The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted data from repository files (Markdown, code comments, JSDoc, and Agent Notes) to identify and fix "leakage" patterns. This creates a surface for indirect prompt injection if the files being analyzed contain malicious instructions targeting the agent.
Ingestion points: The skill reads various source code and documentation files across the repository, including .md, .ts, .py, and other text-based formats as defined in the Workflow and recall-batteries.md files.
Boundary markers: There are no explicit boundary markers or instruction-isolation techniques mentioned to separate the content of the analyzed files from the agent's operational context.
Capability inventory: The skill utilizes shell command execution (via ripgrep) and performs file modifications to fix detected issues, providing a path for injected instructions to trigger actions.
Sanitization: No specific sanitization, escaping, or filtering of the external file content is described before the agent processes it for transformation.
[COMMAND_EXECUTION]: The skill provides ripgrep (rg) command templates in references/recall-batteries.md for the agent to execute. While these are intended for read-only search operations, they involve direct interaction with the host filesystem through shell commands.

dsh-trim-cot-leakage