The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to execute a shell command (python3 scripts/deferred_capture.py) using variables derived from the findings of a code review. Critically, it directs the agent to run this command automatically without prompting the user for confirmation, bypassing standard security oversight.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data (the files and documentation in the repository under review) to generate findings. These findings are subsequently used as parameters for shell commands. There is a lack of clear instruction or mechanism to sanitize or escape these parameters, allowing for potential command injection if a reviewed file contains shell metacharacters.
Ingestion points: Project files, manifests, and git status data analyzed in Step 1.
Boundary markers: Absent in the command-line construction phase.
Capability inventory: Shell execution of the deferred_capture.py script.
Sanitization: Not implemented; data is directly interpolated into the command string.
[DYNAMIC_EXECUTION]: The skill constructs and executes shell commands at runtime based on the textual content of its findings, which is a dynamic code generation pattern.
[REMOTE_CODE_EXECUTION]: The skill references and attempts to execute external logic contained in scripts/deferred_capture.py and workflows/unified-review.js, neither of which is provided in the skill package, making their behavior and safety unverifiable.

unified-review