minhtran3124/agent-harness/finishing-a-development-branch

Trust tier 1
  • community
Source check pending

Checking that the SKILL.md file is still in the repository.

Verify a completed branch, pin required review receipts, push it, and open a pull request. Creates a PR only: never merges, force-pushes, discards work, or amends commits.

Install

npx vskill@latest install minhtran3124/agent-harness/finishing-a-development-branch

SKILL.md

Finishing a Development Branch

Announce this flow. Resolve context once and use its JSON throughout:

bash
python3 scripts/resolve_finish_context.py

The helper reports branch, base, merge base, plan directory, lane, ambiguity, changed files, receipt path, and whether a context-delivery audit is required. If it reports ambiguity, ask the user; if no plan exists, use the tiny/no-plan path instead of guessing.

Verify and gate

  1. Run the repository’s targeted suite, or the full suite if no safe subset is known. The exact

command comes from agents/PROJECT.md → Test execution; never assume a stack or a script name. Fix failures and retry at most twice; never push a failing change.

  1. For non-tiny work with a resolved plan, pin reviews at the resolved base:

``bash python3 scripts/check_review_receipt.py <plan_dir> --require correctness,intent --require-audit-if <base> ``

A missing, stale, failing, or blocking receipt stops the flow. Re-run the affected review; never edit the receipt to pass. Re-run this gate immediately before push. Tiny/no-plan work skips it.

  1. Run python3 scripts/check_open_questions.py --slug <slug>; on findings, tell the user and add an

Open questions closing here list to the PR body (warn mode by default: continue). Run python3 scripts/verify_summary.py --sc-history <slug> --base <base> with the resolved base; relay any finding to the user and the PR body and continue, but an exit 1 under REQUIRE_SC_STABLE=1 stops the flow. It reads committed state only, so commit PLAN.md and ### Deviations edits first; report any exit other than 0 or 1 to the user. Mark the resolved plan status: shipped, append a dated status-log entry, run python3 skills/visual-planner/render_plan.py specs/<slug>/PLAN.md --summarize so the tracked At-a-glance block matches the shipped Status Log, then stage the spec record by name and commit it:

``bash git add specs/<slug>/PLAN.md specs/<slug>/SUMMARY.md specs/<slug>/RUN.json specs/<slug>/events.jsonl ``

(omit a path only if the file does not exist). RUN.json and events.jsonl are tracked artifacts — .gitignore excludes only the .lock — and a fresh RUN.json stays untracked unless named, which the commit gate (Check 1.7) blocks. The helper-selected plan—not a branch-name guess—is authoritative. Do not manually change version/changelog/trust ledger when post-merge automation exists.

Push and PR

Push without force. Create or update one PR against the resolved base, using a concise body with behavioral summary, tasks, and only a useful flow diagram. After creation, best-effort transition the run to ready_to_merge — run python3 runtime/run_state.py transition --slug <slug> --to ready_to_merge --event pr.opened || true (non-fatal). This is the hop the post-merge shipped transition depends on: shipped is legal only from ready_to_merge, so a run left at verifying never terminalizes. The transition rewrites RUN.json and appends to events.jsonl after the last commit, so stage exactly those two paths, commit (chore(specs): record PR #<n> for <slug>), and push once more — the open PR picks the commit up. Without this commit, git keeps the run at its last committed pre-ready_to_merge state and it never terminalizes. Return the URL and stop. A human reviews and merges.

Safety boundary

Never merge, force-push, git add -A, git clean, discard work, or amend a commit. Stage named files only. An external reviewer is recommended for workflow-engine changes but is not a local push gate; the receipt and deterministic checks remain mandatory.

References

  • references/pr-body.md — PR body shape
  • references/workflow-engine-review.md — independent-review guidance
  • scripts/resolve_finish_context.py — context authority

Security

  • Security scanPassedPattern checks on the skill's files.
  • Model reviewPassed

Quality evaluation

  • Task evaluationNot runNo evaluation yet. Run one in Skill Studio.

Source

Popularity

Repository stars
7
7-day trend score
10

Stars and forks are for the whole GitHub repository, not this skill alone. Not refreshed yet.

Works with

Any agent vskill installs to: 56 targets in all. These 39 have a page on this site.

  • Claude Code
  • Cursor
  • GitHub Copilot
  • Windsurf
  • Codex
  • Gemini CLI
  • Cline
  • Amp
  • Roo Code
  • Goose
  • Continue
  • Kilo Code
  • Junie
  • Trae
  • OpenHands
  • Qwen Code
  • AI Agent
  • Tabnine Chat
  • Sourcegraph Cody
  • Amazon Q
  • Firebase Genkit
  • Sweep
  • Devin
  • Phind
  • Mentat
  • Claude Desktop
  • Aider
  • GPT Pilot
  • Plandex
  • bolt.new
  • v0
  • Replit Agent
  • Anse
  • TypingMind
  • BetterChatGPT
  • LibreChat
  • Warp AI
  • Pieces
  • Fig AI
Embed badge
vskill VERIFIED badge
[![vskill VERIFIED](https://verified-skill.com/api/v1/skills/minhtran3124/agent-harness/finishing-a-development-branch/badge)](https://verified-skill.com/skills/minhtran3124/agent-harness/finishing-a-development-branch)

Something wrong with this skill or its page? Tell us.