principle-redesign-from-first-principles

Pass

Audited by Gen Agent Trust Hub on Sep 3, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONNO_CODE
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill instructions create a potential surface for indirect prompt injection by directing the agent to process untrusted data from the local environment. 1. Ingestion points: The agent is instructed to "Read all affected files and understand the current design holistically" (SKILL.md). 2. Boundary markers: There are no instructions providing delimiters or warnings to ignore commands embedded within the files being read. 3. Capability inventory: While this skill defines no tools, it is used by an agent that typically has file system modification and potentially shell execution capabilities. 4. Sanitization: No sanitization or validation of the ingested file content is specified.
  • [NO_CODE]: The skill consists entirely of natural language instructions and YAML metadata without any executable scripts or command-line operations.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 3, 2026, 04:28 PM
Security Audit — agent-trust-hub — principle-redesign-from-first-principles