Skip to content

ThomasMoreAI/legal-skills-open / au/contracts/skills/australian-legal-research-acor8826

australian-legal-research-acor8826

Verified

Comprehensive red-teamed Australian legal research on freely available databases — AustLII, CaseLaw NSW, the legislation registers, BarNet Jade (free tier), and web search. Two stages that always run together: Stage 1 — Orchestration: decompose the task into a research goal and per-agent subgoals; Stage 2 — Diverge-converge: dispatch a fixed roster of agents (legislation, case law, evidence, jurisdiction/court-rules, secondary, red-team) as parallel Claude Code Task subagents, converge their reports, then run a five-check citation-integrity gate before reporting, with AGLC4 citations. Triggers: AustLII, Jade/jade.io, Australian case citation, find/verify a citation, fetch an Act/section, 'is this still good law', 'what's the law on X', 'find authorities for', 'find cases with similar facts', AGLC4, table of authorities. USE BEFORE relying on training knowledge for any Australian citation, quote, statutory text, judge, date, or holding. For paywalled material (logged-in JADE), use jade-research.

references/agent-caselaw.md

View on GitHub
# Case-law Agent — Brief, Strategies, Ranking, Output

The case-law agent is one of the fixed roster the orchestrator dispatches in **Step 3** of the
orchestration loop. It is **one Claude Code Task subagent per cycle** (not one per strategy): it
runs all four search strategies *internally*, converges them, and writes one ranked case table to
`agent-reports/cycle<N>-caselaw.md`. It is dispatched in the same turn as the other engaged
area-agents and runs in an isolated context — everything it needs is in its prompt.

## What this agent owns

Build the case-law that answers the framed question: the **leading/binding authority**, the
authorities **applying** it, the closest **factual analogues**, and the **citator trail** around the
key cases — every one verified on a primary source, pinpointed, and ranked. It assigns each case a
**how-cited** classification so the orchestrator can tell an application of the ratio from a passing
history mention.

It does **not** own: statutory currency / amendment history (legislation agent), secondary framing
(secondary agent), or the systematic contrary-case sweep (red-team). If it trips over a strongly
contrary case while searching, it records it as a flag for the red-team — but it does not build the
adversary brief itself. Stay in lane.

## Dispatch envelope

The orchestrator builds this agent's Task prompt from the **subagent prompt template in
`references/templates.md` §2** plus the role prompt below, and injects: the framed goal (read-only —
question(s), jurisdiction/hierarchy, temporal anchor(s), factual matrix, propositions); this agent's
**case-law task-list slice** (each task tagged major/moderate/minor); the four strategy axes; a
pointer to read `references/retrieval-mechanics.md` before any fetch; and the output path + schema.
Never leave a slot empty — the agent sees nothing else.

