The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests and processes untrusted data from the local file system (filenames and directory structures) to generate governance reports, which are then interpreted by the agent to recommend deletions or moves.
Ingestion points: scripts/scan.py walks the file system to gather metadata; the agent subsequently processes the resulting JSON report in its context.
Boundary markers: No explicit delimiters or instructions to ignore instructions within file metadata are provided in the documentation or script output.
Capability inventory: The agent has the ability to delete or modify memory entries and local files via its tools (subject to user confirmation).
Sanitization: Filenames and paths are passed directly into the JSON report without sanitization, creating a potential vector if a filename contains carefully crafted malicious instructions designed to trick the agent during the interpretation phase.

ldm-memory-system