plugin-workflow

Pass

Audited by Gen Agent Trust Hub on Oct 31, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted or project-specific data from multiple files and passes it to subagents, creating a surface for indirect prompt injection.
  • Ingestion points: The skill reads architecture.md, plan.md, creative-brief.md, and parameter-spec.md as documented in references/stage-2-dsp.md and references/stage-3-gui.md.
  • Boundary markers: Absent. The content of these files is interpolated directly into the subagent prompts (e.g., in the constructPhasePrompt logic) without delimiters or instructions to ignore embedded directives.
  • Capability inventory: The orchestrator can invoke other agents via the Task tool and execute shell commands via the Bash tool, providing a path for injected instructions to reach high-privilege operations.
  • Sanitization: Absent. There is no evidence of filtering or escaping the contents of the ingested files before they are sent to the subagents.
  • [COMMAND_EXECUTION]: The skill performs significant local command execution to manage the build system, state, and file synchronization.
  • Evidence: The skill executes several local scripts including .claude/utils/sync-brief-from-mockup.sh (in references/creative-brief-sync.md), ./scripts/build-and-install.sh (referenced in references/stage-1-foundation-shell.md), and numerous git, sed, and shasum commands for state persistence and auditing in references/state-management.md.
Audit Metadata
Risk Level
SAFE
Analyzed
Oct 31, 2026, 03:40 AM