| name | rlm-scan |
| description | Codebase scanner — architecture mapping, security scan, pattern search across a repo. Cached index auto-loads on session start so Claude doesn't re-learn the codebase every session. |
| context | fork |
| policy_doc_kind | skill |
| classification | canonical |
| canonical_owner | self |
| authority_level | procedural |
| in_verifier_scope | true |
RLM Scan
Recursively decomposes a codebase using rg + ast-grep + semgrep, calls Claude
at each level with bounded context, and caches results to project memory. Future
sessions load the summary automatically via the SessionStart hook.
Invocation
/rlm-scan # scan current directory, general type
python3 ~/.claude/bin/rlm_scan.py <path> [--type general|security|architecture] [--force]
Scan types
| Type | What it does |
|---|
general | Architecture, module roles, dependencies, patterns |
security | Runs semgrep (p/default ruleset) + security-focused analysis |
architecture | High-level structure, coupling, layer boundaries |
When to run
- First time working in a new codebase
- After a major refactor (use
--force)
- Before a security review (
--type security)
- When Claude says "I'm not familiar with this repo"
Do not run on every commit — the incremental hash check handles file-level changes
automatically. Re-scan only when structure changes significantly.
Cache location
~/.claude/projects/<encoded-path>/memory/rlm_scan_<type>.json
~/.claude/projects/<encoded-path>/memory/rlm_scan_<type>_summary.md
Session auto-load
The rlm_session_preflight.py hook fires on SessionStart and injects the scan
summary into Claude's context if a cache exists for the current working directory.
Claude sees it as a [RLM Scan — ...] notice at the top of the session.
Interpreting results
- project_summary: Overall description; use as mental model anchor
- key_findings: Highest-priority items (security issues, architectural concerns)
- modules: Per-directory breakdowns with local summaries and findings
- file_hashes: Used for incremental re-scan; not directly useful for analysis
Workflow
When starting work on an unfamiliar codebase:
- Run
/rlm-scan on the root
- Check
key_findings for known risks before making changes
- Reference module summaries when navigating to unfamiliar directories
- Re-scan changed areas with
--force after significant refactors
