codex-toolkit

Agent auditor — audit Claude Code agent definitions for triggering accuracy, system prompt quality, tool appropriateness, and example coverage

Command auditor — audit Claude Code slash commands for schema, workflow clarity, tool selection, error handling, and output specification

Audit→fix→verify loop — finds issues, fixes them, verifies fixes, repeats until clean or you stop

Repo-wide natural language programming auditor — discover and audit all NL artifacts (prompts, skills, agents, commands, rules, hooks, plugins, specs, plans) in any repository

Plugin auditor — audit Claude Code plugin artifacts for schema, specification, security, and structural defects

Rules auditor — audit .claude/rules/ files for enforceability, token efficiency, conflict detection, and correctness risk

Skill auditor — audit Claude Code SKILL.md files for structure, description quality, content effectiveness, and progressive disclosure

Code auditor — full 9-dimension or fast 5-dimension audit with --full/--mini flag

Root cause analysis for user-described bugs using Codex

Cancel a running Codex background job

Continue a previous Codex session — iterate on findings, request fixes, or drill deeper

Send a plan to Codex MCP for full autonomous implementation — choose model, effort, and sandbox level

Initialize codex-toolkit for your project — generates a .codex-toolkit.md config with project-specific defaults

Check Codex connectivity, authentication, and discover available models

Refresh Claude Code convention knowledge — fetch latest docs via context7 and update the reference skill

Fetch stored output from a completed Codex job

Send a plan to Codex for architectural review — checks consistency, completeness, feasibility, ambiguity, and risk

Check Codex readiness and manage the stop-time review gate

Shared: Codex availability test, developer-instructions builder, canonical call pattern, thread handling

Shared: manual analysis fallback when Codex is unavailable or returns empty

Shared: load project config, run preflight discovery, present model/effort/sandbox choices

Shared: resolve plugin root, validate manifest, discover all artifacts, build cross-reference map

Shared: parse audit scope arguments, enforce skip patterns, detect trivial changes

Show active and recent Codex jobs — background tasks, running audits, completed results

Verification auditor — confirms fixes from a previous audit report

Cross-validate Codex audit findings against Claude's native knowledge of Claude Code conventions. Use after any audit command to catch false positives and hallucinated conventions. <example> Context: Codex returned an audit report flagging a plugin's frontmatter as invalid assistant: "I'll use the cross-validator to verify these findings against Claude's native knowledge." </example> <example> Context: User wants to double-check a Codex audit before acting on it user: "Can you verify these audit findings are correct?" assistant: "I'll dispatch the cross-validator agent to check each finding." </example>

Canonical reference for Claude Code plugin artifact schemas, hook events, frontmatter fields, and naming conventions. Used to inject domain knowledge into Codex audit prompts. Run /codex-toolkit:refresh-knowledge to update from latest docs.

3 hooks across 3 events