What problem does it solve? Maintaining a security-sensitive Rust/axum harness requires disciplined, evidence-first optimization work; ad-hoc changes risk violating architectural invariants, weakening write gates, or duplicating open PRs. This Skill provides a structured scan-to-draft-PR workflow that keeps improvements safe and reviewable. ## Core Features & Use Cases - Time-boxed read-only scan: Sweeps src/agentic, src/server, src/shim, src/llm, CI workflows, and tests to produce 6-10 cited findings grouped into 1-4 earned PR-sized chunks. - Invariant-guarded implementation: Enforces harness guardrails (I6 shim isolation, guard chain order, write gates, clone jail) and requires explicit invariant statements for core-path changes. - Publication discipline: Dedups against open PRs, plans shared-file topology (consolidate or stack), and opens draft PRs only when authorized, verified with cargo fmt, clippy, tests, and check-pr-template.sh. - Use Case: Ask the agent to optimize the repository; it scans the default branch, reports findings with file/line citations, and—once you authorize scope—implements a focused chunk and opens a draft PR. ## Quick Start Ask the agent to run a read-only optimization scan of the CG-agent-harness repository and report grouped findings before implementing anything.
