launch-review

Launch Review Pipeline

A structured, multi-phase agent pipeline that evaluates feature readiness through adversarial debate, creates a reviewed action plan, and executes fixes. Each phase runs as an isolated team to prevent context bloat.

Invocation

/launch-review [topic] [--scope "focus areas"] [--constraints "known constraints"]

Examples:

• /launch-review "checkout experience" --scope "cart, payment, confirmation"
• /launch-review "v2 API launch" --constraints "backwards compatible, feature flag gated"

Phase 0: Constraints Gathering (No Team)

Before spawning any agents, the main agent MUST:

1. Ask the user for known constraints:
   • Legal restrictions (prohibited libraries, data sources, licenses)
   • Business rules (budget limits, timeline, must-have vs nice-to-have)
   • Scope boundaries (what's in/out of this review)
   • Any prior decisions already made
2. Check memory for relevant project constraints
3. Compile a constraints block that gets injected into every agent's prompt

Format:

CONSTRAINTS (non-negotiable):
- [constraint 1]
- [constraint 2]
DO NOT propose solutions that violate these constraints.

This phase prevents the #1 failure mode: agents recommending something that's already been ruled out.

Phase 1: Debate Team

Create team: TeamCreate with name {topic}-debate

Agents (5):

Agent	Role	Type
arch-critic	Code auditor -- finds gaps, bugs, half-baked features	general-purpose
comp-critic	Competitive benchmarker -- researches competitors, holds us to that bar	general-purpose
code-defender	Argues launch readiness from actual code evidence	general-purpose
industry-defender	Argues competitive positioning from market research	general-purpose
moderator	Synthesizes all 4 reports, writes verdict	general-purpose

Flow:

1. Spawn all 5 agents. The 4 debaters run in parallel. Moderator is blocked by all 4.
2. Each debater sends their report to moderator via SendMessage.
3. Moderator writes the report to docs/{TOPIC}_READINESS_REPORT.md.
4. Moderator sends summary to team-lead.