skills/dzhng/skills/implement-spec/Gen Agent Trust Hub

implement-spec

Pass

Audited by Gen Agent Trust Hub on Nov 11, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTIONCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from the repository to drive its logic.
  • Ingestion points: spec README, slices, and repo README serve as the primary instructions for the implementation task.
  • Boundary markers: Absent. The agent is explicitly told that the spec is the "implementation plan" and that user constraints and experimental evidence are binding.
  • Capability inventory: The skill can perform git operations (commit, worktree), write to files (decisions.md, SKILL.md), spawn concurrent subagents, interact with a browser, and execute project tests.
  • Sanitization: None. The agent directly interprets instructions from the project files.
  • [DYNAMIC_EXECUTION]: The skill dynamically loads other skills based on names provided in the specification files ("Load any skills named by the spec"). This allows the content of the repository (data) to control which functional logic (skills) the agent activates at runtime.
  • [COMMAND_EXECUTION]: The agent is directed to "run the focused tests" and "use the real browser/harness" provided in the project. This involves executing arbitrary code or scripts present in the repository being worked on.
  • [PROMPT_INJECTION]: The skill contains instructions that promote high autonomy and reduce human oversight. Phrases such as "If any slice or global TODO is still open, go straight back to step 1... no pause for acknowledgement" and "A goal or implementation NEVER stops to wait on the user" encourage the agent to continue executing a loop of commands without waiting for user verification, which could exacerbate the impact of an injection or a logic error.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 11, 2026, 11:01 AM