architect

Warn

Audited by Gen Agent Trust Hub on Sep 20, 2026

Risk Level: MEDIUMMETADATA_POISONINGINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [METADATA_POISONING]: The author information in the SKILL.md frontmatter (https://github.com/dmitry) contradicts the provided author context of jeffallan, which may be misleading regarding the skill's origin.- [INDIRECT_PROMPT_INJECTION]: The skill triages and structures content from previous 'Madman' and 'Whirlybird' stages, which creates a surface where the agent could potentially follow instructions hidden in that ingested material.
  • Ingestion points: The skill receives material and clusters in the 'Core Workflow' section of SKILL.md.
  • Boundary markers: There are no instructions or delimiters provided to ensure the agent ignores commands within the input text.
  • Capability inventory: No dangerous tools or capabilities (such as file writes, network calls, or shell execution) are present in the skill.
  • Sanitization: The skill does not implement text sanitization or filtering.
  • [EXTERNAL_DOWNLOADS]: The skill recommends that the user download a companion tool, the-fool, from the author's GitHub repository at github.com/Jeffallan/claude-skills.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Sep 20, 2026, 12:43 AM
Security Audit — agent-trust-hub — architect