## Role prompt (paste verbatim into the Task dispatch)

> You are the **Case-law agent** for a legal-research cycle. Your job is to build the case-law that
> answers the framed question: the leading/binding authority, the authorities applying it, the
> closest factual analogues, and the citator trail around the key cases — every one verified on a
> primary source.
>
> **FIRST:** read `references/retrieval-mechanics.md` in full before any fetch, and
> `references/aglc4-citation.md` for citation formatting. Prefer jurisd
> Layer 0 if it is registered (`search_cases` for topic/authority search, `fetch_document_text`
> for full text); for **who-cites-this-case**, jurisd v0.5.0 has **no live citator** — use the
> citator recipe in `references/retrieval-mechanics.md` (Jade's free-tier citation panel is the
> best seed; harvest it via `site:jade.io` search + Chrome, then confirm each candidate on tier 1),
> seeded by `search_cases` on the case name.
> Otherwise use the fallback stack. **Never**
> use AustLII SINO/LawCite (Cloudflare-walled) or bare `site:austlii` index pages.
>
> Run **ALL FOUR** strategies below internally and diverge across them — do not pick one:
> 1. **By-statute** — find decisions on the controlling provision(s) named in the goal/ledger.
>    Search the provision as a phrase and noteup the leading case under it. Catches the line of
>    authority a doctrinal search alone misses.
> 2. **By-doctrine** — find the leading authority on the doctrinal proposition, then its
>    applications. Identify the case everyone cites for the proposition before its downstream uses.
> 3. **By-analogy / factual-matrix** (**highest value**) — distil the factual matrix into 3–6 fact
>    predicates and search for decisions whose *facts* resemble them; rank by factual + doctrinal
>    proximity. This is where the most useful precedent for a self-represented litigant comes from.
> 4. **By-citator-trail** — from each leading case, walk **forward** (who has cited it — use the
>    citator recipe in retrieval-mechanics.md, seeded by `search_cases` on the case name; jurisd's
>    offline `find_citing` is legislation-only today, so it will not list citing cases yet) and
>    **backward** (what it cites — from the judgment's own text). Catches both the line that built on
>    it and the authority it rests on.
>
> For **every** case you rely on:
> - **Fetch the judgment from a primary source and VERIFY it** — parties (correct order), court,
>   year, medium-neutral citation number, authorised report volume/page, bench, and date. Never
>   assert any of these from memory. If you cannot fetch it, mark it `unverified` and say so.
> - Extract the **relevant paragraph numbers** and the **minimum verbatim quote(s)** needed (in
>   quotation marks, each tagged with its `[N]`; keep each quote short).
> - Write a **2–5 sentence factual summary** — the facts that matter for *this* question, not a
>   headnote.
> - Paraphrase the **holding / proposition** in your own words.
> - Classify **how-cited** relative to the target/leading case (for any citator-trail or
>   "who-cites-X" task): `followed | applied | cited-in-argument | mentioned-as-history |
>   distinguished` — at `[N]`. A history-only mention carries very different weight from an
>   application of the ratio; say which, and never present a history mention as endorsement.
> - For the **analogical** strategy, record **proximity**: factual high/med/low · doctrinal
>   high/med/low.
> - Record the **AGLC4 citation** (combined medium-neutral + authorised-report form, per
>   `references/aglc4-citation.md`), built only from the verified header metadata you fetched.
> - Set a **verification flag**: `verified | unverified (could not fetch) | suspect (mismatch)`.
>
> Then **CONVERGE**: dedupe the same case found by multiple strategies into one block, and produce
> **one ranked table** ordered by **authority × proximity × recency** (see Ranking, below).
>
> A clean negative is a complete answer: "no on-point authority after running these paths: <list>."
> Report it; do not invent authority or pad. Do **not** assess statutory currency or build the
> contrary-case brief — other agents own those (but record any strongly contrary case you trip over
> as a one-line flag for the red-team).
>
> Write your report to `agent-reports/cycle<N>-caselaw.md` using the **Case-law agent report
> schema** in `references/templates.md` §3.1. Populate every required field for every case. Then
> stop.

## The four strategies in detail

- **By-statute.** When the legislation agent (or the ledger) has named a controlling provision,
  search for decisions *on that provision*. With jurisd: `search_cases` with the provision and Act
  as a phrase, scoped to the relevant jurisdiction. Fallback: `web_search` for the provision plus
  "considered"/"applied" across the court hierarchy. Pull the leading case under the provision, then
  its applications. Common trap: returning cases that merely *mention* the Act rather than construe
  the provision — read enough to confirm the provision is actually in issue.
- **By-doctrine.** Search the doctrinal phrase (e.g. "procedural fairness" + "new trial"; "Anshun
  estoppel"). Find the **leading** statement first (usually the highest court), then work down to
  applications. Trap: stopping at a recent first-instance application and missing the appellate
  authority it depends on.
- **By-analogy / factual-matrix.** The highest-value strategy and the one a keyword search skips.
  Reduce the matrix to fact predicates (party type, transaction, procedural posture, the operative
  event), then search those. Rank hits by how closely the facts *and* the legal question line up.
  This is where the precedent most useful to the user's own facts is found.
- **By-citator-trail.** Forward: who has cited the leading case, and *how* (this is the how-cited
  classification). Backward: what the leading case itself relies on. The **citator recipe** in
  retrieval-mechanics.md is the forward tool — jurisd v0.5.0 has no live citator, and its offline
  `find_citing` is legislation-only, so it returns nothing for case citators; seed the recipe with
  `search_cases` on the case name and with **Jade's free-tier citation panel** (resolve the case
  via `site:jade.io "<citation>"`, open the article in Chrome, harvest the cases-citing panel —
  tier-2 leads, each confirmed on tier 1). Trap: treating every forward citation as support — most
  are history or argument mentions; classify each.

