The Agent Skills Directory

[SAFE]: The skill implements a verification-first methodology that enhances the reliability and security of multi-agent workflows.
[SAFE]: All identified tool usage (Bash, Grep, Rscript, etc.) is confined to local file system operations for the purpose of validating artifact existence, shape, and consistency.
[PROMPT_INJECTION]: The skill manages potential indirect prompt injection surfaces through its core verification logic. 1. Ingestion points: The skill reads deliverables such as CSV and HTML files to perform validation checks. 2. Boundary markers: It defines strict expected outcome specifications before execution, minimizing reliance on potentially malicious embedded instructions. 3. Capability inventory: Tools like wc, sha256sum, and Rscript are used for read-only validation and range checking. 4. Sanitization: The process focuses on objective metrics (checksums, row counts, and schema validation) which are inherently resistant to injection attacks.

verify-agent-output