skills/dan86de/skills/prove-change/Gen Agent Trust Hub

prove-change

Warn

Audited by Gen Agent Trust Hub on Nov 6, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content from the branch history and code changes, creating a surface for injection attacks.\n
  • Ingestion points: SKILL.md (Step 2) extracts commit messages (Deviations:, possible bug:) and changed files using git log and git diff.\n
  • Boundary markers: The subagent briefs in references/bug-prover-brief.md and references/spec-prover-brief.md use markdown sections but lack rigorous escaping or sanitization for the interpolated git log content.\n
  • Capability inventory: The skill possesses significant capabilities, including executing shell commands (git, subagent-provided scripts) and writing files to the working directory.\n
  • Sanitization: The skill does not sanitize the content of commit messages or diffs before passing them to the subagents.\n- [COMMAND_EXECUTION]: The skill is designed to execute arbitrary shell commands provided by subagent outputs.\n
  • In SKILL.md Step 8, the instructions mandate running the command string returned by the subagents. This creates a risk where a compromised or misaligned subagent could provide malicious shell commands that the primary agent then executes in the host environment or worktree.\n- [DYNAMIC_EXECUTION]: The subagents are instructed to perform environment setup tasks that involve executing code from the branch.\n
  • In references/bug-prover-brief.md and references/spec-prover-brief.md, provers are told to install dependencies and build what a driver needs. This results in the execution of build scripts and package managers on untrusted branch code.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Nov 6, 2026, 08:19 PM