The Agent Skills Directory

[SAFE]: The skill is a documentation file (SKILL.md) that outlines a methodology for AI agents to retrieve codebase context. It uses pseudocode and examples to explain the logic for progressively refining searches.
[INDIRECT_PROMPT_INJECTION]: The design pattern involves reading and analyzing external data (local codebase files) in a loop via the retrieveFiles function. While this creates a potential surface for indirect prompt injection if a file contains malicious instructions, the skill defines a scoring and evaluation phase (evaluateRelevance) to filter content, which is a standard and necessary component of retrieval-augmented generation (RAG) tasks.
Ingestion points: The skill retrieves files from the local filesystem in the DISPATCH phase.
Boundary markers: No explicit boundary markers or XML-style tags are provided in the prompt examples.
Capability inventory: The pattern includes logic to read and score content using evaluateRelevance and identifyGaps functions.
Sanitization: No specific sanitization logic is described beyond relevance scoring.

iterative-retrieval