The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted data in the form of candidate dossiers, accomplishment lists, and existing letter drafts, which could contain malicious instructions or misleading content.\n- Ingestion points: Dossier data, candidate accomplishments, and existing letter drafts provided by the user during the intake process or through file read operations.\n- Boundary markers: The skill utilizes explicit bracketed placeholders (e.g., [recommender: ...]) to separate AI-generated scaffolding from required user-provided evidence, helping to maintain context boundaries and prompt the recommender for missing facts.\n- Capability inventory: The skill uses Read, Write, and Edit tools to manage files, and executes local Python scripts (claim_guard.py, letter_scaffold.py) via the Bash tool to perform linting and scaffolding operations.\n- Sanitization: The skill includes a 'no-fabrication guard' (scripts/claim_guard.py) that specifically lints drafts for unsupported superlatives or unanchored rankings. This acts as a sanitization layer by flagging and requiring human verification for claims that are not anchored to the provided dossier evidence, reducing the risk of the model following instructions to fabricate achievements.

alterlab-recommendation-letters