wfgy-method
Pass
Audited by Gen Agent Trust Hub on Aug 28, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill implements a metacognitive reasoning process that ingests untrusted user-provided goals (G) and task context (I) to calculate reasoning drift (deltaS). Because the skill lacks explicit boundary markers or sanitization for these inputs, it is vulnerable to indirect prompt injection where malicious instructions embedded in a goal could influence the agent's decision-making or checkpoint-and-retry logic.
- Ingestion points:
SKILL.md(Goal G and current task state I processed during reasoning checks). - Boundary markers: Absent; there are no specified delimiters (e.g., XML tags or blockquotes) to isolate untrusted goal/state inputs from the reasoning instructions.
- Capability inventory:
SKILL.md(Instructions to write durable lessons to<project-root>/.wfgy/lessons.md). - Sanitization: Absent; the skill does not instruct the agent to escape or filter content before evaluating it for drift or recording it in the lessons log.
Audit Metadata