reproduce-bug-report

Pass

Audited by Gen Agent Trust Hub on May 20, 2026

Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION]: The skill is vulnerable to Indirect Prompt Injection (Category 8) because it ingests and processes untrusted content from external bug reports (GitHub, Linear, etc.). An attacker could craft a malicious bug report containing instructions to divert the agent from its primary task.
  • Ingestion points: The skill ingests 'Bug report context' and 'Assigned repro path or hypothesis' directly from external issue bodies and comments.
  • Boundary markers: There are no explicit boundary delimiters or 'ignore embedded instructions' warnings around the untrusted data; the skill uses placeholders like ''.
  • Capability inventory: The skill uses run_agents with remote.computer_use_enabled: true, allowing remote shell access, file system writes (screenshots/logs), and application installation.
  • Sanitization: No sanitization or validation of the ingested bug report content is specified before it is processed by the agent.
  • [COMMAND_EXECUTION]: The skill executes commands in a remote environment via 'Oz cloud agents' with computer use enabled to interact with applications and capture screenshots. This is a primary intended function of the skill.
  • [EXTERNAL_DOWNLOADS]: The skill instructs the agent to download and install application artifacts (versions, builds, or channels) to replicate the reporter's environment. The instructions include a preference for 'repository-approved' artifacts, which mitigates risk.
Audit Metadata
Risk Level
SAFE
Analyzed
May 20, 2026, 09:01 PM
Security Audit — agent-trust-hub — reproduce-bug-report