# Code Review > Perform Codex-style full-repository review for Gait (not PR-limited), with severity-ranked findings focused on regressions, fail-closed safety, determinism, portability, and docs/CLI contract correctness. Use when this capability is needed. - Skill: `tomevault-io/code-review-17` (Agent Skill, multi-file: 2 files) - Install (CLI): `npx skillmds@latest add tomevault-io/code-review-17` - Raw SKILL.md: https://api.skillmd.com/api/skills/tomevault-io/code-review-17/raw - Safety review: pending (external: skill-scanner PASS, skillspector PASS) - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: Coding & Dev Tools - Author: tomevault-io (https://skillmd.com/u/tomevault-io) - Updated: 2026-09-17 - Page: https://skillmd.com/skills/tomevault-io/code-review-17 --- # Full-Repo Code Review (Gait) Execute this workflow for: "review the codebase", "audit repo health", "run a full code review", or "find risks in Gait." ## Reviewer Personality - Contract-first: behavior and guarantees over style. - Boundary-first: controls belong at the execution boundary, not prompt/UI layers. - Regression-first: look for latent breakage paths. - Fail-closed safety bias: block safety/control weakening. - Determinism required: same input/state should produce the same decisions and artifacts. - Scenario-driven: each finding includes concrete break path and impact. - Portability-aware: Linux/macOS/CI/toolchain/path behavior. - Signal over noise: findings-first, severity-ranked output. ## Scope - Repository root: `/Users/tr/gait` - Review entire repo, not only current diffs. - Prioritize high-risk surfaces first, then remaining components. ## High-Risk Surfaces (Priority Order) 1. `core/gate`, `core/contextproof`, `core/pack`, `core/runpack`, `core/regress`, `core/jobruntime` 2. `cmd/gait` CLI behavior, flags, exit codes, JSON outputs 3. `core/mcp` and adapter boundaries 4. `sdk/python` wrapper behavior and error mapping 5. `schemas/v1` and compatibility-sensitive artifacts 6. `docs`, `README.md`, `docs-site`, OSS baseline files, and `.github` workflow/template surfaces ## Workflow 1. Build repository map and contract map from code/tests/help text, including stable/internal/deprecated surfaces, schema/version/CLI/exit-code compatibility rules, authoritative-core boundaries, and docs source-of-truth flow. 2. Run baseline validation where feasible (lint/build/tests) and record gaps if not run. 3. Review each subsystem for: - Enforceable boundary violations, including controls living in prompt/UI paths instead of execution/runtime boundaries - Contract drift before logic drift: schema/version/CLI/exit-code breaks, non-additive evolution, or missing dual-reader compatibility - Public APIs that hide side effects, blur `plan` vs `apply`, or weaken scoped approvals, out-of-band stop, and destructive-budget safety patterns - Missing timeout/cancellation propagation, crash-safe state handling, lock/permission discipline, or contention coverage in long-running flows - Fail-closed gaps where ambiguity in high-risk paths can silently allow instead of block - Determinism or reproducibility breaks across decisions, hashes, traces, diagnostics, and artifacts for the same input/state - Authoritative-core leaks where wrappers/SDKs duplicate policy, signing, verification, or other decision logic - Missing machine-readable failures: stable JSON envelopes, error taxonomy, retryability hints, or consistent error mapping on public boundaries - Missing operator observability: deterministic `doctor` output, correlation IDs, or local structured logs - False-green CI/release paths, including risk-tiered lane gaps (fast/core/acceptance/cross-platform/chaos/perf), weak merge policy wiring, supply-chain verification gaps, or missing post-merge regression monitoring - Weak adoption architecture: unclear quickstart, missing expected outputs/integration diagrams, or troubleshooting-only-afterthought docs - Governance and launch drift: missing ADRs, risk register, explicit non-goals, definition-of-done, or OSS trust-baseline artifacts 4. Verify findings with concrete evidence (file refs, commands, test output). 5. Rank findings by severity and confidence. 6. Report minimum blocker set for safe release posture. ## Severity Model - P0: release blocker, severe safety/integrity break, high reputational risk. - P1: major behavioral regression or control bypass with real user impact. - P2: meaningful correctness/portability/docs-contract issue. - P3: minor maintainability concern. ## Finding Format - `Severity`: P0/P1/P2/P3 - `Title`: short and action-oriented - `Location`: file + line - `Problem`: what is wrong - `Break Scenario`: concrete failure path - `Impact`: user/safety/CI/compliance effect - `Fix Direction`: minimal safe correction ## Review Rules - Findings are primary output; summaries stay brief. - Do not report style nits unless they cause runtime/contract risk. - Do not claim tests/commands were run if they were not. - Separate facts from inference. - Prefer the smallest finding that exposes the broken contract or unsafe boundary. - If no findings, explicitly state `No material findings` and list residual risks/testing gaps. ## Command Anchors - `gait doctor --json` to verify baseline runtime diagnostics and dependency posture. - `gait gate eval --policy --intent --json` to validate policy verdict/exit behavior. - `gait pack verify --json` to check artifact integrity and signature status. ## Output Contract 1. `Findings` (required, ordered by severity) 2. `Subsystem Coverage` (Green/Yellow/Red per major area) 3. `Open Questions / Assumptions` (if any) 4. `Residual Risk / Testing Gaps` 5. `Final Judgment`: - technical health today - minimum blockers (if any) - top 3 risk concentrations - merge/release confidence given current CI, observability, and release-integrity posture --- > Converted and distributed by [TomeVault](https://tomevault.io/claim/clyra-ai) — claim your Tome and manage your conversions.