ship（发布交付）

Ship

Run the full pre-merge pipeline in one command: simplify → code review → adversarial review → finalize → update docs → readiness dashboard → create PR.

No steps to remember. /ship handles everything.

Parse Arguments

Check $ARGUMENTS for flags:

• --status — show the readiness dashboard only, don't run anything
• --adversarial — force adversarial review even on small changes

Step 1: Detect Project Context

# Project name from git remote (or directory name as fallback)
PROJECT="$(basename "$(git remote get-url origin 2>/dev/null | sed 's/.*\///' | sed 's/\.git$//')" 2>/dev/null || basename "$PWD")"

# Current branch
BRANCH="$(git branch --show-current)"

# Base branch — try existing PR first, then repo default
BASE="$(gh pr view --json baseRefName -q .baseRefName 2>/dev/null || gh repo view --json defaultBranchRef -q .defaultBranchRef.name 2>/dev/null || echo main)"

# Readiness script
READINESS="$HOME/.claude/skills/ship/scripts/readiness.sh"

Abort if on the base branch. Nothing to ship — tell the user to create a feature branch first.

Ensure readiness directory exists:

"$READINESS" dir "$PROJECT"

Step 2: Status-Only Mode

If --status was passed:

"$READINESS" dashboard "$PROJECT" "$BRANCH"

Print the dashboard and stop. Do not run any pipeline steps.

Step 3: Run the Pipeline

For each step below, follow this pattern:

1. Check readiness — run "$READINESS" check "$PROJECT" "$BRANCH" <skill>
2. If exit code 0 (clean within 2 hours) — the check outputs the relative time. Print: "✓ <Step>: SKIPPED (done <time>)" and move to the next step.
3. If exit code 1 (not clean) — invoke the skill, then log the result.

Step 3a: Simplify

Use the Skill tool to invoke simplify. This is a plugin skill that reviews changed code for reuse, quality, and efficiency, then fixes issues directly.

After it completes, log the result:

"$READINESS" log "$PROJECT" "$BRANCH" simplify clean

If the skill reports issues it couldn't resolve, log failed instead of clean.

Step 3b: Code Review

Use the Skill tool to invoke code-review:code-review (the plugin skill). This dispatches 5 parallel review agents, scores findings by confidence, auto-fixes mechanical issues, and asks for judgment on critical findings.

Disambiguation: If multiple code-review-style skills are available (e.g. a project-local skill like *-code-review, review, or superpowers:requesting-code-review), do not guess. Stop and ask the user which one to run. The intended default is code-review:code-review, but project-local review skills may be preferred in some codebases.

After it completes, log the result with metadata:

"$READINESS" log "$PROJECT" "$BRANCH" code-review clean \
  critical=0 suggestions=2 auto_fixed=2 coverage_ok=true

Replace the numbers with actual counts from the review output. If there are unresolved critical findings, log status: failed.

If code review fails (unresolved critical findings), stop the pipeline. Show the dashboard and report the failure.

Step 3c: Adversarial Review (Optional)

Determine if adversarial review should run:

# Check diff size
DIFF_LINES="$(git diff --stat "$BASE"..."$BRANCH" | tail -1 | grep -o '[0-9]* insertion' | grep -o '[0-9]*')"

Run adversarial review if:

• --adversarial flag was passed, OR
• DIFF_LINES exceeds 500

If neither condition is met, skip this step (it's informational, not required).

If running, use the Skill tool to invoke adversarial-review with argument --pr.

After it completes, log the result:

"$READINESS" log "$PROJECT" "$BRANCH" adversarial-review clean \
  rounds=2 result=clean

Replace with actual values from the review output.

Step 3d: Finalize

Use the Skill tool to invoke finalize. This runs lint, tests, code quality sweep, documentation check, UI audit (if applicable), and git hygiene.

After it completes, log the result:

"$READINESS" log "$PROJECT" "$BRANCH" finalize clean \
  tests_passed=true lint_passed=true

If tests fail or lint fails, log the appropriate fields as false and status: failed. If finalize fails, stop the pipeline. Show the dashboard and report the failure.

Step 3e: Update Docs (Project-Specific)

Check if the current project has an update-docs skill:

ls .claude/skills/update-docs/SKILL.md 2>/dev/null

If found, use the Skill tool to invoke update-docs.

After it completes, log the result:

"$READINESS" log "$PROJECT" "$BRANCH" update-docs clean

If not found, skip this step. It's informational — absence doesn't block shipping.

Step 4: Show Readiness Dashboard

"$READINESS" dashboard "$PROJECT" "$BRANCH"

Print the full dashboard output.

Large changeset advisory: If the diff exceeds 500 lines and adversarial review was not run, add a note:

"Consider running /adversarial-review — this is a large changeset (N lines)"

Step 5: Verdict and PR Creation

Read the dashboard verdict.

If CLEARED

Proceed to create the PR:

1. Push the branch: git push -u origin "$BRANCH"
2. Generate PR description:

- Read the commit log: git log --oneline "$BASE".."$BRANCH" - Read the readiness log for a summary of what was reviewed - Write a PR title (short, imperative, under 70 characters) - Write a PR body with: - Summary of changes (from commits) - Readiness summary (e.g., "Code review: clean, 2 auto-fixed. Finalize: tests pass, lint clean.") - Test plan

1. Create the PR: gh pr create --title "..." --body "$(cat <<'EOF' ## Summary... ## Readiness... ## Test Plan... EOF)"
2. Report:

- Print the PR URL - Check if the project has a release skill: ls.claude/skills/release/SKILL.md 2>/dev/null - If found: "When ready to deploy, run /release"

If NOT CLEARED

Show what's missing or failed:

• List each required gate that isn't DONE
• Ask: "Ship anyway? (y/n)"
• If user confirms: proceed to PR creation
• If user declines: stop

Error Handling

• No git remote: Run the full pipeline (steps 3a-3e). At step 5, report: "No git remote configured. Run git remote add origin <url> or gh repo create, then run /ship again — all steps are cached and will be skipped."
• gh not authenticated: Report: "GitHub CLI not authenticated. Run gh auth login."
• Step failure: Log status: failed, show dashboard with failure visible, stop the pipeline. Report what failed and why. User fixes the issue and runs /ship again — passed steps are skipped via smart skip.
• Ctrl+C / cancel: Readiness file only has entries for completed steps. Next /ship run picks up naturally.