The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses git diff and gh repo view to fetch repository metadata and branch differences. These are standard, well-known development tools used locally for the skill's intended purpose of code auditing.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data (code changes) which constitutes an indirect prompt injection surface.
Ingestion points: Data is ingested through git diff output and reading individual modified files (SKILL.md, Phase 1).
Boundary markers: Absent; there are no specific delimiters or instructions to the agent to treat the code as untrusted data.
Capability inventory: The skill has no capabilities to write files, execute code, or perform network operations (beyond the initial repo metadata fetch). It is explicitly instructed: 'Do not make changes
just report findings'.
Sanitization: No sanitization or filtering is performed on the ingested code content.
Contextual Note: Since the primary purpose of the skill is to audit potentially buggy or malicious code, the ingestion of untrusted data is expected. The risk is mitigated by the agent's limited read-only scope for this task.

find-bugs