The Agent Skills Directory

[SAFE]: The skill is entirely composed of markdown documentation and templates. It does not contain any executable scripts, binaries, or instructions for shell execution, network exfiltration, or persistence.
[PROMPT_INJECTION]: The skill identifies an attack surface for indirect prompt injection as it is designed to ingest external documents like buyer solicitations and Terms of Reference (ToRs). Ingestion points: Untrusted data enters the agent context during the 'Required Inputs' phase defined in SKILL.md. Boundary markers: The skill includes a 'Capability Contract' that explicitly prohibits high-risk actions such as bid submission, publishing content, or changing source records. Capability inventory: Agent actions are restricted to reading workspace files and drafting content; no subprocess or network capabilities are utilized. Sanitization: No explicit input validation or escaping is defined for the external text. The risk is minimized by the skill's restricted operational scope and clear behavioral constraints.

profiles