liability Principle — every line, file, abstraction, and dependency must be earned. Less is more.
• Explicit over clever — code should be so simple it's obviously correct, not so clever it has no obvious bugs.
• What clean does NOT mean — no abstractions for testability, no interfaces with one implementation, no defensive code against impossible internal states. Validate at boundaries, trust internal data.
• Standards are included in every spawned agent's prompt so multi-agent builds maintain quality.

Design decisions

• No new placeholders — uses existing {{PREAMBLE}}, {{BASE_BRANCH_DETECT}}, {{REVIEW_DASHBOARD}}
• Single-agent for small plans — TeamCreate overhead isn't worth it for < 5 sequential tasks
• Self-contained task descriptions — agents only see their task, not the full plan context
• Wave-based parallelism — dependency graph determines what runs in parallel vs sequential
• Trust the plan — /build executes, it does not re-review. Scope debates belong in the review phase
• Persists build log to reviews.jsonl so /ship review dashboard can see the build ran

Test plan

• bun run gen:skill-docs generates build/SKILL.md without errors
• bun run gen:skill-docs --dry-run shows FRESH for all files
• bun test — 383 tests pass, 0 failures
• Skill parser validates all $B commands (none used — this is an orchestration skill)
• Manual: invoke /build after running /plan-eng-review on a feature branch

Files changed

File	Change
build/SKILL.md.tmpl	New skill template (~490 lines)
build/SKILL.md	Generated from template
scripts/gen-skill-docs.ts	Register build template in candidate list (+1 line)