The Agent Skills Directory

[PROMPT_INJECTION]: The skill instructs the agent to resolve decisions by inspecting external sources like the codebase, documentation, and files before consulting the user. This creates a surface for indirect prompt injection where malicious content within these files could influence the agent's logic or recommendations during the "interview" process. * Ingestion points: The skill explicitly directs the agent to check "codebase conventions," "existing implementation," "project docs," and "files" to settle decisions (found in the "Settle it yourself first" section of SKILL.md). * Boundary markers: The skill does not define specific delimiters or warnings for the agent when processing this environmental data. * Capability inventory: The skill utilizes the AskUserQuestion tool and dispatches to a research skill, which may involve further context gathering or execution. * Sanitization: No specific sanitization or validation of the environmental context is mentioned. * Mitigation: The skill includes a mitigation strategy where all decisions settled via defaults or environment inspection are recorded in a "Decision Ledger" and presented to the user in a "Defaults applied" summary for review.

grilling