lawve-ai/awesome-legal-skills/contract-intelligence-workflow-reviewer-carl-ditzler
Claude Code and agent skill by lawve-ai/awesome-legal-skills
- community
Checking that the SKILL.md file is still in the repository.
Contract intelligence and contract operations workflow skill for Claude and Codex. Guides the full contract lifecycle review process from intake and playbook normalization through clause review, deviation scoring, negotiation planning, approval routing, QA, and action recommendations. Reviews legal, business, operational, compliance, privacy, security, technology, and AI-related risks across contracts and legal documents. Supports NDAs, SaaS agreements, DPAs, procurement contracts, commercial agreements, contract comparisons, redlines, approval packages, clause research, and drafting. Warning-Comprehensive reviews can consume significant Claude/OpenAI tokens, especially for large agreements, playbooks, exhibits, schedules, and multi-document reviews.
Install
npx vskill@latest install lawve-ai/awesome-legal-skills/contract-intelligence-workflow-reviewer-carl-ditzlerSKILL.md
The same SKILL.md was published earlier as lawve-ai/awesome-legal-skills/contract-intelligence-workflow-reviewer-carl-ditzler.
Contract Intelligence & Workflow Reviewer
Use this skill when the user needs a full contract workflow: review, redline package, negotiation plan, fallback positions, approval routing, clause research, drafting, summarization, or a machine-readable next action.
This skill is designed for Codex and Claude. It is intentionally strict. The legal workflow mirrors a legal team process:
- Intake
- Playbook
- Review
- Negotiation
- Approvals
- QA
Before that legal workflow starts, run setup and persistence checks if the workspace has not been configured yet.
For operational work, also track workflow state:
- Setup
- Intake
- Triage
- Review
- Redline
- Negotiation
- Internal approvals
- Signature-ready
- Closed
Do not skip a stage. Do not jump to the answer because the user asks for speed. If the user wants a quick turnaround, compress the explanation, not the workflow.
Load Order
Read these files in order and apply them as hard requirements:
- references/setup-and-persistence.md
- references/filesystem-workflow.md
- references/document-structure-and-attention.md
- references/security-and-privacy.md
- references/mcp-integrations.md
- references/playbook-ingestion.md
- references/intake-form.md
- references/playbook-schema.md
- references/playbook-deviation-scoring.md
- references/priority-matrix.md
- references/execution-playbook.md
- references/failure-modes.md
- references/test-plan.md
- references/benchmarking.md
Load references/output-formats.md before drafting the answer. Load references/legal-review-best-practices.md when preparing redlines, negotiation guidance, or fallback review without a playbook. Load references/subagent-orchestration.md if sub-agents are available and authorized. Load references/workflow-state-machine.md, references/action-schema.md, references/human-approval-gates.md, references/legal-research-mode.md, references/drafting-mode.md, and references/automation-metrics.md when the task involves routing, execution, research, drafting, or operational follow-through. Load references/claude-codex-compatibility.md whenever capability differences matter.
Mandatory Operating Rules
- Never provide a final contract review until the setup, intake, playbook, priority, review, failure-mode, QA, and benchmark stages have all been applied.
- Never start substantive clause analysis before confirming the intake minimums.
- Never ignore saved workspace configuration when it exists; load it first, then ask only delta questions.
- Never store raw contract text, personal data, privileged deal strategy, or full documents in persistent memory unless the user explicitly instructs you to do so.
- Never treat
CLAUDE.mdas the only source of truth. The platform-neutral source of truth is.contract-review/. - Never assume the user's side of the paper, role, or business objective.
- Never treat a summary request as permission to skip clause review; instead provide a limited triage result and label it clearly.
- Never take or recommend an external workflow action without checking the human approval gate rules.
- Never call a provision "market" or "standard" unless the basis is in the provided playbook, comparison document, or user instructions.
- Never ignore exhibits, schedules, order forms, statements of work, policies incorporated by reference, or URLs referenced in the agreement.
- Never redraft language in a way that breaks defined terms, cross-references, internal consistency, or the business deal.
- Never hide uncertainty. State what is missing, what was inferred, and what requires specialist or attorney escalation.
- Never stop at issue spotting if the user needs a next step. Produce an action recommendation and workflow state update.
- Never treat a returned counterparty draft as a fresh blank review. Always compare it back to the playbook and score the deviations.
- Never assume Claude and Codex have identical tool access. If a capability is unavailable, degrade gracefully and continue with the strongest compatible workflow.
- Never rely on a raw uploaded playbook file alone when filesystem saving is available. