AGENT SKILL

haiku-adopt

Reverse-engineer an existing feature into H·AI·K·U intent artifacts

License 许可
Apache-2.0
最近更新
2026-11-04
Stars
★ 1
Forks
0

01 前言 Front Matter

name
haiku-adopt
description
Reverse-engineer an existing feature into H·AI·K·U intent artifacts

03 SKILL.md 正文

Adopt

Reverse-engineer an existing feature into H·AI·K·U intent artifacts.

Process

Phase 0 — Pre-checks

  • Check for slug conflicts

Phase 1 — Gather description

  • Get feature description
  • Ask for code paths (specific directories or search whole repo)
  • Ask for git references (PRs, branches, date range)

Phase 2 — Feature exploration (5 parallel subagents)

  1. Code path analysis (modules, entry points, dependencies)
  2. Git history analysis (commit groups, PR boundaries, timeline)
  3. Test analysis (test files, coverage patterns, verified behaviors)
  4. CI configuration analysis (pipelines, quality gates)
  5. Deployment surface analysis (containers, infra, monitoring)

Phase 3 — Propose intent and units (user confirms)

Phase 4 — Reverse-engineer success criteria from tests

Phase 5 — Generate operational plan (if operational surface found)

Phase 6 — Write artifacts (intent.md, unit files, discovery.md, operations/)

Phase 7 — Handoff (summary + next steps)

CRITICAL: MUST NOT modify existing code. Adopted units describe what already exists — they are written with synthetic terminal iterations[] (last entry result: "advance") and signed reviews{} / approvals{} so the cursor sees them as completed and never dispatches a hat against them. Forward-only invariant: completed work is immutable. Wait for user confirmation at each gate (Phase 3, 4, 5).

04 同仓库更多 Skills