The Agent Skills Directory

[DATA_EXFILTRATION]: The helper script scripts/gerrit_api.sh accesses the ~/.netrc file to verify credentials for the Gerrit host. Although this is standard for Gerrit/Git tools, ~/.netrc is a sensitive file containing authentication tokens. The script also handles credentials through environment variables which are passed directly to curl as command-line arguments, potentially exposing them in process listings.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to process untrusted data from Gerrit servers, such as diff contents and review comments, which could contain instructions intended to influence the agent's behavior.
Ingestion points: External data is fetched in scripts/gerrit_api.sh via the get-diff, get-content, and list-comments commands.
Boundary markers: No explicit markers or instructions are provided to the agent to treat Gerrit data as untrusted content.
Capability inventory: The skill allows the agent to make authenticated network requests and perform write operations (e.g., posting reviews, submitting changes) via Gerrit APIs.
Sanitization: The script uses jq to handle JSON data, but the agent receives the raw text of code and comments.
[DYNAMIC_EXECUTION]: The helper script executes a Python snippet at runtime using python3 -c to perform URL encoding for API requests.
[COMMAND_EXECUTION]: The skill relies on executing system commands including curl, jq, and base64 to interact with the Gerrit REST API and process its responses.

gerrit-review