How it works

Three steps from a file tree to AGENTS.md, .cursorrules, CLAUDE.md and related agent instruction files.

Step 1

Paste your repository snapshot

Run git ls-files (or find excluding node_modules) and paste the tree. Optionally add package.json, an entry point, or CI config to upgrade inferences to detections.

Step 2

Deterministic extract, then optional prose

Stage one always runs: stack, commands, entry points, conventions. Stage two is model-assisted prose when AI is configured — otherwise a labelled rule-based draft.

Step 3

Review Context Score and commit carefully

Check open questions and every fact marked inferred. Copy the target files into the repo only after you correct anything that does not match reality.

Honesty: we do not guarantee agents stop hallucinating. See Security for fail-closed AI behaviour and OWASP LLM01 notes.