decodie-ask

Pass

Audited by Gen Agent Trust Hub on Aug 28, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill is instructed to read files from ~/.decodie/rules/*.md and .decodie/rules/*.md and treat their content as 'additional instructions'. This allows untrusted data to influence the agent's behavior without sanitization or boundary markers.
  • Ingestion points: ~/.decodie/rules/*.md, .decodie/rules/*.md, index.json, and entry session files.
  • Boundary markers: Absent. The skill explicitly instructs the agent to 'Treat their content as additional instructions controlling documentation output'.
  • Capability inventory: The agent has the capability to read arbitrary local files if they are listed in the references array of a Decodie entry.
  • Sanitization: Absent. There is no mention of validating the paths in the references array or sanitizing the content of rule files.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 28, 2026, 08:19 PM
Security Audit — agent-trust-hub — decodie-ask