The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from git logs, diffs, and commit messages which could contain instructions intended to influence the agent's analysis.
Ingestion points: SKILL.md (Phase 2: Gather Changes) instructs the agent to read output from git log and git diff commands.
Boundary markers: The skill does not provide specific delimiters or instructions to the agent to treat diff content exclusively as data or to ignore embedded instructions.
Capability inventory: The skill is limited to reading git information and generating text responses; it does not perform network operations, file writes, or privilege escalation.
Sanitization: The skill does not implement sanitization or filtering for the ingested code changes or commit messages.
[COMMAND_EXECUTION]: The skill instructs the agent to execute shell commands to interact with the local git repository.
Evidence: SKILL.md (Phase 1 and Phase 2) lists various standard git commands such as git log, git diff, and git show to be executed to gather context for the analysis.

lesson-learned