The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted external data (git diffs and arbitrary source code) and pass it to multiple subagent reviewers. This creates a surface for indirect prompt injection attacks.
Ingestion points: The skill retrieves content using git diff and file-reading operations in SKILL.md (Step 1).
Boundary markers: The template in references/reviewer-prompt.md places the {DIFF_OR_FILES} content directly into the subagent's prompt without high-entropy delimiters or explicit isolation instructions to prevent the subagent from following instructions embedded within the code itself.
Capability inventory: The skill possesses the ability to execute shell commands (git) and spawn concurrent tasks/subagents.
Sanitization: No sanitization, filtering, or escaping is performed on the code content before it is interpolated into the prompts for the reviewer models.
[COMMAND_EXECUTION]: The skill utilizes shell commands (git diff) to gather context for the review. While appropriate for a development tool, this represents a capability that could be misused if branch names or base references were influenced by untrusted user input.

interrogate