The Agent Skills Directory

[DATA_EXFILTRATION]: The skill transmits user-supplied documents, which may contain sensitive PII like IDs, resumes, and contracts, to the vendor's service at gosnapfill.com. This behavior is the primary function of the skill and is consistent with the vendor identity (Ontos-AI).
[CREDENTIALS_UNSAFE]: The instructions guide users to configure their API key using the platform's standard configuration command (openclaw config set). No secrets are hardcoded within the skill itself.
[PROMPT_INJECTION]: The skill contains a vulnerability surface for indirect prompt injection when processing untrusted external files (PDF, Word, or Excel). Ingestion points: Documents and images are processed via snapfill_ingest_instant_knowledge and snapfill_submit_job. Boundary markers: No explicit delimiters are defined for the ingested content, though the review step acts as a functional boundary. Capability inventory: The skill submits data to a remote API and generates downloadable files. Sanitization: Risk is mitigated by the 'Field Confirmation Rule,' a mandatory human-in-the-loop verification step that requires user approval of all resolved values before the document is finalized.
[PROMPT_INJECTION]: Authoritative instructions (e.g., 'Critical rule', 'Hard block') are used to constrain agent behavior, mandating the use of vendor-specific tools over internal extraction methods and preventing manual workarounds.

snapfill