jupes/agent-forge-harness / .agents/skills/ship
ship
VerifiedRun quality gates, commit, push, and open a PR with the canonical template. Use when the user runs /ship or wants to ship the current changes.
| File | Kind | Size |
|---|---|---|
| SKILL.md | skill md | 5.7 KB |
| agents/openai.yaml | config | 217 B |
SKILL.md
View on GitHub---
name: ship
description: Run quality gates, commit, push, and open a PR with the canonical template. Use when the user runs /ship or wants to ship the current changes.
---
# /ship — Quality Gates → Commit → Push → PR
Runs all quality checks, commits staged/unstaged changes, pushes the branch, and creates or updates a PR.
## Usage
```
/ship # Auto-generate commit message
/ship "add login redirect" # Use provided commit message
/ship --amend # Amend previous commit (use sparingly)
```
---
## Step 1 — Pre-flight Check
```bash
git status # Show what will be committed
git diff --stat # Show changed files summary
```
If working tree is clean (nothing to commit): report "Nothing to ship. Working tree is clean."
---
## Step 2 — Quality Gates
Run all checks. All must pass before committing.
```bash
bun run typecheck # TypeScript — 0 errors required
bun run lint # Lint (skip if no lint script in package.json)
bun test # Tests (skip if no test files exist)
```
**Strict evaluator verdict (optional):** If `AGENT_FORGE_EVAL_VERDICT=strict` is set in the environment where hooks run, the task is shippable only once its run's evaluator verdict is on file and the gate has read it. The quality gate enforces this on `TaskCompleted`, and only for a run it can tie to that Beads issue through a run correlation; without one the strict check fails. Three steps, **with one pointer for all of them**: (1) the run's correlation: a Forge run has one from `forge:phase-gate … --write --bead <TASK-ID>`; otherwise create it **once** with `bun run forge:correlate --bead <TASK-ID>`. Either prints a `pointer`. (2) File the verdict with that pointer: **`bun run forge:verdict --correlation <pointer> …`** (schema 2, written once at the path that run declares: **`.claude/protocols/evaluation-verdict.md`**). (3) Run the gate with the same pointer: `bun run quality-gate --correlation <pointer>`. Running `forge:correlate --bead` again mints a new run, whose verdict has not been filed.
**If any check fails**:
1. Report which check failed and the exact error
2. Fix the issue
3. Re-run the failing check
4. Maximum 2 auto-fix attempts; escalate to user if still failing after 2
---
## Step 3 — Security Scan
Scan the diff for security issues before committing:
```bash
git diff HEAD
```
Check for:
- Hardcoded secrets, API keys, passwords, tokens
- SQL string concatenation
- `eval()` or unsafe dynamic code execution
- Auth checks being bypassed (commented out, `if (true)`)
- Unescaped user input in HTML templates
**If any security issue is found**: stop, fix, do not ship. Never commit secrets.
---
## Step 4 — Commit
Stage all changes:
```bash
git add -A
git status --short # Confirm staged files look right
```
Commit with the provided or generated message:
```bash
# Format: type(scope): description\n\nRefs: <TASK-ID>
git commit -m "<message>"
```
If `--amend` flag: `git commit --amend --no-edit`
### Generating commit messages
If no message was provided, infer from:
1. The Beads task title (if a task is active)
2. The diff summary (most-changed files and purpose)
3. Format: `<type>(<scope>): <description>\n\nRefs: <TASK-ID>`
Types: `feat`, `fix`, `refactor`, `test`, `docs`, `chore`, `build`
---
## Step 5 — PR Stack Detection
Check if the current branch has a stack relationship:
```bash
gh pr list --state open --json number,headRefName,baseRefName
git log --oneline master..HEAD
```
- **If current branch is based off another feature branch**: set `--base <that-branch>` in PR creation
- **If current branch is based off master**: set `--base master`
---
## Step 6 — Push
```bash
git pull --rebase origin <base-branch> # Rebase on latest before pushing
git push origin <current-branch>
```
If push fails due to diverged history:
- `git pull --rebase` and retry once
- If still failing: report the conflict; do NOT force push
---
## Step 7 — Create or Update PR
The PR body MUST follow the canonical template — see **`.claude/skills/pr-description/SKILL.md`**.
Do not hand-roll an ad-hoc body; start from the template, fill every required section with real
content, and validate before posting.
### Check if PR already exists
```bash
gh pr list --head <current-branch> --json number,url
```
### Build the body from the template
```bash
mkdir -p .tmp/work
cp .claude/skills/pr-description/references/pr-template.md .tmp/work/pr-body.md
# Fill every section in .tmp/work/pr-body.md:
# What Changed · Why It's Needed · How It Was Tested · Test Evidence
# (paste real `bun test` / typecheck output) · Risk & Rollback · Linked Issues & AC Trace.
# Pull AC from: bd show <TASK-ID>
bun run .claude/skills/pr-description/scripts/check-pr-body.ts .tmp/work/pr-body.md
# Must print "ok": true before continuing.
```
### Create new PR
```bash
gh pr create \
--title "<TASK-ID>: <task title>" \
--base <base-branch> \
--body "$(cat .tmp/work/pr-body.md)"
```
### Update existing PR
```bash
gh pr edit <number> --body "$(cat .tmp/work/pr-body.md)"
```
---
## Step 8 — Post-Ship
```bash
# Close Beads task
bd close <TASK-ID> 2>/dev/null || true
bd comments add <TASK-ID> "worklog: shipped — PR #<n>" 2>/dev/null || true
# Beads is local-only here: nothing to push (see "Tracker is local-only" in CLAUDE.md)
# Deploy dashboard
bun run build-pages 2>/dev/null || true
# Final status
git status
echo "✓ Shipped: <branch> → PR #<n>"
```
---
## Output
```
Quality gates: PASS (tsc, lint, tests)
Security scan: PASS
Committed: abc1234 — feat(auth): add refresh token rotation
Pushed: origin/feat/T-42-refresh-tokens
PR: https://github.com/org/repo/pull/17 (created)
Beads: T-42 closed
```
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| receiving-code-reviewobra/superpowersUse when receiving code review feedback, before implementing suggestions, especially if feedback seems unclear or technically questionable - requires technical rigor and verification, not performative agreement or blind implementation | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| requesting-code-reviewobra/superpowersUse when completing tasks, implementing major features, or before merging to verify work meets requirements | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| using-git-worktreesobra/superpowersUse when starting feature work that needs isolation from current workspace or before executing implementation plans - ensures an isolated workspace exists via native tools or git worktree fallback | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| verification-before-completionobra/superpowersUse when about to claim work is complete, fixed, or passing, before committing or creating PRs - requires running verification commands and confirming output before making any success claims; evidence before assertions always | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |