Skill flagged — suspicious patterns detected
ClawHub Security flagged this skill as suspicious. Review the scan results before using.
Trade Audit
v2.0.0Mandatory audit gate for all trading and transfer decisions. Sends agent-prepared decision material to Apus deterministic inference on an NVIDIA H100 TEE and...
⭐ 0· 80·0 current·0 all-time
byAlex@alex-wuhu
MIT-0
Download zip
LicenseMIT-0 · Free to use, modify, and redistribute. No attribution required.
Security Scan
OpenClaw
Suspicious
medium confidencePurpose & Capability
Name/description say it will accept agent-prepared decision material, send it to Apus deterministic inference, and return an attested verdict. The included analyze.py implements those behaviors (builds a normalized bundle, posts to an APUS endpoint, parses a JSON packet, and writes a local audit log). No unrelated credentials, binaries, or install steps are requested.
Instruction Scope
SKILL.md confines work to agent-prepared inputs and instructs the agent to fetch public data and distill it before calling analyze.py. The script itself posts the bundle to a remote Apus endpoint and appends a record to ~/.trade-audit/audit.jsonl. Two issues: (1) analyze.py contains at least one clear coding bug (in normalize_packet it sets norm['missing_information'] = normal, where 'normal' is undefined) which will likely cause crashes or exceptions at runtime; (2) the skill will transmit whatever the agent includes in the prepared bundle (addresses, amounts, possibly other sensitive details) to an external service by default — the SKILL.md warns to strip extraneous material but there is no technical safeguard to prevent leaking sensitive fields.
Install Mechanism
No install spec; the skill is instruction-plus-a-single-python script using only stdlib. Nothing is downloaded or written beyond the script and template files included in the bundle.
Credentials
The skill declares no required environment variables or credentials and the code uses hard-coded APUS_BASE_URL and MODEL_NAME. There is no request for unrelated credentials. However, because it posts bundle contents to an external endpoint, users must ensure they don't include secrets in prepared bundles.
Persistence & Privilege
The skill is not always-enabled and does not request elevated privileges. It does create and append an audit log at ~/.trade-audit/audit.jsonl on each run; that persistent local storage could accumulate sensitive decision material and should be considered when deploying (encryption, rotation, or opt-out may be desirable).
What to consider before installing
This skill is broadly consistent with its description, but take these precautions before installing or using it:
- Review and fix the code bug: analyze.py contains an undefined name ('normal') in normalize_packet which will likely raise an exception; ask the author for a corrected release or patch before relying on it.
- Audit data sent to Apus: the script posts whatever is in the prepared bundle to https://hb.apus.network; ensure you never include private keys, wallet seeds, or any confidential PII in prepared bundles. The SKILL.md recommends stripping extraneous text, but that is a manual step — consider adding explicit sanitization or local vetting.
- Verify the Apus endpoint and attestation claims: confirm the endpoint, attestation format, and expected guarantees (hardware TEE attestation, integrity proofs) independently. Hard-coded endpoints are harder to rotate; you may prefer an environment-variable override so you can point to a test or internal endpoint.
- Be aware of local logs: runs append records to ~/.trade-audit/audit.jsonl which may contain sensitive decision material; decide whether to encrypt, rotate, or disable logging.
- Test in an isolated environment: run the script with non-sensitive sample bundles to see behavior and confirm output formatting and exit codes (gate mode) before integrating into any automated trading workflow.
If you want, I can: (1) point out exact lines to patch for the undefined-variable bug; (2) produce a hardened variant that prompts for explicit approval before sending bundles externally and optionally redacts sensitive fields; or (3) draft a short checklist for safe operational use (logging policy, bundle sanitization, endpoint verification).Like a lobster shell, security has layers — review code before you run it.
latestvk970v2tcqthce1x5qvy4ypttnx841767
License
MIT-0
Free to use, modify, and redistribute. No attribution required.
