Review For Ship

Pre-Ship Review

Run all relevant review agents for this project's tech stack, then invoke rad-code-review as the final gate. This is the comprehensive quality gate — use it before deploying, merging a major feature, or shipping a release.

Announce at start: "Running pre-ship review — dispatching specialist reviewers for your stack, then rad-code-review as final gate..."

Step 1: Read Stack Profile

Read .claude/stack-profile.local.md from the project root.

If the file does not exist: Tell the user: "No stack profile found. Run /detect-stack first to scan your project and identify which review agents apply. Then come back and run /review-for-ship." Stop here.

If the file exists: Parse the YAML frontmatter. Extract:

• project_name
• project_type
• technologies (the list)
• review_agents (the list of agents with their subagent_type values)
• plugins.core, plugins.supporting, plugins.pre_ship, plugins.final_gate

Profile Drift Warning

Parse the detected date from the YAML frontmatter. If it is older than 90 days:

"Your stack profile is [N] days old. Dependencies and technologies may have changed. Consider running /detect-stack --update to refresh before review. Continue with current profile?"

Wait for user confirmation. If they want to update, stop and suggest /detect-stack --update.

Step 2: Confirm with User

Before dispatching agents, show the user what will run:

## Pre-Ship Review Plan

**Project:** [project_name]
**Type:** [project_type]
**Stack:** [comma-separated technology names]

### Phase 1: Specialist Reviewers (run in parallel)
[For each agent in review_agents:]
- **[agent name]** — [one-line description of what it reviews]

### Phase 2: Final Gate
- **rad-code-review** — AI slop detection, architecture, security, release readiness verdict

This will run [N] specialist agents in parallel, then rad-code-review as the final pass.

Proceed? (This uses context and tokens proportional to the number of agents.)

Wait for user confirmation before proceeding. If the user says no or wants to skip certain agents, adjust accordingly.

Step 3: Dispatch Specialist Review Agents

Launch all specialist review agents in parallel using the Agent tool. For each agent:

• Set subagent_type to the value from the stack profile
• Set run_in_background to true so all agents run concurrently
• Give each agent a clear prompt:

Review the project at [current working directory] for [what this agent reviews].
This is a [project_type] project using [technology list].
Scan the full codebase and report findings organized by severity (Critical, Warning, Info).
Be thorough but practical — focus on issues that would cause real problems in production.

Available Specialist Review Agents

Only dispatch agents that are listed in the stack profile's review_agents field:

Step 4: Final Gate — rad-code-review

After all specialist agents have completed and their results are collected, run rad-code-review as the final gate.

Invocation

Invoke the rad-code-review skill with these parameters:

• Scope: the current project (full codebase)
• Strictness: production (default) — user can override to mvp or public
• Let rad-code-review run its full workflow: bug finder, architecture reviewer, release gate

What the Final Gate Covers

rad-code-review provides the generalist layer that complements the specialist reviewers:

1. AI Slop Detection — 14 patterns for hallucinated imports, fake error handling, placeholder stubs, cargo-cult patterns, silent failures
2. Architecture Review — Coupling, boundaries, naming, extensibility, test gaps
3. General Security �� OWASP 12 categories (catches gaps the specialists might miss)
4. Release Readiness — Config separation, secrets, dependencies, migrations, monitoring, documentation
5. Documentation — README, setup instructions, env var documentation
6. Dependency Audit �� CVEs, outdated packages, lockfile integrity

Final Gate Verdict

rad-code-review's release verdict becomes the ship/no-ship decision:

• Ship — No critical or major findings blocking release
• Conditional — Issues found but not release-blocking at the specified strictness
• No Ship — Critical or major findings that block release

Step 5: Compile Unified Report

Once all specialists AND the final gate have reported, compile a unified report.

Report Format

# Pre-Ship Review Report

**Project:** [name]
**Date:** [today's date]
**Stack:** [technologies]
**Specialist Agents:** [count] run
**Final Gate:** rad-code-review (strictness: [level])

---

## Overall Assessment

**Verdict:** [Ready to Ship / Needs Work / Critical Issues Found]

| Source | Critical | Warning | Info |
|--------|----------|---------|------|
| [Each specialist agent] | [n] | [n] | [n] |
| rad-code-review (final gate) | [n] | [n] | [n] |
| **Total** | **[n]** | **[n]** | **[n]** |

---

## Specialist Findings

### [Agent Name]
**Status:** [Clean / Issues Found]

[If issues found, list organized by severity:]

#### Critical
- [Issue description + file:line + recommended fix]

#### Warnings
- [Issue description + file:line + recommended fix]

#### Info
- [Improvement suggestion]

[Repeat for each specialist agent]

---

## Final Gate: rad-code-review

**Verdict:** [Ship / Conditional / No Ship]
**Strictness:** [mvp / production / public]

### Findings
[rad-code-review findings organized by its standard severity model]

### AI Slop Detection
[Any AI-generated code issues flagged]

### Release Readiness
[Release gate assessment]

---

## Recommended Actions

Priority order for fixing issues before shipping:

1. **[Critical item from any source]** — [why it matters, what to do]
2. **[Critical item]** — [why it matters, what to do]
3. **[Warning item]** — [why it matters, what to do]
...

---

## What's Next?

[Based on overall verdict:]
- **Critical Issues Found:** "Fix the critical issues above, then run /review-for-ship again to verify."
- **Needs Work:** "The warnings above are worth addressing but may not be blockers depending on your timeline."
- **Ready to Ship:** "Your project looks good! No critical issues found across all reviewers and the final gate."

Overall Verdict Logic

1. If rad-code-review (final gate) says No Ship �� overall = Critical Issues Found
2. If any specialist has Critical findings ��� overall = Needs Work
3. If rad-code-review says Ship and no specialist Criticals ��� overall = Ready to Ship
4. If rad-code-review says Conditional and no specialist Criticals → overall = Needs Work

Step 6: Offer Follow-Up

After the report, offer:

• "Want me to fix any of these issues?" (if issues were found)
• "Want me to run a specific reviewer again after fixes?" (for targeted re-check)
• "Want me to save this report to a file?" (write to .ucr/history/ or user-specified path)

Edge Cases

• Specialist agent fails or times out: Note which agent failed, report results from successful ones, suggest re-running the failed agent individually.
• rad-code-review fails: Report specialist results without the final gate. Note that the release verdict is incomplete.
• No review agents in profile: Still run rad-code-review as the final gate (it works on any project). Note that specialist coverage is missing and suggest /detect-stack --update.
• User wants partial review: Only dispatch the requested subset. Still offer the final gate unless explicitly skipped.
• User wants to skip final gate: If the user says "skip code review" or "specialists only", compile the report from specialist findings only.