The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data by scanning the developer's codebase using Grep and Read operations to identify integration details and security risks. While this scanning process could theoretically expose the agent to malicious instructions embedded in comments or code within the analyzed project, the skill uses this information exclusively for generating a structured audit report.
Evidence Chain:
1. Ingestion points: The local codebase files searched during the discovery and evaluation steps (Steps 1, 2, and 3).
1. Boundary markers: None explicitly specified for the search results.
1. Capability inventory: Calls to official Mercado Pago quality and homologation tools, and interaction with the user via AskUserQuestion.
1. Sanitization: None specified for the read content.
Conclusion: The behavior is assessed as safe as it is fundamental to the skill's purpose as a security auditor.

mp-review