For Claude: REQUIRED SUB-SKILL: Use superpowers:executing-plans to implement this plan task-by-task.
Goal: Create a SEA repo skill that guides agents to debug why code generation fails or produces wrong/incomplete output by tracing the spec→AST→IR→manifest→code pipeline.
Architecture: A single skill with two diagnostic tracks (hard failure vs wrong/incomplete output), aligned with SEA spec-first rules and generator-first guardrails. Keep SKILL.md concise and move detailed checks into a short reference file if needed.
Tech Stack: Markdown skill format with YAML frontmatter; repo-local skill under .claude/skills/.
Files:
.claude/skills/debugging-codegen-pipeline/notes/baseline.mdStep 1: Draft 3+ pressure scenarios
Step 2: Run baseline without skill
Step 3: Summarize violations
Files:
.claude/skills/debugging-codegen-pipeline/SKILL.mdStep 1: Create YAML frontmatter
debugging-codegen-pipelineStep 2: Draft core sections
Files:
.claude/skills/debugging-codegen-pipeline/pipeline-checklist.mdStep 1: Add concise stage-by-stage checks
Files:
.claude/skills/debugging-codegen-pipeline/SKILL.mdStep 1: Add a single worked example
Files:
.claude/skills/debugging-codegen-pipeline/SKILL.md.claude/skills/debugging-codegen-pipeline/notes/baseline.mdStep 1: Add rationalization table Step 2: Add explicit counters for each excuse Step 3: Add red flags list
Files:
.claude/skills/debugging-codegen-pipeline/notes/baseline.mdStep 1: Re-run the scenarios with the skill
Step 1: If needed, run packaging
scripts/package_skill.py .claude/skills/debugging-codegen-pipeline