The Agent Skills Directory

[SAFE]: The skill operates entirely on local files within the project directory. The main script, ingest-library.py, reads source documents from library/inbox/ and writes structured output to library/ and brand/boilerplate/. These operations are strictly limited to the intended scope of organizing a proposal document library.
[SAFE]: The skill does not perform any network requests, external data transfers, or remote code downloads. All data processing, classification, and indexing are handled locally by the provided Python scripts.
[SAFE]: No hardcoded credentials, API keys, or access to sensitive system paths (e.g., .ssh, .aws) were found. The script manages its own manifest and index files using project-relative paths.
[SAFE]: The inclusion of a vendor attribution link to the official pawbytes.io store in SKILL.md is a documented practice for the author and does not involve any suspicious tracking or data exfiltration mechanisms.
[PROMPT_INJECTION]: The skill processes documents from the library/inbox/ directory, which constitutes an indirect prompt injection surface. This is assessed as safe due to the skill's purpose as a local tool for user-managed content.
Ingestion points: Markdown, text, and JSON files in library/inbox/ are read by ingest-library.py and subsequently interpreted by the agent.
Boundary markers: None explicitly defined in the script for isolating ingested document content from agent instructions during refinement.
Capability inventory: File read/write access is restricted to the project's library and boilerplate directories.
Sanitization: Data is extracted using regular expressions, but no sanitization or escaping of external content is performed before presentation to the agent.

paw-pa-library