The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes the suspec command-line utility to validate review packets and check them against specification files.
[COMMAND_EXECUTION]: The instructions mandate that the agent re-run verification checks by resolving and executing shell commands from the Commands table found in an external AGENTS.md file. This introduces a risk where a malicious repository could include dangerous commands intended to be executed during the automated review process.
[PROMPT_INJECTION]: The skill is vulnerable to Indirect Prompt Injection (Category 8) because its primary function involves ingesting and acting upon untrusted project artifacts. • Ingestion points: External files including AGENTS.md, code diffs, and specification documents. • Boundary markers: The skill recommends a 'refute-by-default' stance and spawning fresh subagents for independence, which acts as a process-level boundary, though it lacks technical data delimiters. • Capability inventory: Shell execution via suspec and arbitrary command execution sourced from AGENTS.md. • Sanitization: The instructions do not specify sanitization or validation of the commands retrieved from the project's command table before execution.

review-output