<!-- JARVIS skill: semantic code intelligence via Serena (oraios/serena, GPL-3.0). Serena is used as an INDEPENDENT TOOL — its code is not copied into any MIT repo. Binary: ~/workspace/.venv-tools/bin/serena -->

Serena — Semantic Code Toolkit

Serena exposes LSP-powered code intelligence over MCP (stdio) or CLI. Use it as an external tool; never copy its code into MIT-licensed repos.

Quick CLI usage

SERENA=~/workspace/.venv-tools/bin/serena
$SERENA --help
# Index a project, then query symbols/references via its MCP or CLI surface

Typical agent flows

1. Blast-radius before a change: find the symbol, list references and callers, then decide the edit scope. (Complements codegraph's static graph with live LSP accuracy.)
2. Precise refactor: rename/move a symbol across files via LSP instead of sed/grep, then run the project's typecheck/tests.
3. Onboarding: ask for the definition/call sites of an unfamiliar symbol rather than reading whole files.

Rules

1. Serena is GPL-3.0: run it, don't vendor it. Findings (facts about code) are fine to use; its source stays out of MIT repos.
2. Prefer Serena over grep when the question is about symbols (definitions, references, renames). Prefer grep/ast-grep for plain text searches.
3. After any Serena-assisted refactor, run the repo's own verification (typecheck, tests) — LSP edits are precise, not infallible.