The Agent Skills Directory

[SAFE]: The skill implements a folder indexing mechanism. It uses standard tools (Read, Write, Edit, Glob) to scan files and update a JSON index.
[COMMAND_EXECUTION]: The skill calls a local script python tools/allocate-id.py to manage document IDs. This is a legitimate internal tool usage for preventing ID collisions in a multi-user environment.
[DATA_EXPOSURE]: The skill restricts its scope to Markdown and JSON files within the project, explicitly skipping sensitive directories like .git, node_modules, and .venv to prevent accidental exposure of configuration or credentials.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted file content (H1/H2 headers and first paragraphs) to extract topics. While this is an ingestion point, the output is restricted to metadata fields in a structured JSON file, posing low risk of influencing agent behavior.

index-folder