The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted documents provided by the user (PDF, Word, LaTeX exports), which presents a potential surface for indirect prompt injection attacks.
Ingestion points: Reads source document text directly from files or user-pasted content as described in SKILL.md.
Boundary markers: Includes an explicit instruction in SKILL.md to ignore any directives embedded in the source documents.
Capability inventory: Executes internal Python scripts via subprocesses, writes proposal and notes files to the workspace, and performs network requests to api.crossref.org.
Sanitization: Implements automated stripping of personal data (names, IDs, emails) and forbidden content types during the conversion process.
[COMMAND_EXECUTION]: The skill invokes included Python utility scripts (scripts/check.py and scripts/validate_refs.py) to verify the structural integrity and bibliographic accuracy of the imported proposal. It also mentions the use of standard system tools like pdfimages or mutool for image extraction if available.
[EXTERNAL_DOWNLOADS]: Fetches metadata from Crossref's official API (api.crossref.org) to verify and complete bibliographic entries. This communication targets a well-known scholarly service and is essential for the skill's reference enrichment functionality.

proposal-import