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-validation.md

View on GitHub
# Validation Gate — Five Checks, Dispatch, Reconciliation

The citation-integrity gate runs **every cycle**, over every candidate authority that cycle
produced, before any of it reaches the cycle memo. It replicates the `submissions-verification`
skill's checks so this skill stays standalone.

**Dispatch rule (same as Step 3).** The five checks are dispatched as **five named Claude Code Task
subagents in a single turn** — so they run in genuinely parallel, isolated contexts and each is
separately visible. Name them explicitly; do not collapse two checks into one dispatch; if a check
fails or returns malformed output, re-dispatch only that check. The orchestrator then runs a sixth
**reconciliation** pass itself. (Single-threaded fallback — running the checks sequentially — applies
**only** where the Task tool is absent, e.g. claude.ai chat.)

Certainty states: **verified / suspect / overridden** — never "probably fine". An authority that
cannot be fetched cannot be verified; mark it suspect and say so.

## Common dispatch contract

Every check subagent receives: the **candidate-authority list** for this cycle (citations +
pinpoints + the propositions/quotes attributed to them, from the area-agents' reports); the
**temporal anchor**; a pointer to read `references/retrieval-mechanics.md` before any fetch (jurisd
Layer 0 first — `resolve_citation`, `fetch_document_text`; for treatment/citing-cases the citator
recipe (jurisd v0.5.0 has no live citator; Jade's free-tier citation panel is the best seed); else
the fallback stack) and `references/aglc4-citation.md` (the citation-format rules the existence
check tests against); its role prompt (below); and its output path
`agent-reports/cycle<N>-validation-<check>.md`.

Hard constraints in every check's prompt: work only your own check; ground every finding in a fresh
fetch performed during this dispatch (never from training-data memory); confirm on the highest source
tier reachable (see ranking); if an authority cannot be located after reasonable search, return
`not_located` with the attempts listed — do not guess; write strictly to the §3.7 schema.

## The five role prompts

**Check 1 — Existence** *(paste verbatim):*
> You are the **Existence check**. For every candidate authority, confirm it exists on a primary
> source in the form cited: party names (correct order), court, year, medium-neutral citation number,
> and authorised report volume/page all match. Use jurisd `resolve_citation` (or the fallback) to
> locate it by neutral cite, by report cite, and by party name — whichever resolves first. A Jade
> article (tier 2 — resolve via `site:jade.io "<citation>"`, open in Chrome) is acceptable evidence
> of existence, but anything destined for a filed document still needs tier-1 confirmation. Check
> the **AGLC4 citation string** against the fetched metadata per `references/aglc4-citation.md`: a
> formatting slip is corrected in your note; an element mismatch (wrong year, court, number, or
> report volume/page) is a `FAIL`. Record
> `PASS` (exists, form correct), `FAIL` (cannot locate after listed attempts, or the form is wrong —
> state the discrepancy), or `DISPUTED`. Do not check pinpoint, quote, or treatment. Write to
> `references/templates.md` §3.7.

**Check 2 — Pinpoint** *(paste verbatim):*
> You are the **Pinpoint check**. For every candidate authority with a paragraph/page pinpoint,
> retrieve that pinpoint **verbatim** from the source and compare it to the proposition attributed to
> it. Record `PASS` (pinpoint supports the proposition), `FAIL` (it does not — quote the retrieved
> text and say why), or `DISPUTED`. Where there is no pinpoint, note `not_applicable`. Do not check
> existence, quote fidelity, or treatment. Write to §3.7.

**Check 3 — Quote** *(paste verbatim):*
> You are the **Quote check**. For every candidate authority carrying text in quotation marks,
> retrieve the cited paragraph and compare the quoted text to the source **word-for-word**. Treat as
> discrepancies: changed/dropped/added words, ellipses that hide a qualification, capitalisation or
> punctuation that alters sense, and silent emphasis changes. Record `PASS` (verbatim), `FAIL`
> (substantive discrepancy — show source vs submission side by side), or `DISPUTED`. Pay particular
> attention to ellipses that elide a condition in the original. Skip entries with no quote. Write to
> §3.7.

