TDD discipline for Codex: per-lane gates (unit/integration/e2e/contract), spec-strength levels S1-S6 with property tooling, adversarial spec review, red receipts, per-path critical thresholds, 9-format coverage merging and mutation testing, as 11 workflow skills ($tdd-guardian-init, -plan, -design-tests, -implement, -gate, -workflow, ...). A PreToolUse hook denies git commit/push until the gating lanes are fresh; Claude Code's TaskCompleted auto-gate has no Codex event, so $tdd-guardian-gate runs those lanes.
32 skills · 16 commands · 7 agents · hooks
Documentation quality and freshness for Codex: $docs-guardian-audit checks staleness, accuracy, coverage and quality into one report with a fixing plan; $docs-guardian-coverage finds undocumented public APIs; $docs-guardian-generate writes missing docs in the project's framework format (Markdown, MkDocs, VitePress, Docusaurus, Sphinx). A PreToolUse hook warns on, or blocks, code-only commits.
35 skills · 5 commands · 5 agents · hooks
Natural-Language Programming reference knowledge as Codex skills — the 50 Rules, the 100-point scoring rubric, per-tool conventions (Claude Code / Codex CLI / Antigravity), anti-patterns, vocabulary discipline, and authoring guides. Knowledge skills, not commands: load $nlpm-rules + $nlpm-scoring to score an artifact, $nlpm-conventions-codex for the Codex skill layout. The interactive /nlpm:* linting commands remain a Claude Code plugin; cross-tool deterministic checks use the standalone bin/nlpm-check.
17 skills · 15 commands · 8 agents · hooks
English coach for non-native speakers: a UserPromptSubmit hook shows the corrected English of each prompt, translates prompts in other languages, and refines :: prompts, and Codex works from the clean version. Corrections are logged locally for $claude-english-buddy-today, -stats, -mistakes and -drill; $claude-english-buddy-review and -preview check any text for grammar, tone and clarity. The hook calls Claude Haiku with your own Anthropic credentials and runs once you trust it with /hooks.
23 skills · 10 commands · 4 agents · hooks
Personal IT guy for non-technical macOS users — checkups, safe cleanup, undoable file organising, evidence-first diagnosis, backups with a restore drill, home network, automation that leaves reusable tools behind, and a tutor grounded in your machine's own numbers. PreToolUse guard and SessionStart digest both run under Codex. macOS only.
8 skills · 12 commands · 4 agents · hooks
Recursive governance for Executable Operating Units (EOUs). Faceted classification, generating-EOU constraints, ECP-governed change, no-self-approval. v0.7.0: Stage 0 (captured_workflow + domain_values constitutional layer + Rule 96). v0.8.0: agentic judgment (judgment_authorized flag, value_invocations trace, F14–F17 taxonomy, $audit-judgment skill, Rule 97, counterfactual-swap audit).
14 skills · 1 command
Deep codebase interrogation — multi-angle architecture review with 7 specialized analysis skills, 5 review styles (including Paranoid Mode), and 8 add-on pressure tests
8 skills · 1 command · 6 agents
Deep argument interrogation — map a claim, then attack it from five adversarial angles (logic, evidence, counter, assumptions, integrity) via one orchestrator skill ($reason-grill-roast). A six-check calibration gate refuses to manufacture objections, so a sound argument mostly returns [SOUND]; returns a calibrated verdict (HOLDS / HOLDS-IF-NARROWED / WOUNDED / DEFEATED).
8 skills · 1 command · 6 agents
Per-file pure-LOC limits for Codex: $loc-guardian-scan counts with tokei through a tested reducer script (no model arithmetic), flags files over the limit with per-glob overrides and ignores, and proposes extractions whose savings are measured with tokei; $loc-guardian-init sets the limit and extraction rules.
8 skills · 2 commands · 2 agents
XROS — an eXecutable Research Operating System. Four skills: compile a long-horizon investigation into a verifiable methodology spec; sharpen an un-checkable question (orient a newcomer in a field's vocabulary → falsifiable claim → best-available check, with its ceiling); run the Tier-A/B engine (diverse independent routes, adversarial refutation) gated on a real oracle's exit code; or reason through a no-oracle Tier-C claim with premises, pre-mortem, and dated tripwires. Founding rule: no verification, no claim. Ships a bundled dependency-free validator and conformance suite.
4 skills