agent-repo-preflight: Open-Source Read-Only Audit of AGENTS.md, MCP and Hook Configs Before Cloned Repos Hit Your Coding Agent
Digital-Fortress-Sec · reddit · 2026-10-28
Digital Fortress released agent-repo-preflight (v0.1.0, MIT, stdlib-only Python), a read-only script that audits what a freshly cloned repo would feed your coding agent: rule files (AGENTS.md, CLAUDE.md, .cursorrules), declared MCP servers, and hooks.
It flags hidden Unicode (zero-width/bidi/tag characters), package-install and curl-pipe-to-shell lines in rule files, auto-approve/permission-bypass settings, MCP entries with command/args, and hook definitions, with an optional --known-packages allowlist. FAIL/WARN exits nonzero for CI gating; it never executes anything or touches the network.
The author notes limits: line-based heuristics, fixed filename list, and a clean result ≠ safe repo. It complements their earlier mcp-tool-pin-check, which pins live tool definitions after connection — one checks before you open the repo, the other after you connect.
More from coding & agent
- Fan runs One Punch Man inside Fallout 4 with Claude: 20 days, 27B tokens, ~$324K — imjustnewatai · 2026-10-28
- Dumb chunk-and-rerank baseline beats top agentic memory systems in 6 of 7 benchmarks — JeremyCMorgan · 2026-10-28
- Microsoft's MANDATE: 23 failure categories from 163k Reddit posts on AI agents — dair_ai · 2026-10-28
- LangSmith now tracks cache hit rates for your agents — LangChain · 2026-10-28
- AI amplifies the startup advantage: more hours, higher risk tolerance, less coordination — yenkel · 2026-10-28
- MOSS-Image ecosystem roundup: ComfyUI 0.43, QwenVL layout node, new style LoRAs — optimisticalish · 2026-10-28