The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it processes external data (reviewer comments) and uses it to perform high-privilege actions like code modification.
Ingestion points: Reviewer comment bodies are fetched from GitLab's API via the scripts/fetch_mr.py script and provided to the agent.
Boundary markers: The instructions do not define clear boundaries or provide warnings to ignore instructional content found within the comments themselves.
Capability inventory: The agent has access to local code via LSP tools and Grep, and is explicitly instructed to apply code fixes to the working tree.
Sanitization: There is no evidence of sanitization or validation of the comment content before it is interpreted by the agent.
[COMMAND_EXECUTION]: The skill relies on several local CLI tools to perform its tasks, which is expected for its development-focused purpose.
Evidence: The helper script fetch_mr.py executes glab api calls using subprocess.run to fetch MR metadata.
Evidence: SKILL.md directs the agent to execute standard git commands, including git rev-parse, git diff, and git switch to manage the local environment.
[EXTERNAL_DOWNLOADS]: The skill fetches data from external GitLab instances.
Evidence: It uses the glab CLI tool to download Merge Request discussions and metadata from the host provided in the user's MR URL.

audit-reply