skills/dzhng/skills/explore-unknowns/Gen Agent Trust Hub

explore-unknowns

Pass

Audited by Gen Agent Trust Hub on Oct 29, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill's primary function is to perform a systematic sweep of the user's codebase ("the territory") to identify project constraints and resolve unknowns. This creates a surface for indirect prompt injection where malicious instructions embedded in the scanned files could manipulate the agent's analysis or its decision-making during the "Preference Checkpoint" phase.
  • Ingestion points: Identified in references/stage-1-known-knowns.md ("scan the territory") and references/stage-4-unknown-unknowns.md ("Sweep every file the task touches"), where the agent reads arbitrary files from the environment.
  • Boundary markers: The instructions lack specific guidance on using delimiters or safety warnings to treat content from the territory as untrusted data.
  • Capability inventory: The agent can read files, generate interactive HTML mocks, and automate decisions on the user's behalf if granted permission via the "Preference Checkpoint" logic.
  • Sanitization: No sanitization or escaping of ingested file content is mentioned before it is processed by the agent or subagents.
  • [DYNAMIC_EXECUTION]: The skill supports the creation of HTML artifacts with inline CSS and JavaScript for user interaction and mocks in references/stage-3-unknown-knowns.md. Although the skill explicitly forbids external network requests in these artifacts (as stated in SKILL.md), the dynamic generation of executable code based on analysis of potentially untrusted data from the "territory" presents a vector for malicious scripts to be rendered if the source material is compromised.
Audit Metadata
Risk Level
SAFE
Analyzed
Oct 29, 2026, 05:08 PM