build-from-issue
Pass
Audited by Gen Agent Trust Hub on May 14, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill executes various local shell commands using
gh(GitHub CLI) for issue management,gitfor repository operations (branching, commits, pushes), andmisefor running pre-commit hooks and end-to-end test suites. - [EXTERNAL_DOWNLOADS]: The skill retrieves issue content, comments, and repository metadata from GitHub, which are used to inform the implementation plan and workflow state.
- [PROMPT_INJECTION]: Because the skill processes untrusted content from GitHub issues and comments, it has an inherent surface for indirect prompt injection. This risk is effectively mitigated by several design choices:
- Ingestion points: Untrusted data enters the agent context via
gh issue viewcommands inSKILL.md(Steps 1 and 2). - Boundary markers: The skill uses explicit markers (
🏗️ build-planand🏗️ build-from-issue-agent) to distinguish agent-generated content from human feedback and third-party inputs. - Capability inventory: The agent has the capability to write code to the local file system, execute shell commands through the project's test runner (
mise), and push changes to remote repositories. - Sanitization: The skill relies on sub-agent review (
principal-engineer-reviewer) and, most critically, a mandatory human authorization gate (state:agent-readylabel) before any execution occurs.
Audit Metadata