The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted source code via git diff results and passes the content to various specialized reviewer agents, creating a potential vector where instructions within the analyzed code could attempt to manipulate the review output. -- Ingestion points: Files detected by git diff in SKILL.md (Step 1) and processed in Step 3. -- Boundary markers: The prompt provided to sub-agents lacks strong delimiters or specific 'ignore embedded instructions' directives to isolate the untrusted code content. -- Capability inventory: The skill utilizes the Agent tool to perform multi-role analysis and generate unified reports based on filesystem content. -- Sanitization: No validation, escaping, or filtering of the code content is performed before analysis.

effect-review