demand-draft
Pass
Audited by Gen Agent Trust Hub on Sep 20, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from an external file (
intake.md). This ingestion point represents a potential surface for indirect prompt injection if the source file contains instructions meant to override the agent's behavior. However, the risk is mitigated by the skill's mandatory 'Pre-draft Gate' and 'Post-sending Checklist,' which force the agent to explicitly verify facts, tone, and confidentiality before generating final output. These boundary markers ensure that external content is treated as data to be analyzed rather than instructions to be followed. - Ingestion points: Reads registration data from
~/.claude/plugins/config/claude-for-legal-zh/litigation-legal/demand-letters/[slug]/intake.md. - Boundary markers: Implements a multi-step 'Pre-draft Gate' (Confidentiality Filter, Admission Risk, Rights Reservation, etc.) to validate facts and posture before drafting.
- Capability inventory: File system read/write operations restricted to specific application configuration directories; no network access or shell execution tools are defined.
- Sanitization: The skill relies on structured verification steps (Yes/No prompts and explicit checklists) to filter and validate input data.
Audit Metadata