keyblade-pro-workflow

Pass

Audited by Gen Agent Trust Hub on Sep 15, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
  • [SAFE]: The skill's primary function is process orchestration. It focuses on enforcing a specific sequence of operations (Analysis -> Planning -> Execution -> Coding) and requires explicit user consent before and after every action. No commands related to privilege escalation, data exfiltration, or persistence were found.
  • [INDIRECT_PROMPT_INJECTION]: The workflow involves processing user-provided natural language requirements to generate technical plans and source code. This architecture presents a potential surface for indirect prompt injection where malicious instructions could be embedded in user requirements to influence downstream tasks.
  • Ingestion points: User requirement inputs, feature requests, and bug reports described in the 'brainstorming' phase (SKILL.md).
  • Boundary markers: The skill implements strict boundary markers through mandatory confirmation dialogues ("即将开始...是否确认执行?") and output summaries that require user validation before the agent continues.
  • Capability inventory: The skill orchestrates sub-skills like 'executing-plans' and 'keyblade-pro-code-rules'; however, this specific orchestrator file contains no direct network, file-system, or shell execution capabilities.
  • Sanitization: Security relies on the human user reviewing the summaries, task lists, and code changes provided at each checkpoint.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 15, 2026, 03:23 AM
Security Audit — agent-trust-hub — keyblade-pro-workflow