termination-review
Pass
Audited by Gen Agent Trust Hub on Jul 3, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [SAFE]: The skill follows security best practices by implementing explicit warnings and gates for high-risk actions (terminating an employee) and identifying legal consequences.
- [SAFE]: It enforces the use of external research tools and primary sources over model memory to prevent legal misinformation and hallucinations.
- [SAFE]: File system operations are restricted to specific, non-sensitive paths within the agent's own plugin configuration directory (~/.claude/plugins/config/).
- [PROMPT_INJECTION]: The skill ingests untrusted data through user-provided descriptions and attached documentation (Step 1). While no malicious intent is evident in the skill itself, this creates an attack surface where instructions embedded in processed documents could potentially influence agent behavior.
- Ingestion points: basic fact collection and documentation attachments in SKILL.md.
- Boundary markers: Absent for user-provided data.
- Capability inventory: File read/write access to ~/.claude/plugins/config/claude-for-legal/employment-legal/ and access to legal research tools.
- Sanitization: None explicitly mentioned for external content.
Audit Metadata