The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/letta_fs_to_memfs.py executes the qmd CLI utility using subprocess.run to manage semantic search indices and queries.
[INDIRECT_PROMPT_INJECTION]: The skill ingests content from untrusted external URLs and local documents, processing them into memory chunks accessible by the agent.
Ingestion points: File paths and URLs provided via the --source argument in the ingest command of scripts/letta_fs_to_memfs.py.
Boundary markers: Chunks are stored in markdown files prefixed with YAML frontmatter metadata.
Capability inventory: The skill writes to the filesystem and executes external search and indexing commands.
Sanitization: Input filenames are normalized using a slugification function, and frontmatter values are escaped via JSON encoding.
[EXTERNAL_DOWNLOADS]: The download_source function in scripts/letta_fs_to_memfs.py uses urllib.request to fetch content from remote URLs. The documentation includes examples targeting well-known academic repositories like arXiv.
[REMOTE_CODE_EXECUTION]: The skill suggests the installation of a third-party Node.js package (@tobilu/qmd) from an unverified source to enable semantic search functionality. It also uses subprocess.run to execute this tool, representing a dependency on external code that runs with the agent's privileges.

letta-filesystem-to-memfs