Release — Pre-Ship Gate

Final verification before shipping. Maps to Protocol P12 (Release Readiness) and P13 (Security Gate).

Pattern: This skill runs after all beads are closed. It verifies all work is done, security is clean, and the system is ready to ship.

When This Applies

Signal	Action
All beads closed	Run release checklist
User says "/release" or "ship"	Run release checklist
Before production deployment	Run release checklist

Prerequisites

Before running /release:

All execution phases complete (or user explicitly skipping)
No beads in in_progress status
Agent Mail available (if multi-agent)

Tool Reference

Verification Commands

Command	Purpose
`bd list --json`	Verify all beads closed
`bv --robot-summary`	Check for orphaned/blocked beads
`bv --robot-alerts`	Surface any issues
`ubs --staged`	Security scan (MANDATORY)
`pytest` / `npm test`	Run test suite
`git status`	Check for uncommitted changes
`git log --oneline -10`	Recent commit history

Agent Mail (MCP)

Tool	Purpose
`fetch_inbox(agent_name, urgent_only=True)`	Check for unresolved urgent messages
`release_file_reservations(project_key, agent_name)`	Release any held reservations
`send_message(subject="[RELEASE READY]")`	Announce release readiness

Message Subjects

Pattern	When
`[RELEASE READY] Project Name`	All checks passed
`[RELEASE BLOCKED] Project Name`	Blockers found

Architecture

┌─────────────────────────────────────────────────────────────────┐
│                      RELEASE ORCHESTRATOR                        │
│  You are here. You:                                              │
│  - Verify all beads closed                                       │
│  - Run security scan                                             │
│  - Check for uncommitted work                                    │
│  - Validate against rigor tier                                   │
│  - Generate release report                                       │
└─────────────────────────────────────────────────────────────────┘
                              │
         ┌────────────────────┼────────────────────┐
         │                    │                    │
         ▼                    ▼                    ▼
┌─────────────────┐  ┌─────────────────┐  ┌─────────────────┐
│  Bead Status    │  │  Security Gate  │  │  Multi-Agent    │
│  bd list        │  │  ubs --staged   │  │  Cleanup        │
│  bv --robot-*   │  │  Language audits│  │  Agent Mail     │
└─────────────────┘  └─────────────────┘  └─────────────────┘
         │                    │                    │
         └────────────────────┼────────────────────┘
                              │
                              ▼
                    ┌─────────────────┐
                    │  Release Report │ → Present to operator
                    │  GO / NO-GO     │
                    └─────────────────┘

Execution Flow

Step 1: Setup

markdown

1. Identify rigor tier from North Star Card:
   - Read PLAN/00_north_star.md (or project config)
   - Determine Tier 1, 2, or 3

2. Initialize TodoWrite with release phases:
   - [ ] Verify beads complete
   - [ ] Run security scan
   - [ ] Check git state
   - [ ] Multi-agent cleanup
   - [ ] Generate report

Step 2: Verify Beads Complete

bash

# Check for incomplete beads
bd list --json | jq '.[] | select(.status != "closed")'

# Get summary
bv --robot-summary

# Check for alerts
bv --robot-alerts

Blocking if:

Any P0 beads not closed
Any beads in in_progress status (orphaned work)

Step 3: Security Gate (MANDATORY)

bash

# Run security scan
ubs --staged

# For Java code (72% OWASP failure rate)
# ADDITIONAL manual SQL injection review required

# For web code (JS/TS)
# ADDITIONAL manual XSS output encoding review required

Blocking if:

Any high/critical findings
Medium findings without documented justification

Step 4: Run Tests

bash

# Run test suite (project-specific)
pytest                    # Python
npm test                  # Node.js
go test ./...             # Go
cargo test                # Rust

Blocking if:

Any test failures

Step 5: Check Git State

bash

# Uncommitted changes?
git status

# Recent commits look right?
git log --oneline -10

# On correct branch?
git branch --show-current

Blocking if:

Uncommitted changes (unless intentional)
On wrong branch

Step 6: Multi-Agent Cleanup

python

# Check for unresolved urgent messages
urgent = fetch_inbox(
    project_key=PROJECT_PATH,
    agent_name=AGENT_NAME,
    urgent_only=True
)

if urgent:
    # Handle or acknowledge before release
    for msg in urgent:
        acknowledge_message(project_key, AGENT_NAME, msg.id)

# Release any file reservations
release_file_reservations(
    project_key=PROJECT_PATH,
    agent_name=AGENT_NAME
)

# Announce release readiness
send_message(
    project_key=PROJECT_PATH,
    sender_name=AGENT_NAME,
    to=["Coordinator"],  # or all agents
    subject="[RELEASE READY] Project Name",
    body_md="""
Release verification complete.

