The Agent Skills Directory

[PROMPT_INJECTION]: The skill operates on untrusted data from git diffs and pull requests. This represents an indirect prompt injection surface where malicious instructions could be embedded within the code changes to attempt to influence the agent's analysis.
Ingestion points: DIFF_SOURCE input and the content of the diff or patch file analyzed in SKILL.md.
Boundary markers: No specific delimiters are used to isolate untrusted code content from the agent's instructions.
Capability inventory: The workflow includes executing shell-based tools such as git, semgrep, npm audit, pip-audit, and govulncheck (SKILL.md).
Sanitization: No explicit sanitization or validation of the input content before processing is described.
[COMMAND_EXECUTION]: The skill relies on shell commands for its core functionality (e.g., git diff, semgrep). The use of user-supplied inputs such as branch names or commit references in these commands is a standard pattern for developer tools, but could potentially lead to command injection if the inputs (like <BASE_REF> or <HEAD_REF>) are maliciously crafted.

differential-review