The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill reads uncommitted code which may contain malicious instructions designed to influence the agent's analysis.
Ingestion points: SKILL.md uses git diff and reads local file content to provide context to specialists.
Boundary markers: While the specialist briefs in references/specialists.md provide role-specific context, the analyzed code snippets are not wrapped in delimiters that instruct the model to ignore embedded commands.
Capability inventory: The skill executes several git commands via shell (SKILL.md) and calls external Codex MCP tools for synthesis.
Sanitization: There is no evidence of filtering or sanitizing the content of the analyzed files before processing.

goodreview