The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill establishes an automated workflow for processing raw data from the context/pipeline/ directory. If an attacker can place files in this staging area, they could potentially influence the agent's behavior during the ingestion process.
Ingestion points: The context/pipeline/ folder and its subdirectories (audio/, documents/) serve as the entry point for external, untrusted content.
Boundary markers: The instructions do not define specific delimiters or instructions for the agent to ignore potentially malicious content embedded within the files being processed.
Capability inventory: The skill utilizes file system writes, file deletions, git commits, and the spawning of task agents.
Sanitization: There are no requirements for sanitizing or validating the contents of the files before they are parsed and incorporated into the context store.
[DYNAMIC_EXECUTION]: The skill generates markdown-based subagent templates (pipeline-processor.md and context-organizer.md). These templates contain detailed instructions that are intended to be read and executed by the agent in subsequent tasks.
[COMMAND_EXECUTION]: The skill instructs the agent to perform several file system operations, including the creation of directories and files. Notably, it includes a cleanup instruction to "delete the source file" after processing, which is a destructive operation that could be abused or lead to unintended data loss if the agent misidentifies the target files.

chunk-claude-md