Review Agentic Setup

Detect agentic coding infrastructure and flag gaps across Claude Code and Codex CLI conventions. Analysis only. Does not install or configure anything.

Scope

Agentic setup review always operates at the project level. Scope parameters (diff commands, file lists) are accepted but ignored.

When called standalone, use the git repository root as the project root (fall back to the current working directory if not in a git repo).

Step 1: Detect Infrastructure

Search for agentic coding configuration in the project. Classify findings into five categories:

Agent Instructions

File	Read by	What to check
CLAUDE.md (project root)	Claude Code	Exists? Sections present (project structure, conventions, build commands, rules)?
CLAUDE.md (subdirectories)	Claude Code	Any nested CLAUDE.md files for sub-module guidance?
AGENTS.md (root and subdirectories)	Codex	Exists? Content areas (setup, conventions, testing, architecture)?
AGENTS.override.md	Codex	Exists? Overrides AGENTS.md at the same directory level.
README.md	Both	Mentions AI-assisted development, agent workflows, or prompting guidance?

Read each file that exists and summarize its coverage areas. Note which tools can use it.

Claude Code does not read AGENTS.md directly. Projects share instructions across both tools by:

• Import: CLAUDE.md includes @AGENTS.md to pull in shared content
• Symlink: CLAUDE.md symlinked to AGENTS.md (or vice versa) so both tools read the same file

Check for both patterns when the project has agent instruction files.

Installed Skills

Check both skill locations. For each skill directory, read the SKILL.md frontmatter to extract name and description. Group skills by type:

Location	Used by
.claude/skills/	Claude Code
.agents/skills/	Codex

Skill types:

• Analysis — review, audit, or inspection skills
• Workflow — multi-step process skills
• Pipeline — skills that compose other skills
• Utility — formatting, staging, committing, or other focused tools

Both tools use the same SKILL.md format (YAML frontmatter with name and description). Projects can symlink one location to the other (e.g., .agents/skills/ → .claude/skills/) to share skills across both tools. Check for symlinks.

MCP Servers

File	Used by	What to check
.mcp.json (project root)	Claude Code	Configured MCP servers with their types (stdio, http)
.claude/settings.json	Claude Code	MCP server entries in project settings
.claude/settings.local.json	Claude Code	MCP server entries in local settings (gitignored)
.codex/config.toml ([mcp_servers])	Codex	MCP server entries in project config

List each detected server by name, type, and which tool configuration it belongs to.

Hooks

File	Used by	What to check
.claude/settings.json	Claude Code	Hook configurations (PreToolUse, PostToolUse, etc.)
.claude/settings.local.json	Claude Code	Hook configurations in local settings (gitignored)
.codex/hooks.json	Codex	Hook configurations (PreToolUse, PostToolUse, etc.)

List each detected hook with its trigger event, command, and which tool configuration it belongs to.

Documentation References

Search CLAUDE.md and AGENTS.md content for file paths, URLs, or section headings that reference documentation. Also check the project root for:

File or directory	What it indicates
docs/ directory	Project documentation
CONTRIBUTING.md	Contributing guidelines
TROUBLESHOOTING.md	Known issues and fixes
ARCHITECTURE.md	Architecture or design docs
API docs or schema files referenced in agent instructions	API reference material

Step 2: Analyze Gaps and Compatibility

For each category, assess agentic readiness:

• Present and configured — infrastructure detected with meaningful content
• Partially configured — file exists but content is minimal or boilerplate
• Missing — no infrastructure detected for this category

Gap analysis considers the project's complexity. A single-file script doesn't need installed skills. A multi-module application benefits from CLAUDE.md with build commands and architecture notes.

Cross-Tool Compatibility

After detection, assess which agentic tools the project supports:

Signal	Compatibility
Has AGENTS.md but no CLAUDE.md	Codex-ready only. Claude Code does not read AGENTS.md.
Has CLAUDE.md but no AGENTS.md	Claude Code-ready only. Codex does not read CLAUDE.md.
Has both, but CLAUDE.md lacks @AGENTS.md import	Both files exist but instructions are siloed. Claude Code misses AGENTS.md content.
Has both, and CLAUDE.md imports @AGENTS.md	Both tools supported. Shared instructions with tool-specific additions.
Instruction files are symlinked	Both tools supported. Identical instructions.
Has .claude/skills/ but no .agents/skills/	Skills available to Claude Code only.
Has .agents/skills/ but no .claude/skills/	Skills available to Codex only.
Skill directories are symlinked	Both tools supported. Shared skill set.
MCP configured in one tool's config only	MCP servers available to that tool only.

Flag any compatibility gaps as findings. A project that invests in one tool's config but not the other has a portability gap that may matter for teams using both.

Output Format

Return findings as a numbered list. For each finding:

### [P<N>] <title (imperative, <=80 chars)>

**Category:** <Agent Instructions | Installed Skills | MCP Servers | Hooks | Documentation References | Compatibility>

<one paragraph: what is missing or underdeveloped and why it matters for agentic workflows>

After all findings, add:

## Agentic Setup Summary

| Category | Status | Details |
|---|---|---|
| Agent Instructions | <Present/Partial/Missing> | <detected files or "—"> |
| Installed Skills | <Present/Partial/Missing> | <N skills or "—"> |
| MCP Servers | <Present/Partial/Missing> | <detected servers or "—"> |
| Hooks | <Present/Partial/Missing> | <detected hooks or "—"> |
| Documentation References | <Present/Partial/Missing> | <referenced docs or "—"> |

## Tool Support

| Tool | Supported | Key config |
|---|---|---|
| Claude Code | <Yes/Partial/No> | <detected Claude Code files> |
| Codex CLI | <Yes/Partial/No> | <detected Codex files> |

## Overall Verdict

**Agentic readiness:** <well-equipped | gaps found | not configured>

<1-3 sentence summary>

If all categories are adequately covered, report that and highlight what the project does well.

Priority Levels

• P0 — No agent instruction files (CLAUDE.md, AGENTS.md) in a project with meaningful source code
• P1 — CLAUDE.md exists but missing build/test commands or project structure
• P2 — No installed skills, or agent instructions don't reference available documentation
• P3 — No MCP servers, no hooks, or minor coverage gaps