project-context-generation

Guidance for analyzing a repository and creating concise, high-value AGENTS.md and related instruction context for future agent sessions.

SKILL.md

GitHub

Core approach

  • Build a mental model of the repository first.
  • Capture only the instructions future sessions are likely to need repeatedly.
  • Prefer concise, actionable guidance over long prose.
  • Focus on facts visible from the repo unless a key ambiguity requires a clearly labeled assumption.

What to capture

  • build, test, lint, and verification commands
  • command ordering when it matters
  • project structure and module boundaries that are not obvious from names alone
  • important stack conventions, coding patterns, and workflow expectations
  • setup quirks, operational gotchas, and environment assumptions
  • references to other existing rules or standards files when they should be pulled in through config
  • code intelligence output (from hero scan) — package structure, key symbols, dependency graph, and hot files that help orient future sessions

Guardrails

  • Do not restate the whole codebase.
  • Do not add generic advice that is not specific to the project.
  • Improve existing AGENTS.md in place when present.
  • Keep the file useful for future agent execution, not as a general project README.