fix-broken-integration-tests

Pass

Audited by Gen Agent Trust Hub on Jun 17, 2026

Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill uses the Bash tool to execute shell commands for repository management, GitHub CLI interactions, and Jira queries. This includes operations like creating branches, pushing code, and opening pull requests.\n- [COMMAND_EXECUTION]: The workflow involves dynamic generation and modification of Go source code (e.g., test sweepers and timestamp adjustments) based on predefined templates in fix-patterns.md, followed by test execution via a helper skill.\n- [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it processes data from external systems:\n
  • Ingestion points: Jira work item descriptions (SKILL.md, Phase 1) and CI failure logs (SKILL.md, Phase 2).\n
  • Boundary markers: Absent; the skill does not use specific delimiters or instructions to ignore potential commands within the ingested data.\n
  • Capability inventory: The skill has extensive permissions including arbitrary command execution (Bash), file modification (Write, Edit), and the ability to invoke other skills or agents.\n
  • Sanitization: None; the ingested content is used directly to influence the agent's diagnostic and fixing logic.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 17, 2026, 08:38 PM
Security Audit — agent-trust-hub — fix-broken-integration-tests