triage-pr

Pass

Audited by Gen Agent Trust Hub on Aug 31, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONPROMPT_INJECTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection due to its automated processing of untrusted external content.
  • Ingestion points: The agent ingests data from GitHub Actions failure logs (SKILL.md Step 2) and PR review comments/threads via the scripts/review-threads.mjs fetcher (SKILL.md Step 8).
  • Boundary markers: Absent. The instructions do not provide technical delimiters or specific guidance to ignore embedded instructions when the agent parses log content or review feedback.
  • Capability inventory: The agent has extensive capabilities including file modification (Read, Edit, Write), shell execution via gh, git, node, pnpm, and npx, and the ability to manage Linear issues via MCP tools.
  • Sanitization: Absent. While basic markdown escaping is performed for table formatting, no sanitization or filtering is applied to the ingested content to prevent the execution of injected instructions.
  • [PROMPT_INJECTION]: A static analysis hint regarding action concealment was identified in SKILL.md. Upon review, the instruction to "stay quiet while watching" is a documented design choice intended to reduce user interruptions during long-running CI tasks (Step 6), not a malicious attempt to hide unauthorized actions. This pattern is considered benign in the context of a triage automation tool.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 31, 2026, 12:05 PM
Security Audit — agent-trust-hub — triage-pr