work-integrate

Pass

Audited by Gen Agent Trust Hub on Sep 10, 2026

Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill directs the agent to execute series of Git commands to manage local branch merges, directory moves, and index adjustments. This includes using complex patterns like git merge --no-ff --no-commit followed by git rm with globbing to prevent audit reports from entering the base branch (SKILL.md, references/archive.md). \n- [INDIRECT_PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it processes untrusted data from repository files. \n
  • Ingestion points: It reads content from progress.md, docs/specs/, and .agents/MEMORY.md to verify task completion and resolve language preferences (SKILL.md, references/archive.md). \n
  • Boundary markers: There are no explicit instructions to the agent to treat the contents of these markdown files as untrusted or to ignore instructions embedded within them. \n
  • Capability inventory: The agent can execute Git commands and invoke other automated skills such as git-commit, quality-check, code-review, and trace-validate. \n
  • Sanitization: The skill parses specific status tokens (e.g., 'Done') but does not perform general validation or sanitization of the file contents before they enter the LLM context.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 10, 2026, 08:06 PM
Security Audit — agent-trust-hub — work-integrate