Scopecheck

Analyze an OpenClaw SKILL.md and extract its permission scope — what env vars, CLI tools, filesystem paths, and network URLs it accesses. Compares declared requirements against actual usage and flags undeclared access.

Install

openclaw skills install @mirni/scopecheck

ScopeCheck

Find out what a skill actually accesses vs what it claims to need.

Start the server

bash
uvicorn scopecheck.app:app --port 8002

Check a skill's scope

bash
curl -s -X POST http://localhost:8002/v1/check-scope \
  -H "Content-Type: application/json" \
  -d "{\"skill_content\": $(cat path/to/SKILL.md | jq -Rs)}" | jq

Returns declared (env vars and bins from metadata), detected (what the skill actually references), and undeclared_access (detected but not declared — potential risk).

What it extracts

  • env_vars — environment variables like $HOME, $API_KEY
  • cli_tools — binaries used in Run/Execute commands
  • filesystem_paths — /etc/hosts, ~/.ssh/config, etc.
  • network_urls — any http/https URLs referenced

Undeclared access format

Each undeclared item is prefixed with its type: env:SECRET_KEY, bin:curl, fs:/etc/passwd, net:https://example.com.