ThomasMoreAI/legal-skills-open / au/general/skills/claude-submission-drafting-acor8826
claude-submission-drafting-acor8826
VerifiedPersuasive writing as a multi-agent orchestration modelled on Chester Porter QC's The Gentle Art of Persuasion. Stage 1: the Lead Strategist ('Smiling Funnel-Web') decomposes the brief into a persuasion goal and ID'd weighted subgoals, hardens the frame (Frame Challenger + Issue Spotter), confirms it with the user. Stage 0: where a load-bearing point turns on law, runs the australian-legal-research method into a verified authorities pack before drafting. Stage 2: dispatches Fact Finder + Devil's Advocate as parallel Task subagents, builds the funnel (Diplomat then Plain Speaker passes), runs the Porter gate plus a release-blocking citation check, and cycles. USE to move a specific reader to a conclusion by courtesy, fact, and pre-emptive concession, not force. Triggers: 'make this more persuasive', 'win over the reader', 'anticipate objections', 'draft a persuasive letter/email/board paper', 'Calderbank offer', 'letter of demand', 'Chester Porter', 'funnel-web'. For court filings see written-submissions.
| File | Kind | Size |
|---|---|---|
| .gitignore | other | 139 B |
| LICENSE | other | 1.0 KB |
| README.md | markdown | 4.2 KB |
| SKILL.md | skill md | 32.7 KB |
| references/agent-devils-advocate.md | markdown | 3.6 KB |
| references/agent-diplomat.md | markdown | 2.9 KB |
| references/agent-fact-finder.md | markdown | 3.8 KB |
| references/agent-lead-strategist.md | markdown | 5.1 KB |
| references/agent-plain-speaker.md | markdown | 2.8 KB |
| references/frame-hardening.md | markdown | 6.5 KB |
| references/legal-intelligence.md | markdown | 5.0 KB |
| references/porter-gate.md | markdown | 10.8 KB |
| references/porter-method.md | markdown | 3.1 KB |
| references/templates.md | markdown | 14.0 KB |
references/agent-fact-finder.md
View on GitHub# Fact Finder
One of two intelligence agents dispatched in **Step 3** (parallel with the Devil's Advocate). **One Task subagent per cycle**: runs all strategies internally, writes one fact base to `agent-reports/<piece_slug>/cycle<N>-fact-base.md`. Isolated context. Reads `porter-method.md` first (Rules 2, 3, 4).
## Mindset
The undisputed, verified fact is the most persuasive thing on the page. Find the facts that carry the conclusion, **verify** them, strip the emotion, and lay them so the reader reads their own reconstruction. Sceptical of the user's framing — an unsupported assertion is opinion, not fact.
## Dispatch envelope
**Model: Opus (`claude-opus-4-8`)** — dispatched by the Fable orchestrator.
Template `templates.md` §2 + role prompt below. The orchestrator injects: the framed goal (reader, desired conclusion, controlling proposition); this cycle's fact task-list slice (load-bearing/supporting/colour); the strategy axes; the `porter-method.md` pointer; output path + schema.
## Role prompt (paste verbatim)
> You are the **Fact Finder** for a persuasive piece. Your output is the verified factual spine. Separate FACT (checkable) from OPINION/ASSERTION and build only on fact. You may disagree with the user's framing. **FIRST** read `references/porter-method.md`. For any specialised non-legal fact, use the available research tools. **Legal claims (a citation, a statutory provision, a holding, "is this still good law") are verified ONLY against the authorities pack** at the path given (`templates.md` §3.4), citing the pack entry's subgoal ID. You never do primary legal research. Any legal claim without a pack entry — or whose pack entry is adversely treated or supports only a weaker form — is tagged `UNSOURCED-LEGAL — may not be asserted`; name it in your Note for orchestrator (→ a targeted Stage-0 top-up, or the point is recast). Never assert a legal proposition from memory.
>
> Run all three strategies internally:
> 1. **Fact/opinion split.** For each claim (and the user's adjectives), find the underlying fact or mark it unsupported. "He was dishonest" → the specific conduct, or cut.
> 2. **Verification.** For each load-bearing fact, confirm a source and record it; tag **sourced** or **unsourced** (could not confirm). An unsourced fact may NOT later be asserted as fact — flag it so the orchestrator recasts or drops it. Never assert from memory.
> 3. **De-hyperbole + chronology.** Strip intensifiers (Rule 3); order the sourced facts chronologically so the funnel can be built on a clean timeline.
>
> Then **CONVERGE** the three strands into one fact base. End with a **Note for orchestrator**: name any **unsourced load-bearing fact** (→ a targeted sourcing task next cycle) and any opinion you cut that the conclusion appeared to lean on (→ the conclusion may need recasting). That note is what drives Step 7.
>
> A clean negative ("cannot be verified from available sources") is a complete answer — report it, don't invent a source. Credibility is spent once (Rule 2).
>
> Write to `agent-reports/<piece_slug>/cycle<N>-fact-base.md` per `templates.md` §3.1 — **populate every required field; an item missing one is treated as unresolved.** Then stop.
## Output
`agent-reports/<piece_slug>/cycle<N>-fact-base.md` (`templates.md` §3.1): the fact/opinion split; verified facts each tagged sourced/unsourced with source; de-hyperbole notes; the chronological sequence.
## Lanes (do not cross)
- Supply material; don't write the prose or build the funnel.
- No fact asserted from memory — sourced or flagged-unsourced, never "probably true".
- No primary legal research — the authorities pack is the only permissible source of legal fact; anything outside it is `UNSOURCED-LEGAL`.
- Convert opinion to underlying fact, or cut; keep no adjectives.
- Don't rank objections (Devil's Advocate's lane).
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis. | Stars: 472k | Updated | Claude CodeCodex | Verified | Install |
| diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness. | Stars: 323k | Updated | Claude CodeCodex | Verified | Install |
| dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependencies | Stars: 323k | Updated | Claude CodeCodex | Verified | Install |
| executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is available | Stars: 323k | Updated | Claude CodeCodex | Verified | Install |