plan-fix-spec - SKILL.md Agent Skill

name: plan-fix-spec description: Sync the implementation plan with the current specs AND iteratively fix all CRITICAL/HIGH issues in both specs and the plan until a clean pass is achieved — using domain-scoped re-reviews, diff-based prompts, issue deduplication, fix verification, and deferred commits to minimize token usage and latency.

Plan Fix Spec

Sync the implementation plan from the current architecture specs, then delegate all review and fix cycles to the fix-implementation skill.

Configuration

Parse both variables from the user's invocation before doing anything else — they will be forwarded to fix-implementation unchanged.

Severity filter (optional, default: CRITICAL+HIGH): controls which issue levels are reported and fixed. If the user specified a list (e.g. critical, critical+high+medium, all), use those levels. If nothing was specified, default to CRITICAL and HIGH.

Phase scope (optional, default: all phases): controls which phases are reviewed and fixed. If the user specified phases (e.g. phase-3, phase-3 phase-5), use those. Otherwise default to all phases.

Process

Step 1 — Product Owner syncs the implementation plan

Launch the prod-owner agent:

You are a Senior Product Owner for AI Town, a 3D city simulator built with C++, Irrlicht, and OpenAL Soft. Read ALL specification files under architecture/ and CLAUDE.md, then update the implementation plan files under ./implementation/ (per-phase files + INDEX.md) so they accurately and completely reflect the current specs. Follow the Core Rule: Spec Consistency and the File Layout rules defined in your agent instructions. When done, confirm what was written and list any spec contradictions you flagged.

Wait for the prod-owner agent to finish before continuing.

Step 2 — Run fix-implementation

Use the Skill tool to invoke fix-implementation, passing the user's original invocation arguments verbatim as the args parameter so fix-implementation receives the same severity filter and phase scope without re-parsing.

fix-implementation handles all remaining work: parallel domain-agent reviews, deduplication, spec and plan fix tracks, iteration until clean, markdown lint, and the single deferred commit.