The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill has a surface for indirect prompt injection as it processes external, potentially untrusted code diffs and source files. \n- Ingestion points: The instructions in SKILL.md direct the agent to 'Read the full diff and nearby code needed to establish conventions.' \n- Boundary markers: The skill includes protective instructions such as 'Host permissions and explicit safety constraints still bind the work.' \n- Capability inventory: The skill is restricted to 'read-only' behavior and 'narrow non-destructive checks' (e.g., reading files), with no network or write capabilities defined. \n- Sanitization: There are no explicit sanitization or filtering steps for the ingested code content. \n- [SAFE]: No malicious command execution, data exfiltration, or obfuscation techniques were detected in the skill instructions.

code-quality-review