legal-research
Skill1 starsChanged 26 days agoApache-2.0View on GitHub
Internal grounding function (alias: legal-ground) for any legal system: deep investigation used inside any skill when authority or context is missing. Uses web search to collect provisions, case law, articles and commentary, then grounds with a citation ledger. Also answers standalone legal questions (CS: rešerše, judikatura, jak to je podle zákona?). Load alongside the legal engine; other skills invoke its phases inline.
10 other projects use a version of this.
Adopted by
Most recently taken up by
- joaoguirunas/team-ostheir own version, first seen Nov 15, 2026
- Cure-Consulting-Group/ProductEngineeringSkillstheir own version, first seen Nov 12, 2026
- Mharbulous/probanda-devtheir own version, first seen Nov 8, 2026
- Sprintlaw/legal-skillstheir own version, first seen Oct 17, 2026
- soloveliness-a11y/legal-risk-workflowtheir own version, first seen Oct 17, 2026
- code-lawyer/Legal-Agent-Skillstheir own version, first seen Aug 22, 2026
Other versions10 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 10
- moyupeng0422/legal-tools★ 25same name
- Youchu-lawhub/cn-litigation-toolkit★ 23same name
- joaoguirunas/team-os★ 6same name
- paulieb89/bouch-mcp-skills★ 4same name
- Cure-Consulting-Group/ProductEngineeringSkills★ 3same name
- code-lawyer/Legal-Agent-Skills★ 2same name
- kingselyjoe/legal-research-dsh★ 2same name
- soloveliness-a11y/legal-risk-workflow★ 2same name
- Mharbulous/probanda-dev★ 1same name
- Sprintlaw/legal-skills★ 0same name
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository5 other files it gives its agents.
- legal-disputeSkill
- legal-intakeSkill
- legal-projectSkill
- legal-reviewSkill
- legalSkill
The file
We split this skill into the rules it gives an agent
5 rules. 0 are common, 1 is distinctive, and 4 are very specific to this project.
0
Common
Most of the 4 edited versions we read keep these.
1
Distinctive
This author's own choices. Worth a look.
4
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 22
Read ../legal/SKILL.md first if the legal engine isn't loaded.
- Very specific: line 25
No profile (or a placeholder one): offer setup, or run provisional — mark it [PROVISIONAL — configure for tailored output] …
- Very specific: line 95
Write memo-<topic>.md plus a mandatory memo-<topic>.docx beside it (../legal/scripts/md2docx.py, after scan.py outbound).
- Distinctive: line 116
Stop: when the next source no longer changes the answer (minimum sufficient inquiry).
- Very specific: line 121
grounding: always: every relied-on authority has a ledger row.
“Common” is measured: kept by at least 6 in 10 of the 4 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
legal-research (alias: legal-ground)
Object of work. A legal question grounded in retrieved, checked authority (alias legal-ground).
Imports. Obey the capability files named in the front-matter
metadata.imports; do not restate them. This file carries only the mode's delta.
The internal grounding function. Other skills invoke these phases inline whenever a premise lacks authority — it is not a separate track the user must choose. It also answers standalone legal questions.
Read ../legal/SKILL.md first if the legal engine isn't loaded. The ledger
format and procedure are in ../legal/references/citation-ledger.md.
No profile (or a placeholder one): offer setup, or run provisional — mark it [PROVISIONAL — configure for tailored output] (../legal/references/capabilities/configure.md).
Variants (pick in the Gate Plan)
| Variant | When | Output |
| quick | One question, settled law, low stakes (fast-track) | Short answer + rule + authorities with ledger results, inline |
| memo (default for architect path) | A question the client or a decision relies on | Research memo + authorities.md |
| leads | Early exploration, unfamiliar area | Leads, not cites: where to look and what to check. Nothing in it is citable; it says so on top |
Phases
- Frame. Restate the question. Pin
JURISDICTION(split multi-jurisdiction questions into one sub-question per jurisdiction; never blend them) andEPOCH(the date the facts happened or the act will happen, which picks the version of the law). Decompose into sub-questions. Separate the facts the answer depends on ([USER], fact ids), and name any missing fact that would flip the answer. - Source plan. Per sub-question, follow the hierarchy in pack §1 (Sources & hierarchy) and format citations per pack §2 (Citation format):
constitutional and EU/international law → statute in the version in force at
EPOCH(pack §3 (Version in force)) → binding or unifying case law → supranational courts (e.g. CJEU/ECtHR) → lower courts → commentary. Name the portals to use and the search terms (anonymised: the legal point, never client facts). Note which sources are paywalled and whether the user has access. - Retrieve and read. Open each provision and decision. Read the
surrounding provisions (definitions, exceptions, transitional rules) and the
decision's reasoning, not just its headnote. Results from search engines and
other AI tools are leads.
