Use when a lawyer, legal team, or client needs to assess AI agents that can act on someone's behalf: send messages, search, draft, file, pay, delete, connect to accounts, use tools, or rely on external data. Produces a practical delegation, oversight, accountability, and control audit for legal operations.
SkillScore86Maintained
Automatic
◆Legal & Contracts pack
Last verified 8/16/2026
What it changes
Passed 19 of 22 test cases with the skill, and 5 of 22 without it.
Without skill23%
With skill86%
Checks it passed that the model alone missed
The verdict is PASS
The verdict is BLOCK
The overall verdict is PASS
and 10 more in the full report
14 fixed5 passed either way3 still failing0 broke
Pass-rate lift
+64 pp
Tokens
+1%
Agent turns
0%
Response time
-44%
Model gemini-3.6-flashMeasured 8/16/2026v119 of 22 cases passed with skill
Tokens, turns and time describe resource use, not a dollar price.
See it in action
One of 14 test cases the skill fixed. Same model, same prompt, with and without the skill.
1 of 14
Prompt
A law firm is using a tool that reads counterparty redlines on standard non-disclosure agreements and inserts proposed counter-revisions into Word document files saved locally for attorney review. The tool cannot send emails or touch production databases. Conduct a delegation governance assessment.
What was checked, without and with the skill
Checked
Without skill
With skill
The verdict is PASS
Missed
Passed
The autonomy level is classified as draft_only
Missed
Passed
case-08 · gemini-3.6-flash · 8/16/2026 · answers are capped and redacted
What's inside
The instructions your model reads when the skill loads.
SKILL.md128 linessingle file
Lawve Agentic Delegation Audit
Use this skill to help lawyers evaluate whether an AI agent workflow is safe enough to use in legal or business operations.
Core idea:
When AI only answers, the user evaluates an output.
When AI acts, the user delegates authority.
Delegated authority needs controls, logs, revocation, and accountability.
Use When
A client wants to use an AI agent for legal, compliance, business, or administrative work.
A product, law firm, or legal ops team is connecting an agent to email, files, CRM, court portals, payments, messaging, calendars, databases, browser tools, or code execution.
A lawyer needs to explain the risks of agentic AI in practical governance terms.
A workflow may send, file, delete, approve, purchase, publish, sign, deploy, or change access.
A system claims to be "autonomous", "agentic", "assistant with tools", "AI employee", "legal copilot", or "workflow agent".
Do Not Use For
Pure doctrinal legal research with no agentic action.
Final legal opinions without jurisdiction-specific legal review.
Technical implementation of an agent runtime.
Approving production use of an agent without evidence, logs, and controls.
Intake
Collect only what is necessary:
What task does the agent perform?
Who is the human principal?
Who deploys or operates the agent?
What systems, accounts, files, channels, or tools can it access?
What actions can it take without step-by-step approval?
Which actions are irreversible, external, financial, legal, confidential, or reputation-sensitive?
What logs exist and who can read them?
How can the user pause, revoke, appeal, or correct the agent?
What data can influence the agent, including emails, webpages, chats, files, tickets, and prompts?
Routes any legal task to the right LLM, like OpenRouter but for legal work and grounded in benchmarks instead of brand loyalty. Built from mid-2026 legal evals (legalbenchmarks.ai, Vals AI × Stanford LegalBench across 124 models, Harvey's Legal Agent Benchmark, the Atticus Project's CUAD/MAUD/ACORD) plus translation evidence (WMT25, SwiLTra-Bench, ArabLegalEval). Covers five verticals: contract drafting, info extraction, legal research, contract review, and legal translation (including Arabic/ME