vibe-coding
Pass
Audited by Gen Agent Trust Hub on Oct 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill acts as a central router that ingests and processes untrusted data, including user instructions, repository history (Git diffs), and external plan/specification artifacts, to determine execution flow.
- Ingestion points: User-provided instructions, git-backed diffs, task lists, and plan files processed during the
review,plan-execution, anddebug-and-repairphases defined inSKILL.md. - Boundary markers: The skill implements a 'Goal-Alignment Gate' (in
SKILL.md) and 'shared-contract' blocks (inSKILL.mdand all reference files) that explicitly define execution limits and audit requirements to prevent unintended command execution. - Capability inventory: The router has the capability to perform state-changing edits, execute shell commands (e.g.,
sed,mv,git checkout), and conduct repository maintenance as authorized in the 'Workflow Phases' table ofSKILL.md. - Sanitization: Employs mandatory human-risk checkpoints and explicit confirmation gates for any operations involving irreversible changes, security boundaries, or external side effects.
Audit Metadata