Back to skill
Skillv1.0.0

ClawScan security

black-box · ClawHub's context-aware review of the artifact, metadata, and declared behavior.

Scanner verdict

SuspiciousFeb 21, 2026, 6:55 PM
Verdict
suspicious
Confidence
medium
Model
gpt-5-mini
Summary
The skill mostly does what it says (remote audit logging to TiDB) but has implementation inconsistencies and behaviors (auto-provisioning, persistent DSN file, potential exfiltration of sensitive agent actions) that deserve caution before installing.
Guidance
This skill is coherent with remote audit logging, but you should not install it without consideration: it will transmit agent actions to an external TiDB service and persist a DSN file (~/.openclaw_black_box_dsn) that may contain credentials. Verify you trust the TiDB provider (zero.tidbapi.com), prefer supplying your own DB credentials instead of relying on auto-provisioning, and audit run.py to confirm what exactly is logged (avoid logging raw commands or secrets). If you proceed, restrict network access and rotate any credentials used for the logger; consider running it in an environment where agent logs won't leak sensitive data.

Review Dimensions

Purpose & Capability
noteThe declared requirements (TiDB env vars, python3, curl) and included code align with a remote audit-logging purpose. Small mismatches exist: SKILL.md mentions a table named `agent_logs` while the code and DESIGN.md use `flight_recorder`/`flight_recorder` schema. Otherwise the required pieces are proportionate to the stated goal.
Instruction Scope
concernThe SKILL.md and PROTOCOL instruct the agent to log 'before executing High-Risk commands' which means potentially sensitive commands/intents may be transmitted to a remote DB. The runtime code writes/reads a DSN from ~/.openclaw_black_box_dsn and will auto-provision an instance by POSTing to https://zero.tidbapi.com — both behaviors are within the stated purpose but increase exposure of agent activity and credentials. Also the SKILL.md/schema name mismatch is an incoherence to note.
Install Mechanism
okNo install spec; code is instruction-only with a bundled run.py and a simple external dependency (pymysql) listed. The script uses curl to contact a third-party API (tidbapi.com) but does not download or execute arbitrary archives. Risk is limited to runtime network interactions rather than arbitrary install-time code fetches.
Credentials
noteRequested env vars are TiDB connection parameters, which are appropriate for a remote-DB logger. However, the skill caches a connection string to ~/.openclaw_black_box_dsn in plaintext (including credentials from auto-provisioning), which is a persistence-of-secrets risk. Also logs may contain sensitive agent context or secrets if the agent logs full command text.
Persistence & Privilege
noteThe skill does not request always:true and does not modify other skills. It persists a DSN file in the user's home and can auto-provision an external DB — these enable long-lived access to logged data and credentials, which increases blast radius if the external service or DSN file is compromised.