Agent Supply Chain Guard
Your AI agent can be compromised by a file it reads.
An innocent-looking SKILL.md, MCP manifest, plugin or README can carry instructions that redirect an agent, expose credentials, run commands, or ask for far more access than it needs.
- Flags hidden instruction language, shell and process execution, credentials and private keys, dangerous URL schemes, wildcard permissions and invisible Unicode controls.
- Local-first and dependency-free. Clean scans exit 0, findings exit 1, and a GitHub Action is included.
- Optional second-pass review with your own OpenAI or Anthropic key; only matched evidence is sent.
Know the limit
A signal scanner, not proof of exploitability. Findings need human review and a clean result does not make a dependency safe.
$ git clone https://github.com/ppradyoth/agent-supply-chain-guard.git
$ cd agent-supply-chain-guard
$ python -m agent_supply_chain_guard scan examples/poisoned-skill.md
examples/poisoned-skill.md:5: [hidden-instruction] instruction-like text found
examples/poisoned-skill.md:6: [process-execution] process execution reference
examples/poisoned-skill.md:7: [unrestricted-permission] broad or unrestricted permission
3 finding(s)