The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze untrusted data in the form of local code diffs during its input gathering phase.
Ingestion points: The agent gathers the full branch diff using git diff and reads changed files individually in Phase 1.
Boundary markers: The instructions do not define clear delimiters or include explicit directives for the agent to ignore instructions that may be embedded within the source code being audited.
Capability inventory: The skill utilizes git and gh (GitHub CLI) to read repository data and file contents. While the skill's defined commands are read-only, the agent's broader environment may include tools with write or network capabilities.
Sanitization: There is no evidence of sanitization or escaping of the content retrieved from the diffs before it is processed by the agent.
[COMMAND_EXECUTION]: The skill uses local shell commands to facilitate its auditing functionality.
Evidence: Phase 1 executes git diff combined with gh repo view --json defaultBranchRef --jq '.defaultBranchRef.name' to identify changes relative to the default branch. These are standard operations for the skill's stated purpose of reviewing code changes.

find-bugs