maintainer-review

Pass

Audited by Gen Agent Trust Hub on Jul 29, 2026

Risk Level: SAFECOMMAND_EXECUTIONREMOTE_CODE_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill enables the agent to execute shell commands using the GitHub CLI tool.
  • Evidence: The workflow instructions (SKILL.md, Section 1) allow the agent to run the gh command on the host machine if the user explicitly requests it.
  • [REMOTE_CODE_EXECUTION]: The skill defines a process for executing code patches retrieved from remote GitHub Pull Requests.
  • Evidence: The "Stage 2: approved runtime probe" (SKILL.md, Section 4) instructs the agent to perform focused local reproductions of claims found in pull requests.
  • Mitigation: The skill mandates explicit user approval before execution and requires the use of a specific tool ($runtime-behavior-probe) that is expected to manage environment gates and cleanup.
  • [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it is designed to ingest and analyze untrusted data from external sources.
  • Ingestion points: The agent reads content from GitHub issue URLs, PR bodies, comments, and reproduction steps (SKILL.md, Section 1).
  • Boundary markers: The skill does not implement explicit delimiters or boundary markers to distinguish instructions from untrusted external data.
  • Capability inventory: The agent possesses powerful capabilities, including command execution and code reproduction, which could be exploited if the agent follows instructions embedded in a malicious GitHub issue.
  • Sanitization: No explicit sanitization or filtering of the ingested external content is described in the workflow.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 29, 2026, 03:18 PM
Security Audit — agent-trust-hub — maintainer-review