| name | legal-mdl-audit-ignacio-adrian-lerer |
| description | Audits legal AI outputs and workflows for honest compression: unnecessary complexity, false simplicity, excessive caveats, hidden uncertainty and poor cost per legally acceptable output. |
| license | agpl-3.0 |
| metadata | {"author":"Ignacio Adrián Lerer","license":"agpl-3.0","version":"2026-05-31"} |
Legal MDL Audit
What this skill does
This skill reviews whether a legal AI answer, prompt, workflow, benchmark result, contract review, memo, compliance report, or agent chain is as simple as it can safely be.
It is inspired by Minimum Description Length: good legal reasoning should explain more with less structure, but never by hiding material uncertainty.
Audit categories
Classify the material as:
- APPROVE: lean and still legally safe.
- APPROVE WITH CONSTRAINTS: complexity is justified, but reliance needs stated limits.
- REWRITE: the output is too complex, repetitive, expensive, or hard to audit.
- QUARANTINE: the output is falsely simple and hides material uncertainty.
Checklist
Review:
- Rules: how many legal propositions are needed?
- Exceptions: how many carve-outs or qualifications are doing real work?
- Conditions: what facts, dates, forums, sources or procedural states must hold?
- Sources: are citations enough, excessive, or missing?
- Uncertainty: what must remain visible?
- Workflow cost: how many model/tool/human steps were needed?
- Output value: did added complexity improve legal acceptability?
Output format
Return:
- Verdict.
- Complexity drivers.
- Hidden uncertainty or omitted hard cases.
- What can be simplified.
- What must not be removed.
- Safer shorter version, if requested.
Rules
- Do not reward short answers that erase legal uncertainty.
- Do not reward long answers that add caveats without improving reliance.
- Prefer cost per legally acceptable output over cost per token or API call.
- Preserve source gaps, authority boundaries and human-review gates.