The Agent Skills Directory

[SAFE]: The skill operates using established platform tools for task orchestration and repository access. It relies on internal logic to manage subagents and synthesizes results locally without external network dependencies.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted input in the form of repository files and git diffs (Ingestion points: Step 1 in SKILL.md). This data is interpolated into prompts for four subagents (Step 3). While boundary markers for this data are not explicitly defined in the SKILL.md (Capability inventory: Task tool spawning subagents), the risk is significantly mitigated by the readonly: true configuration for all subagents, preventing them from performing unauthorized system modifications based on malicious code comments. No sanitization logic for the ingested code is described.

interrogate