**Check 4 — Treatment** *(paste verbatim):*
> You are the **Treatment check**. For every case, determine whether it remains good law as at the
> temporal anchor. Use the **citator recipe** in `references/retrieval-mechanics.md` (jurisd v0.5.0
> has no live citator; LawCite/SINO are Cloudflare-walled), optionally seeded by jurisd `search_cases`
> on the case name. Apply
> the **treatment taxonomy** below and assign Good Law / Reinforced / Qualified / Overruled. Where
> Qualified or Overruled, record the name, citation, and URL of the later decision and a 1–3 sentence
> note on what it held. Also record the **how-cited** classification (followed / applied /
> cited-in-argument / mentioned-as-history / distinguished). Record `PASS` (Good Law or Reinforced),
> `FAIL` (Overruled, or Qualified in a way that defeats the proposition relied on), or `DISPUTED`.
> Skip statutes — the currency check owns those. Write to §3.7.

**Check 5 — Red-team (currency)** *(paste verbatim):*
> You are the **Red-team currency check**. Adversarially test the cycle's authorities: (a) for every
> statutory citation, confirm via the register that the section exists, bears the content claimed,
> and the **version was in force as at the temporal anchor** (or the relevant past date) — state any
> version mismatch precisely; (b) flag any common-law authority cited as live where statute has since
> codified, modified, or displaced the principle; (c) flag mis-characterised holdings (obiter as
> ratio, plurality as majority, dissent as accepted). Record `PASS` / `FAIL` / `DISPUTED` per item
> with the issue stated. Write to §3.7.

## Treatment taxonomy (decision tree for Check 4)

- **Overruled** — a higher court, or the same court in a later case, has held it wrongly decided / no
  longer good law. → **Do not rely.** Mark suspect; tell the orchestrator to replace it.
- **Qualified / Distinguished / Doubted** — later authority limits it, distinguishes it on the
  facts/law, or expresses doubt. → Rely **with care**; record the limitation and the limiting case.
- **Reinforced** — later higher authority approves or follows it. → Strong; record the approving case.
- **Good Law** — no adverse subsequent treatment found after running the defined paths. → Stands (a
  verified-negative finding, not an absence of work).

A **history-only** citation (mentioned-as-history) does not make a case authority for a proposition —
never report it as endorsement.

## Source-authority ranking (every check)

Confirm on the highest tier reachable; anything to be **relied on in a filed document** must be
confirmed on a **tier-1 primary source** (AustLII / CaseLaw NSW / HCA site / legislation registers,
or a jurisd fetch of same). Tiers below primary — official-party host / authentic aggregator
(including free-tier Jade), chambers/law-society summary, general aggregator — may *surface* an
item (Jade is good enough for existence and text) but are not sufficient for court reliance. Record
the tier consulted for each authority.

## Reconciliation (orchestrator's sixth pass)

A **dispute** is any disagreement between the checks on existence, pinpoint match, quote fidelity, or
treatment, **or** any single `FAIL` on a substantive question, **or** any Red-team flag the other
checks did not address. For each disputed item the orchestrator runs an **independent re-fetch on a
primary source** (it does not vote — the checks are input, not ballots) and records the resolution in
the **Reconciliation Log** in the cycle memo: what each check found, where they diverged, and how it
was settled.

## Outcome feeding the memo

- **Verified** — passes all applicable checks on a tier-1 source; treatment ≠ Overruled. Reported as
  good law with its treatment + how-cited.
- **Suspect** — fails a check, cannot be fetched, or is Overruled. Flagged; **not** presented as good
  law; the orchestrator decides whether to replace it (may spawn a follow-up case-law task).
- **Overridden** — only if the user explicitly overrides (e.g. a decision too recent to be indexed);
  recorded with the user's stated reason.
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