setup-github-project
Pass
Audited by Gen Agent Trust Hub on Sep 6, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from GitHub repository issues and project items to assess the current state and propose reconciliation actions.
- Ingestion points: Untrusted content such as issue titles, labels, and project item metadata is ingested via 'gh issue list' and 'gh project item-list' in 'scripts/inspect.ps1'.
- Boundary markers: The 'SKILL.md' file defines a 'Non-negotiable approval gate' that requires the agent to present all proposed changes to a human executor and wait for explicit approval before any write operations occur.
- Capability inventory: The skill possesses capabilities to create repositories, milestones, and projects, and modify repository settings using the 'gh' CLI (as seen in 'scripts/reconcile-milestones.ps1').
- Sanitization: Data is handled using structured PowerShell objects and output as JSON. The manual verification step in 'SKILL.md' ('Run verification... require zero proposed mutations') serves as a primary defense against malicious data influencing agent behavior.
Audit Metadata