- Use your web search tool to collect wider context first: relevant
statutes and their current wording, case law, doctrinal articles, blog
posts and practitioner commentary, regulatory guidance. Anonymise every
query (the legal point, never client names, amounts, or identifiers).
Everything found this way is a lead until its ledger row is
verified.
- Use your web search tool to collect wider context first: relevant
statutes and their current wording, case law, doctrinal articles, blog
posts and practitioner commentary, regulatory guidance. Anonymise every
query (the legal point, never client names, amounts, or identifiers).
Everything found this way is a lead until its ledger row is
- Ledger. One row per authority per proposition, in
authorities.md: version in force, verbatim quote match, weight, status. Contrary authority gets its own rows. A row isverifiedonly when every column supports it. - Synthesise. Per sub-question: the rule
[AUTH:A…]→ how it applies to these facts[INFER]→ contrary authority and how strong it is → open points and what would resolve them. Say where the law is unsettled, and present both readings instead of picking one silently. - Answer and memo. Lead with the answer.
Memo structure
# <Question, as restated>
⚠ <banner if any relied-on authority is not verified>
Jurisdiction: <…> · Epoch: <…> · Audience: <…>
## Short answer
<2–5 sentences: the answer, its confidence, the condition it depends on>
## Facts relied on
- <fact> [USER|DOC|ASSUMED] (F-id) ← a change here changes the answer
## Analysis
### <Sub-question 1>
Rule: … [AUTH:A1]
Application: … [INFER: A1 + F2]
Contrary view: … [AUTH:A4] — weight: …
Open point: …
## Recommendation and next steps
## Authorities
<the ledger, or a pointer to authorities.md>For AUDIENCE = client or business, add a plain-language summary above the
short answer and move the analysis below it.
Write memo-<topic>.md plus a mandatory memo-<topic>.docx beside it
(../legal/scripts/md2docx.py, after scan.py outbound). Screen follows
../legal/references/display.md: print the full memo on screen in the turn
that delivers it, and close with FILES — clickable absolute links to the
.docx, the .md, and authorities.md beside them. The links are additional,
never a substitute for the text.
Rules
- No memory citations. An authority recalled but not retrieved is a lead:
[ASSUMED] (verify),unverifiedin the ledger, and never quoted. If no retrieval tool is connected, say so at the start, deliver as leads, and put the engine's banner on top. - Version in force. A provision quoted in its current wording for a
question whose
EPOCHpredates an amendment is an error. Check the transitional provisions. - Binding vs persuasive is stated for every case cited. A lower-court decision is never presented as settled law.
- Quotes match verbatim, or they're paraphrased without quotation marks.
- Deadline questions give the rule, the trigger, the computation (pack §7 (Computing time & delivery rules))
and the conservative date, and go to
deadlines.mdwhen a matter folder exists. - Stop when the next source no longer changes the answer (minimum sufficient inquiry). Depth follows the stakes.
Checks
- grounding always: every relied-on authority has a ledger row; the
banner is on top if any isn't
verified. - trace for memos:
trace.py memo.md --authorities authorities.md --facts facts.md. - For
RISK = high, or advice a client will act on, propose the challenge capability (../legal/references/capabilities/challenge.md, lens: opposing counsel or judge) before final output.
Fan-out
Research fans out at the source: one read-only packet per sub-question and per jurisdiction (never blend jurisdictions in one packet), or per authority to retrieve-and-verify. The parent owns the memo and the ledger. Protocol: ../legal/references/capabilities/orchestrate.md.
What this mode does not do
- Give legal advice: it grounds a question in authority; the answer and its application are the lawyer's.
- Cite from memory: a proposition with no retrieved, checked source is tagged
[verify], never asserted. - Skip currency: where the law can move, the retrieval is required and its absence is stated.
Counter-example (illustration)
Illustration, not law. Asked whether a clause is enforceable, it answers from a case it "recalls". Mistake to catch: the citation was never retrieved, has no ledger row, and is presented as settled. Correct: retrieve it, record the ledger row, tag its tier — or say the law was not checked.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

