tech-design

Warn

Audited by Gen Agent Trust Hub on Sep 4, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill instructs the agent to execute shell commands (grep, git) using variables like NNN-slug derived from directory names and user input. The lack of explicit sanitization instructions for these variables could potentially lead to command injection if a directory name contains shell metacharacters. \n- [INDIRECT_PROMPT_INJECTION]: The skill processes content from spec.md and other project files which are considered untrusted inputs. \n
  • Ingestion points: Specification files (meta/specs/NNN-slug/spec.md), project conventions (AGENTS.md), and rule files (.claude/rules/). \n
  • Boundary markers: The protocol does not define delimiters or specific instructions to disregard instructions embedded within the spec files. \n
  • Capability inventory: The skill can write to the file system (spec.md), execute shell commands (git, grep), and spawn additional subagents. \n
  • Sanitization: There are no requirements to sanitize or escape the content of the specification before it is used to influence the design or the logic of subagents. \n- [DYNAMIC_EXECUTION]: The skill dynamically spawns various subagents (Opus subagents, propose-alternatives, deep-modules, find-gaps) to handle different parts of the design and verification process based on the input specification's complexity.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Sep 4, 2026, 02:45 AM
Security Audit — agent-trust-hub — tech-design