The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from various sources to define the code review specifications.
Ingestion points: The skill reads pull request descriptions, Git commit messages, and specification files located in directories such as docs/, specs/, and .scratch/.
Boundary markers: There are no explicit instructions or delimiters to isolate these external inputs or warn the agent to ignore embedded instructions.
Capability inventory: The skill executes Git commands, reads local files, and delegates tasks to sub-agents for parallel processing.
Sanitization: The instructions do not specify any validation, escaping, or filtering of the content retrieved from external metadata or documentation files.
[COMMAND_EXECUTION]: The skill relies on shell commands to gather the necessary context for the review process.
Evidence: The workflow explicitly calls for executing git diff, git status, and git log commands to identify changes and project history.
[EXTERNAL_DOWNLOADS]: The skill attempts to fetch pull request and issue data using external tracker integrations (e.g., GitHub or GitLab via issue-tracker.md logic).
[METADATA_POISONING]: The skill references a setup command /setup-ouyangjiahong-skills in the boundary cases table, which is an external vendor-specific tool used for configuration.

code-review