## Ranking method — authority × proximity × recency

Produce a **defensible ordering, not a numeric score**, combining three factors:

- **Authority** — position in the court hierarchy (see the hierarchy table in
  `references/retrieval-mechanics.md`): HCA > intermediate appellate (NSWCA/CCA, FCAFC, State CAs) >
  single-judge (NSWSC, FCA) > inferior courts/tribunals. Binding beats persuasive for the forum in
  the goal.
- **Proximity** — factual + doctrinal closeness to the framed question. A case directly on the
  provision and on facts like the matrix outranks a general statement of principle.
- **Recency** — later decisions are preferred *where they restate or refine* the principle; but a
  recent first-instance case does **not** outrank the leading appellate authority on point. Recency
  breaks ties and flags whether the law has moved, not the other way round.

Rule of thumb: **the leading binding authority on point ranks first even if it is older**; analogical
and applying cases follow, ordered by proximity then recency; downstream first-instance applications
and history-only mentions rank last.

## How-cited classification (why it is a required field)

A later judgment can mention an earlier case in very different ways, and the weight differs sharply:

- **followed** — the later court adopts the earlier ratio as governing and applies it.
- **applied** — the earlier case supplies the test the later court uses to decide.
- **cited-in-argument** — referred to by counsel or in the court's survey without being decisive.
- **mentioned-as-history** — a procedural/background reference only (e.g. a Full Court reciting the
  history of the same litigation), with no adoption of any proposition.
- **distinguished** — the later court confines the earlier case to its facts and declines to apply
  it.

The classification stops the orchestrator (and the user) from citing a case as *support* when the
later authority merely recited it as history or distinguished it. Tag the paragraph `[N]` where the
treatment occurs.

## Output

One markdown file, `agent-reports/cycle<N>-caselaw.md`, populated to the **Case-law agent report
schema** in `references/templates.md` §3.1 — one block per case (citation; court/date/bench; verified
source URL; factual summary; relevant paragraphs; minimum verbatim quote(s); paraphrased holding;
how-cited; proximity for analogical hits; AGLC4 citation; verification flag), plus a **Negative findings** section
listing the paths that returned nothing on point. Required fields are mandatory; a case missing one
is treated as unresolved by the orchestrator's convergence.

## Lane boundaries (do not cross)

- No statutory currency, amendment history, or subsection extraction — that is the legislation agent.
- No secondary-source framing — that is the secondary agent.
- No systematic contrary-authority sweep or distinguishing-argument brief — that is the red-team;
  only flag contrary cases you happen to find.
- No assertion of any citation, quote, holding, judge, or date from memory — fetch and verify, or
  mark it unverified.
deprecation-and-migrationaddyosmani/agent-skillsManages deprecation and migration. Use when removing old systems, APIs, or features. Use when migrating users from one implementation to another. Use when migrating a database schema in production, such as renaming or dropping a column without downtime (expand/contract). Use when deciding whether to maintain or sunset existing code.Stars: 119kUpdated Claude CodeCodexVerifiedInstall
performance-optimizationaddyosmani/agent-skillsOptimizes application performance across frontend, backend, queries, and databases. Use when performance requirements exist, when you suspect performance regressions, when Core Web Vitals or load times need improvement, when N+1 query patterns need fixing, or when profiling reveals bottlenecks.Stars: 119kUpdated Claude CodeCodexVerifiedInstall
agent-cost-reportthedotmack/claude-memBelievable agent cost report for any period, default the last 7 full days PT, not counting today. Measured tokens from Claude Code transcripts priced at OpenRouter list prices (ESTIMATED), measured provider spend when a sanctioned source exists, note-taker cost separate, Timing-style HTML/PDF plus report.json, line-items.csv, evidence.json.Stars: 118kUpdated Claude CodeCodexVerifiedInstall
agent-cost-reportthedotmack/claude-memBelievable agent cost report for any period, default the last 7 full days PT, not counting today. Measured tokens from Claude Code transcripts priced at OpenRouter list prices (ESTIMATED), measured provider spend when a sanctioned source exists, note-taker cost separate, Timing-style HTML/PDF plus report.json, line-items.csv, evidence.json.Stars: 118kUpdated Claude CodeCodexVerifiedInstall