The Agent Skills Directory

[COMMAND_EXECUTION] (LOW): The skill utilizes curl commands to fetch metadata and logs from the Read the Docs API. This is a standard diagnostic function but involves executing shell commands.
[DATA_EXFILTRATION] (LOW): The skill includes instructions to pass an API token (RTD_TOKEN) in a header to a variable host (RTD_HOST). While the documentation points to legitimate Read the Docs domains, a malicious user could potentially supply a different host URL to capture the token.
[PROMPT_INJECTION] (LOW): Indirect Prompt Injection Surface Detected. The skill is specifically designed to read and interpret raw build logs which could contain attacker-controlled content.
Ingestion points: Raw build logs are fetched via the RTD text endpoint defined in SKILL.md.
Boundary markers: None are specified; the logs are processed as raw text.
Capability inventory: Includes network reads via curl. The agent is also instructed to "apply fixes," which might lead to file modifications if the agent has those permissions.
Sanitization: The skill does not implement sanitization for the log content, relying on the agent to interpret potentially malicious strings inside the build output.

readthedocs-build-failure-triage