Caveat verdict
ironclaw
Fetches from network AND uses eval/Function
Agent safety classifier that calls its own API at ironclaw.io to label content; registration and API key flows are transparent, and requesting periodic version checks via its heartbeat mechanism is standard update behavior with no evidence of exfiltration or manipulation.
Automated static analysis โ not a human review. Caveat flags capabilities, not confirmed intent, and can produce false positives. Disagree with this verdict? Use Dispute below.
What it does
These are capability combinations: each listed behavior occurs in the skill, but Caveat detects co-occurrence โ it does not verify that one flows into another. Read the code to confirm a live chain.
Fetches from network AND uses eval/Function โ the remote-code-execution pattern (data-flow not verified)
LLM05 ยท LLM06 ยท ASI05
Accesses credentials AND makes external network calls โ potential credential theft
LLM02 ยท ASI03
Permission integrity
network_out
credential_access
Findings (3)
Pipe to bash โ executes piped content as shell commands
SKILL.md ยท code ยท |bash
Sets world-executable permissions
SKILL.md ยท code ยท chmod 777
Popular HTTP library โ network access
SKILL.md ยท prose ยท downgraded ยท Got
Why the tier is capped
Execution sink present in raw bytes (Hard Floor: class A). Final tier capped at Caution โ cannot be lifted by any downgrade, example-payload opt-in, or allowlist.
Permissions & capabilities
No declared permissions โ minimal attack surface.
network_outdynamic_evalcredential_access Thanks โ recorded.