hoklims/semctx

Govern code-intelligence routing, freshness, refresh, consumer generations, provider cost, and fallback across Codex and Claude Code. Use this skill whenever a task asks which code index to use, reports slow or stale indexing, touches CCC/CocoIndex, Graphify, Semctx, Serena, LSP or SCIP, changes index hooks or MCP exposure, opens a substantial Git worktree, or needs architecture/impact retrieval. Prefer this control plane even when the user only says the agent is slow, cannot find code, or seems to use the wrong context.

Use semctx through its MCP tools for top-down root-cause diagnosis, repository impact analysis, authored goals and invariants, proof-carrying change contracts, handoffs, bounded target proposals, and fail-closed migration planning. Use for non-trivial code changes, refactors, migrations, architecture reconstruction, semantic trace requests, invariant preservation, generic project demonstrations, or pre-commit verification in a semctx-enabled repository.

Carry intent, invariants, decisions, evidence and unknowns through a non-trivial change using the semctx semantic layer. Use when starting substantial work: open or select a change contract, pull a bounded semantic slice, then keep the contract honest as you edit — verify impact, compose the change verdict, run the recommended tests, and record proofs/unknowns. Never conclude on a BLOCK.

Verify the semantic blast radius of a code change before finishing. Use after any non-trivial edit and before a commit or push: check impacted invariants/contracts and run the recommended tests. Invoke when the user asks to commit, open a PR, or "make sure this change is safe".