The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill instructs the agent to automatically search for and read credentials from sensitive file paths including ~/.env, ~/.openclaw/secrets.env, and workspace .env files.
[CREDENTIALS_UNSAFE]: The agent is specifically directed to read the first two lines of a .env file using sed and treat them as raw API keys and secrets if the standard KEY=VALUE format is not detected, posing a risk of harvesting unrelated sensitive data from those files.
[COMMAND_EXECUTION]: The skill provides complex shell script logic for the agent to execute, including the use of grep, sed, openssl, and curl to manipulate secrets and perform authenticated requests.
[COMMAND_EXECUTION]: The skill implements automated signing procedures using openssl for HMAC, RSA, and Ed25519, which involves the agent accessing and processing private keys (private_key.pem).
[DATA_EXFILTRATION]: The skill reads local credentials and transmits them to https://api.binance.com. While this targets the vendor's official API, the automated retrieval and transmission occur without explicit user confirmation for certain configuration methods.
[PROMPT_INJECTION]: The skill contains instructions for the agent to 're-read' secrets files upon user request and to bypass confirmation steps for transactions when specifically configured, which reduces human-in-the-loop oversight.

fiat