This skill is recommendation-only. No edit.
Workflow
- Use every lens below for a broad audit, or only the lens named by the user.
- Inspect the relevant source and installed evidence. Hunt deeply through recent task threads for repeated friction, confusion, detours, failures,
manual work, and context waste. For context audits, run
scripts/analyze-codex-sessions.ts --helpand use its current interface. - For the external-tools lens, search GitHub Trending and current GitHub results for agentic-AI tools that could remove observed friction. Search other current external sources only when needed.
- Rank candidates by recurring future value, confidence, implementation effort, and risk.
- Report the strongest recommendations first. For each, give the evidence, expected recurring benefit, and concrete implementation direction. Mention rejected ideas only when they are genuinely tempting.
Lenses
- Context: always-loaded instructions, session growth, compaction, noisy tool output, screenshots, and avoidable token use.
- Skills: missing, overlapping, stale, broad, duplicated, or script-worthy capabilities.
- Rules: scope, duplication, source-of-truth placement, and workflow guidance that belongs in a skill.
- Install and tooling: install, doctor, shell sync, system tools, helpers, generated drift, and installed configuration.
- Active projects: inventory, paths, environment contracts, lane status, and stale assumptions.
- External tools: trending and current agentic-AI tools that remove repeated work, reduce context, or simplify the setup enough to justify integration cost.