The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze untrusted data from git history and source files, which could contain malicious instructions.
Ingestion points: git log output, git diff output, and source code files accessed during the audit tiers.
Boundary markers: The skill provides structural templates for output findings but lacks specific instructions to disregard embedded commands in the data being audited.
Capability inventory: The agent can execute git commands and a Python script, and write output files.
Sanitization: No sanitization is performed on the ingested data before analysis.
[DYNAMIC_EXECUTION]: The skill executes a Python script (citation_verifier.py) from a local path within the project. This execution relies on the integrity of the script file at that location.
[COMMAND_EXECUTION]: The skill runs shell commands including git log and git diff. The use of {base} as a variable in these commands presents a minor risk of injection if the variable source is untrusted.

tiered-audit