contract-review: Skill for Claude Code

pipeline-state Skill

Manage pipeline state persistence for resume support and round-level diffs.

Capabilities

1. Save State (scripts/save-state.py)

   • Writes/updates pipeline-state.json after each step completion
   • Usage: python3 save-state.py '<json_params>'
   • Parameters: state_path, pipeline, matter_id, round_num, step, step_name, status, output, review_mode
   • Optional v2 parameters: session_id, validation, metrics, error
   • Reads v1 state and writes back v2-compatible state without changing the file name

2. Load State (scripts/load-state.py)

   • Reads state and determines resume point
   • Migrates v1 state in memory for resume decisions
   • Usage: python3 load-state.py <state_path>
   • Exit code 2 = resumable pipeline found

3. Round Diff (scripts/diff-rounds.py)

   • Compares clause records between two negotiation rounds
   • Classifies: unchanged, modified, added, removed
   • Usage: python3 diff-rounds.py <current_clauses_dir> <prior_clauses_dir> <output.json>

When to Use

• After every pipeline step completion: save state
• Before starting any pipeline: check for existing state (resume detection)
• WF4 Step 3: diff between negotiation rounds

Pipeline State Schema

{
  "schema_version": 2,
  "pipeline": "review",
  "matter_id": "deal-2026-001",
  "round": 1,
  "session_id": "review-deal-2026-001-round-1-20260306T100000Z",
  "last_completed_step": 7,
  "review_mode": "moderate",
  "step_artifacts": {
    "step_1": {
      "name": "Target document normalization",
      "status": "completed",
      "output": "working/normalized/clean.md",
      "completed_at": "2026-03-06T10:00:00Z",
      "validation": {
        "artifact_exists": true
      }
    }
  },
  "metrics": {
    "reference_full_load_count": 0,
    "reference_tokens_estimated": 1200,
    "retrieval_tokens_estimated": 900,
    "hydrated_candidate_count": 3
  },
  "started_at": "2026-03-06T10:00:00Z",
  "updated_at": "2026-03-06T10:45:00Z"
}