gh-address-comments

Warn

Audited by Gen Agent Trust Hub on Feb 17, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • Privilege Escalation (MEDIUM): The skill instructions explicitly direct the agent to run commands with "elevated network access" and "escalated permissions" (specifically sandbox_permissions=require_escalated). This is a deliberate attempt to bypass or override the agent's security sandbox constraints.
  • Persistence Mechanisms (MEDIUM): The PostToolUse hook is configured to append data to ~/.claude/gh-address-comments.log. Unauthorized writing to hidden directories in the home folder is a persistence-like behavior used to maintain a footprint on the host system outside the active workspace.
  • Indirect Prompt Injection (LOW): The skill is designed to fetch and process GitHub PR comments via scripts/fetch_comments.py. These external, attacker-controllable inputs are then used to guide the agent in "applying fixes," creating a surface for malicious instructions to hijack the agent's task.
  • Ingestion points: External GitHub PR/Issue comments fetched via scripts/fetch_comments.py.
  • Boundary markers: None; the agent is instructed to summarize and act on comments without delimiters or safety warnings.
  • Capability inventory: GitHub CLI (gh) execution and file system modification (to "Apply fixes").
  • Sanitization: None; the skill lacks any logic to sanitize or escape the content of the fetched comments before processing.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Feb 17, 2026, 04:56 PM