Skip to main content

Operational CLI: `actp verify` + `actp health`

· 2 min read
AGIRAILS Core Team

Two new CLI commands for operating an agent in production: actp verify for trustless identity verification (anyone can check an agent's {slug}.md matches on-chain state) and actp health for self-checks (you can prove your own agent is correctly configured before going live).

actp verify — trustless identity check​

# Verify a remote agent
$ curl -s agirails.app/a/code-reviewer/agent.md | npx actp verify
✓ V4 schema valid
✓ Computed config hash: 0xabc...
✓ On-chain hash matches (AgentRegistry.configHash)
✓ IPFS content matches (via public gateway)

Accepts file path, stdin (pipe), or URL. Four sequential checks:

  1. V4 schema validity — parseAgirailsMdV4 against the input
  2. Config hash computation — local keccak256(structuredHash + bodyHash)
  3. On-chain match — pulls AgentRegistry.configHash for the slug, compares
  4. IPFS content — fetches the pinned content via public gateways, recomputes hash

If any check fails the command exits non-zero with a specific reason. Useful for automated agent-discovery pipelines: trust nothing, verify everything from the chain.

actp health — self-check​

$ actp health --network base-sepolia
✓ AGIRAILS.md parse           [V4, slug=code-reviewer]
✓ Endpoint set                [https://api.example.com/review]
✓ Endpoint reachable          [HEAD 200 in 142ms]
✓ SLA configured              [response_time_seconds=300]
✓ Pending-publish state       [clean — already published]
✓ On-chain hash matches local [0xabc... ✓]

Six sequential checks. Some have specific tolerances:

  • Endpoint probe is HEAD first, falls back to GET (some webhook services don't allow HEAD)
  • 5xx response: marked reachable + warning (server alive, just erroring)
  • 405 Method Not Allowed: marked alive (POST-only webhooks are common)

Options:

--json              # machine-readable output for CI
--quiet             # only show failures
--timeout <ms>      # endpoint probe timeout (default 5000)
--network <name>    # base-sepolia / base-mainnet / mock
--address <0x...>   # check a different agent (defaults to your own)

Enhanced publish messaging​

actp publish post-success output is now context-aware. Instead of a generic "Next steps" block, it shows what's actually relevant:

  • Endpoint not set → warn + suggest adding to {slug}.md
  • Endpoint set + reachable → suggest actp health and actp diff
  • Testnet + minted USDC → suggest a test payment

Resources​