A prompt for auditing stale custom instructions, skills, and memory as models improve
morganb · x · 2026-10-15
morgan shares a prompt he uses to keep his workflow current as models improve: have the AI audit its own auto-loaded context—custom instructions, skills, AGENTS.md, CLAUDE.md, memory—flagging useful knowledge, redundancies, conflicting instructions, obsolete tools, overly broad skill triggers, and stale assumptions.
Key design choices:
- Distinguish confirmed defects from judgment calls; treat file dates as clues, not proof
- Preserve useful domain knowledge and operational safeguards
- Output a read-only audit first with prioritized keep/simplify/verify/retire recommendations, and only clean up after approval
More from coding & agent
- DuoDiscover turns evaluation into agent search, 2x experts on SWE-fficiency — OfirPress · 2026-10-15
- Claude Platform ships OTel traces in beta; CoreWeave Agent Lens ingests natively — altryne · 2026-10-15
- Free LLM judges and automated agent-failure discovery on OTEL traces until Jan 1 — _ScottCondron · 2026-10-15
- micro1's ERU paper finds rater bias and 5-week drift make AI agent evals unreliable — Exp_Mark · 2026-10-15
- ChipAgents and NVIDIA launch Build-A-Chip-Bench: top AI model scores just 16.6% on end-to-end chip design — WilliamWangNLP · 2026-10-15
- Alexandria Launches Product Search for Agents: 50+ Providers, 1 API Key — devdigest · 2026-10-15