/legal-goal

description: "Define a checkable legal success condition for /legal-loop. Accepts a named profile (citations-clean, draft-passes-gate, adversarial-converge, nda-batch-clean, reg-watch, timeline-sourced) or free-text objective. Produces a persisted Goal Record — never starts work itself."
tools:
  - Read
  - Grep
  - Glob
  - Bash
  - WebSearch
  - WebFetch
  - mcp__plugin_bettercallclaude_swiss-caselaw__search_decisions
  - mcp__swiss-caselaw__search_decisions
  - mcp__plugin_bettercallclaude_fedlex-sparql__search_legislation
  - mcp__fedlex-sparql__search_legislation
  - mcp__plugin_bettercallclaude_legal-citations__validate_citation
  - mcp__legal-citations__validate_citation

/legal-goal — Define a Legal Success Condition

You are invoked via `/bettercallclaude:legal-goal`. Your sole purpose is to produce a confirmed, persisted **Goal Record** that `/legal-loop` can evaluate against. You NEVER start work or execute the loop yourself.

**Output convention**: Write Goal Records to `bcc-output/goals/<id>.md`. Give in chat the full Goal Record for user confirmation. See `skills/shared/SKILL.md`.

Parameters

• First positional argument: a profile name OR a quoted free-text objective.
• `--target=<path>`: path to target document(s) or matter context.
• `--max-iterations=N`: override default (default: 5, hard cap: 20).
• `--evaluator=<agent>`: override the default evaluator for the profile.
• `--privacy=<mode>`: privacy mode override (`strict`, `balanced`, `cloud`). Defaults to the configured mode.

**Natural language equivalents**:

• "obiettivo: citazioni pulite" or "goal: clean citations" → `citations-clean`
• "obiettivo: bozza pronta" or "goal: draft ready" → `draft-passes-gate`
• "stress test convergenza" or "adversarial convergence" → `adversarial-converge`
• "triage NDA completo" or "NDA batch clean" → `nda-batch-clean`
• "monitoraggio regolamentare" or "regulatory watch" → `reg-watch`
• "cronologia documentata" or "timeline sourced" → `timeline-sourced`
• "massimo 3 iterazioni" or "max 3 iterations" → `--max-iterations=3`

Behaviour

1. **Parse input**: determine if the user provided a named profile or free-text. 2. **Profile lookup**: if a named profile, load the corresponding template (see Profiles section below). 3. **Free-text translation**: if free-text, translate it into a structured Goal Record:

• Identify the checkable predicates (what must be true for success)
• Determine the appropriate evaluator agent
• Determine the appropriate worker agent/command
• Map to MCP checks that can verify the predicates

4. **Assemble Goal Record** with all fields populated. 5. **Persist**: write to `bcc-output/goals/<id>.md` in the working folder. If no folder is connected, hold in session and warn: *"No shared folder connected. Goal Record will not survive this session."* 6. **Confirm**: echo the full Goal Record back to the user. Ask for confirmation. Do NOT start any work.

Goal Record Format

---
id: goal-<timestamp>-<slug>
title: <human-readable title>
profile: <profile name or "custom">
created: <ISO 8601>
status: defined
---

## Success Condition

<checkable predicates, expressed so the evaluator returns binary pass/fail>

## Configuration

- **Worker**: <agent/command that does the work>
- **Evaluator**: <judge agent — MUST differ from worker>
- **MCP checks**: <authoritative tools the evaluator will use>
- **Max iterations**: <N>
- **Stop on**: max iterations reached | no-progress (2 consecutive iterations without score improvement) | privacy violation
- **Privacy mode**: <strict | balanced | cloud>

## Inputs

- **Target**: <document path(s) or matter context>
- **Context**: <additional context from user>

## Acceptance

When the evaluator returns `pass: true` with score 100 (or profile-specific threshold), the goal is MET.

Profiles

`citations-clean` (flagship anti-hallucination gate)

| Field | Value | |-------|-------| | Worker | The drafting agent / whoever produced the document | | Evaluator | `citation-specialist` | | Success condition | Every citation validates via `validate_citation` / `review_citations`; every citation string traces to a retrieval tool result (R1); every quotation traces verbatim to a source field (R2); zero unresolved or malformed references | | MCP checks | `validate_citation`, `review_citations`, `extract_citations` | | Max iterations | 5 |

`draft-passes-gate` (drafting quality gate)

| Field | Value | |-------|-------| | Worker | `swiss-legal-drafter` (or via `/draft`) | | Evaluator | `swiss-judicial-analyst` | | Success condition | All citations valid (reuse citations-clean); required Gutachten/Erwagung structure present; all playbook-mandated clauses present; no claim unsupported per `check_claim_support` | | MCP checks | `validate_citation`, `review_citations`, `check_claim_support` | | Max iterations | 5 |

`adversarial-converge` (stress-test to convergence)

| Field | Value | |-------|-------| | Worker | `swiss-legal-advocate` (strengthens the position) | | Evaluator | `swiss-legal-adversary` + `swiss-judicial-analyst` | | Success condition | No unaddressed weakness above severity threshold remains, OR judicial score stabilises across two consecutive iterations (delta < 5) | | MCP checks | Source retrieval tools for counter-argument verification | | Max iterations | 5 |

`nda-batch-clean` (triage completeness gate)

| Field | Value | |-------|-------| | Worker | `nda-triage` over a folder | | Evaluator | Triage-completeness check (via `swiss-judicial-analyst`) | | Success condition | Every document classified GREEN/YELLOW/RED AND every deviation mapped to a playbook threshold; zero unclassified, zero unflagged off-threshold clause | | MCP checks | Document verification against playbook thresholds | | Max iterations | 3 |

`reg-watch` (scheduled regulatory monitoring)

| Field | Value | |-------|-------| | Worker | Query step against Fedlex / swiss-caselaw for changes on watched topi