The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill defines a process where an agent programmatically retrieves and evaluates the contents of files from a codebase to refine its context. This pattern establishes a surface for indirect prompt injection, where an attacker could place malicious instructions inside the files being searched.
Ingestion points: The 'エージェントとの統合' (Integration with Agent) section and the iterativeRetrieve pseudocode instruct the agent to ingest and analyze the contents of files found in the project directory.
Boundary markers: The instructions do not include requirements for the agent to wrap retrieved content in delimiters or to ignore embedded instructions within the ingested data.
Capability inventory: The pattern relies on the agent's ability to perform filesystem searches and read file content via standard tools.
Sanitization: There are no provisions for sanitizing, escaping, or validating the content of the files before the agent evaluates their relevance.

iterative-retrieval