spec-driven-implementation

Pass

Audited by Gen Agent Trust Hub on Nov 19, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill manages untrusted data ingestion from external sources such as GitHub issue titles, descriptions, and comments. These inputs influence the creation of product and technical specifications.
  • [INDIRECT_PROMPT_INJECTION]: The skill establishes security boundaries by instructing the agent to ignore prompt injections, secret extraction attempts, or role-change requests embedded in untrusted data.
  • [INDIRECT_PROMPT_INJECTION]: While the skill orchestrates implementation tasks that write code to the repository, it mitigates risk through explicit warnings that external data cannot override output paths, validation requirements, or safety protocols.
  • [INDIRECT_PROMPT_INJECTION]: Mandatory Evidence Chain: (1) Ingestion points: GitHub issue metadata and content; (2) Boundary markers: Explicit instructions to treat data as untrusted; (3) Capability inventory: Orchestrates file writes and code implementation via related skills; (4) Sanitization: Direct instructions to disregard injections and safety bypass attempts.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 19, 2026, 01:13 AM