Reinforce Skills & Agents

Inject compressed skill-mapping and agent-mapping directives into a project's CLAUDE.md file so skill names, agent IDs, and invocation triggers persist across the entire session without being lost to context drift.

Problem

LLMs lose track of skill names, agent IDs, and invocation details as conversation context grows. Skills and agents load at session start but fade from working memory mid-session, causing the agent to guess names, skip invocations, or fail to delegate to the right specialist.

Solution

Vercel's research found that passive context embedded in project files outperforms active skill retrieval — 100% vs 79% pass rate in agent evals. Embed a compressed skill map directly in CLAUDE.md using the same directive pattern.

Reference: https://vercel.com/blog/agents-md-outperforms-skills-in-our-agent-evals

For detailed research findings, consult references/vercel-research.md.

File Placement