菜单
Use when running structured, adversarial analysis across large case-file directories — extracts facts, claims, and legal views into XML metadata via a stateless R.A.L.P.H. loop, then synthesizes contradictions and timeline across all files. Proof of concept; not a finished product.
默认使用会先检查来源的 Prompt；你也可以切换为直接命令，或下载本地副本。
决定是否安装前，请先阅读 SKILL.md，以及 SkillsMP 当前展示的配套文件。
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
直接命令不会经过审查 Prompt；运行前请先检查来源。
npx skills add https://github.com/ThomasMoreAI/legal-skills-open --skill case-file-analyzer命令会保持在同一行。复制前请横向滚动并检查完整内容。
想先保存到本地？可下载 SkillsMP 当前能够提供的文件。
Use when retrieving arbitration institutional rules (ICC, LCIA, SCC, SIAC, HKIAC, VIAC, МКАС/МАК при ТПП України, UNCITRAL) — fetching current version, verifying redaction applicable to the date of arbitration agreement, constructing URLs for official rule texts
Use when choosing the Polish legal regime for letters, requests, applications, complaints, petitions, public-information requests, KPA filings, PPSA complaints, RODO access requests, registry extracts, court-file access, tax/ZUS/cudzoziemcy/USC procedures, or professional lawyer letters. Prevents mixing UDIP, KPA, PPSA, RODO, registry, special-procedure, and advocate/radca letter regimes.
Use when preparing applications for recognition and enforcement of foreign arbitral awards in Poland, applications for setting aside arbitral awards under KPC art. 1205–1211, or opposing such applications — mapping Article V of the 1958 New York Convention to art. 1214–1215 of the Polish KPC, identifying grounds for refusal, structuring public policy arguments
基于 SOC 职业分类
正在显示 SKILL.md
| name | case-file-analyzer |
| title | Stateless Case File Analyzer (Ralph Loop Edition) |
| description | Use when running structured, adversarial analysis across large case-file directories — extracts facts, claims, and legal views into XML metadata via a stateless R.A.L.P.H. loop, then synthesizes contradictions and timeline across all files. Proof of concept; not a finished product. |
| author | LegalQuants |
| author_url | https://github.com/LegalQuants/lq-skills/tree/main/skills/case-file-analyzer |
| license | Apache-2.0 |
| version | 0.1.0 |
| execution_mode | open |
| jurisdiction | general |
| practice | litigation |
| language | en |
| tags | ["case-files","multi-agent","adversarial","ralph-loop","verification"] |
⚠️ Proof of concept. This is an open-source demo showing how far agentic case file revision can be implemented efficiently. It is not a finished ready-to-use product.
| Attribute | Value |
|---|---|
| Description | Stateless, autonomous AI agent utilizing the R.A.L.P.H. pattern for adversarial and structured case file analysis. Runs parallel verification for factual claims, logic, and completeness across massive document directories. |
| Author | Attorney Dennis G. Jansen / J-Law.de |
| Version | 3.1.0 |
| Jurisdiction | Agnostic |
| Last Reviewed | 2026-05 |
| Last Reviewed By | LegalQuants (QA remediation) |
| Tags | legal-analysis, quality-control, multi-agent, verification, adversarial, certificate, ralph-loop |
What it does:
What it doesn't do:
PRD.md or a state file, it does not exist.When triggered via "scan case files", the agent orchestrator must execute the following sequence via the R.A.L.P.H. loop:
Initialize the workspace with the following reference modules:
Audience. Qualified litigators, dispute-resolution lawyers, paralegals, and litigation-support staff who already work with structured case files and understand the role of fact-vs-claim distinction. Not for unsupervised use by non-lawyers, opposing parties, or self-represented litigants. Treat outputs as worked input for a lawyer, not a deliverable.
Work shape — dual. This skill operates in two distinct shapes across its pipeline. Both shapes apply on every run:
references/definitions.md. The model checks each unit against a known template. Outputs are lists of extracted units with XML tags, not conclusions.The dual shape matters because confidence and escalation rules differ between the two stages, and downstream readers must not treat the synthesis output as if it were the same kind of artifact as the extraction output.
This skill provides legal support, not legal advice. A completed synthesis is a structured restatement of what is in the case files, with timeline and contradiction flags. It is not a case theory, a litigation recommendation, an admissibility ruling, or a sign-off on any factual or legal conclusion.
Privilege and confidentiality of case files being scanned. Case files routinely contain privileged client communications, work product, settlement-privileged material, and material covered by data-protection regimes (GDPR, attorney-client privilege under the applicable jurisdiction, professional-secrecy rules). Running this skill against such files moves their contents through whichever model API is configured in references/configuration.md and writes structured extracts to disk in the output directory.
Before invoking on real matter files:
summary_*.xml, perspective_review.xml, and final_synthesis.xml artifacts as work product — apply the same retention, access, and destruction controls as you would to a paralegal's litigation summary.Accountability gap. Stateless agents are designed to forget. The R.A.L.P.H. loop deliberately discards context after each iteration, which is good for context-window hygiene but means there is no in-model record of why a given extraction was made, what was discarded, or what edge cases the agent encountered. The on-disk artifacts (PRD.md, progress.txt, summary_*.xml) are the only record. A qualified lawyer must:
summary_*.xml outputs against the source files before any extracted fact is relied upon.final_synthesis.xml against the underlying corpus before any contradiction, timeline entry, or holistic assessment is acted upon.Confidence is reported separately for the extraction phase and the synthesis phase, because the two work shapes warrant different reliability profiles.
Extraction (Steps 1-2):
<fact>, opinion, legal view); quotes match the source verbatim; no [FAILED] markers in progress.txt for this file.[FAILED], classification rules in definitions.md could not be applied confidently, or the source contains material that does not fit the schema (e.g., diagrams, images, handwritten annotations). Do not include in synthesis without lawyer triage.Synthesis (Step 3):
summary_*.xml files have High extraction confidence; the timeline is reconstructable from <fact> tags alone; identified contradictions are between explicit, quoted claims.<status>REVIEW REQUIRED</status> in final_synthesis.xml and stop.Stop the loop and route to the responsible lawyer when:
progress.txt, and exit.<fact> across runs, or the synthesis step flags a contradiction that depends on which sub-agent's extraction is trusted. Do not paper over the disagreement in the synthesis; surface both readings and stop.references/configuration.md, and the substitution was not approved — halt rather than proceed with a model that may not satisfy the reasoning or confidentiality profile required for the matter.On escalation, the agent writes the trigger and the file path to progress.txt, marks the synthesis status as REVIEW REQUIRED if Step 3 is in progress, and exits cleanly. The responsible lawyer decides whether to override, restart, or abandon the run.
LegalQuants applied design-quality QA against the Legal Skill Design Framework and added the Audience-and-Work-Shape, Scope-and-Legal-Use (covering legal-advice-vs-support, privilege/confidentiality of case-file inputs, and the accountability gap created by stateless operation), Confidence Bands, Out of Scope, and Escalation sections. The frontmatter was normalized to standard Anthropic Skills format and the file renamed from agent.md to SKILL.md. The technical content (R.A.L.P.H. pattern, 3-step execution process, attribute table, references) is unchanged. Dennis G. Jansen / J-Law.de is retained as author of the underlying skill.