**Status:** READY TO SHIP
**Beads:** X closed, Y blocked
**Security:** Clean
**Tests:** Passing
    """,
    importance="high"
)

Step 7: Generate Release Report

Present to operator:

markdown

## Release Readiness Report

**Project:** [name]
**Rigor Tier:** [1/2/3]
**Date:** [timestamp]
**Agent:** [agent_name]

### Verification Summary

| Check | Status | Details |
|-------|--------|---------|
| Beads Complete | ✓/✗ | X/Y closed |
| Security (ubs) | ✓/✗ | Clean / N findings |
| Tests | ✓/✗ | All passing / N failures |
| Git State | ✓/✗ | Clean / uncommitted |
| Agent Mail | ✓/✗ | No urgent / N unresolved |

### Blockers
[List any blocking issues, or "None"]

### Recommendation
**[READY TO SHIP]** or **[BLOCKED - see above]**

### Next Steps
- [ ] Operator sign-off
- [ ] Tag release: `git tag -a vX.Y.Z -m "Release X.Y.Z"`
- [ ] Deploy to [environment]
- [ ] Monitor logs/metrics

Release Checklist by Rigor Tier

Tier 1 (Personal/Hobby)

Check	Required
All planned beads closed	✓
`ubs --staged` clean	✓
Basic smoke test	✓
No uncommitted changes	✓

Tier 2 (Startup MVP / Business Tool)

Check	Required
All P0/P1 beads closed	✓
All tests passing	✓
`ubs --staged` clean	✓
Medium findings documented	✓
No secrets in codebase	✓
P0 requirements verified	✓
Basic observability	✓
Operator sign-off	✓

Tier 3 (Enterprise)

Check	Required
All beads closed	✓
Full test suite passing	✓
`ubs --staged` zero findings	✓
Language-specific audits	✓
All REQ-* traced to tests	✓
Threat model reviewed	✓
Observability complete	✓
Rollback plan tested	✓
Audit trail complete	✓
Stakeholder sign-off	✓

Agent Mail Protocol

Subject	Sender	Recipients	When
`[RELEASE READY] Project`	Release Agent	Coordinator, All	All checks passed
`[RELEASE BLOCKED] Project`	Release Agent	Coordinator	Blockers found

Error Handling

Beads Not Complete

List incomplete beads with bd list --json
Present to operator: complete or descope?
If descoping: document decision, update requirements
Re-run release check

Security Findings

List findings from ubs --staged
For high/critical: MUST fix before release
For medium: document justification or fix
Re-run after fixes

Test Failures

List failing tests
Fix failures or document known issues
Known issues require operator sign-off
Re-run after fixes

Orphaned File Reservations

Check for stale reservations
Force release if agents are inactive
Document in release notes

Post-Release

After shipping:

bash

# Tag the release
git tag -a v1.0.0 -m "Release 1.0.0"
git push origin v1.0.0

# Update beads (if any tracking beads remain)
bd close <release-bead-id> --reason "Released v1.0.0"

python

# Notify team
send_message(
    project_key=PROJECT_PATH,
    sender_name=AGENT_NAME,
    to=["all"],
    subject="[RELEASED] v1.0.0",
    body_md="Release complete. Monitoring for issues.",
    importance="high"
)

Mandatory Rules (2025 Research)

Rule	Why	Evidence
`ubs --staged` mandatory	~40% of LLM code has vulnerabilities	`research/052-llm-security-vulnerabilities.md`
Java: extra SQL review	72% OWASP failure rate	`research/061-llm-security-2025.md`
JS/TS: XSS review	Only 14% XSS caught	`research/061-llm-security-2025.md`
Human verification for P0	AI helps most when human understands domain	`research/051-metr-rct.md`

Evidence Base

Research	Finding
`research/021-swe-bench-plus.md`	Benchmark hygiene matters; don't trust passing tests alone
`research/047-humaneval-pro.md`	Progressive verification catches drift
`research/044-iris.md`	Security verification is essential
`research/052-llm-security-vulnerabilities.md`	~40% of LLM code has vulnerabilities
`research/061-llm-security-2025.md`	Language-specific vulnerability rates
`research/051-metr-rct.md`	Human verification required for production

Search AI Tools

release

Install this agent skill to your Project

SKILL.md

Release — Pre-Ship Gate

When This Applies

Prerequisites

Tool Reference

Verification Commands

Agent Mail (MCP)

Message Subjects

Architecture

Execution Flow

Step 1: Setup

Step 2: Verify Beads Complete

Step 3: Security Gate (MANDATORY)

Step 4: Run Tests

Step 5: Check Git State

Step 6: Multi-Agent Cleanup

Step 7: Generate Release Report

Release Checklist by Rigor Tier

Tier 1 (Personal/Hobby)

Tier 2 (Startup MVP / Business Tool)

Tier 3 (Enterprise)

Agent Mail Protocol

Error Handling

Beads Not Complete

Security Findings

Test Failures

Orphaned File Reservations

Post-Release

Mandatory Rules (2025 Research)

Evidence Base

See Also