ClawAudit verdict

autoresearch

autoresearch-karpathy

88
🟢 Trusted
Low risk — reviewed by ClawAudit, behavior matches stated purpose

An autonomous ML research skill that modifies training code and runs experiments locally on a GPU; all operations are documented and constrained to local neural network training with no external exfiltration.

⚠ Flagged for review — coarse, uncorroborated signal, not a confirmed exploit. Review the config yourself before installing.

Automated static analysis — not a human review. ClawAudit flags capabilities, not confirmed intent, and can produce false positives. Disagree with this verdict? Use Dispute below.

0
security
70
transparency
70
maintenance

Findings (4)

Pattern match critical

Pipe-to-shell pattern (curl | sh) — supply chain attack vector

README.md · code · curl -LsSf https://astral.sh/uv/install.sh | sh

Pattern match critical

Pipe to sh — executes piped content as shell commands

README.md · code · | sh

Pattern match high

Possible hardcoded credential

prepare.py · prose · downgraded · TOKEN = "<|reserved_0|>

Pattern match high

Uses eval() — can execute arbitrary code

train.py · prose · downgraded · eval(

Why the tier is capped

Execution sink present in raw bytes (Hard Floor: class A). Final tier capped at Caution — cannot be lifted by any downgrade, example-payload opt-in, or allowlist.

Permissions & capabilities

No declared permissions — minimal attack surface.

Is this flag fair?

Check another skill Browse the registry Auditing your own skills or configs? Use the API