| name | auditing-agent-infra |
| description | Audits the AI coding agent ecosystem and compares against a project's current agent infrastructure. Researches Claude Code, Cursor, Codex, Gemini CLI, and other tools for new features, skill standards, and best practices, then produces a gap analysis with proposals. Use when asked to audit agent infrastructure, compare agent tools, or review agent setup. |
Agent Infrastructure Audit
Process
Audit checklist:
- [ ] Snapshot project's current agent setup
- [ ] Research each platform's latest state
- [ ] Compare findings against project infrastructure
- [ ] Identify gaps and rate by impact/effort
- [ ] Draft report and present for review
1. Snapshot current state
Read the project's agent configuration — look for CLAUDE.md, AGENTS.md, .claude/skills/, .cursor/rules/, copilot-instructions.md, or equivalent files.
2. Research platforms
For each platform, investigate the latest state of:
- Claude Code — skills discovery, MCP integrations, hooks, memory systems
- Cursor — rule format changes, memory features, agent mode, MCP support
- OpenAI Codex — AGENTS.md conventions, memory, skill format, sandbox capabilities
- GitHub Copilot — instruction formats, agent mode, MCP support
- Gemini CLI — conventions, agent capabilities
- Agent Skills standard — spec changes, new tooling, distribution methods
Cross-cutting concerns:
- Memory and knowledge persistence (new standards?)
- Skill/tool distribution (package managers, registries)
- MCP server ecosystem (relevant to project's stack)
- Context window management best practices
3. Compare and report
Use web search and web fetch. Prioritize official docs and release notes.
Report template
# Agent Infrastructure Audit — <date>
## Current Setup
[Project's agent config, installed skills, memory setup]
## Ecosystem Updates
[Per-platform findings — only notable changes]
## Gap Analysis
| Area | Current | Available | Priority |
|------|---------|-----------|----------|
| ... | ... | ... | High/Med/Low |
## Proposals
[Numbered, actionable recommendations with effort estimates]
## Sources
[Links to docs, release notes consulted]
