The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external, potentially untrusted code files as part of its core functionality, which creates a surface for indirect prompt injection.
Ingestion points: The skill reads files located at review_path, as well as docs/project.md, feature.md, and plan.md from the project repository.
Boundary markers: The instructions do not define clear delimiters or use "ignore embedded instructions" warnings when the agent processes the content of the files being reviewed.
Capability inventory: The skill has the ability to read files, execute git commands, write the review.md report, and modify project files to "fix" identified issues.
Sanitization: No explicit sanitization or filtering of the ingested code content is specified before the agent evaluates it.
[COMMAND_EXECUTION]: The skill uses local shell commands to determine the scope of the review.
Evidence: Step 0 executes git diff main...HEAD --name-only to identify changed files.
Context: This is a standard operation for a code review tool and does not involve network exfiltration or sensitive file access.

sdd-review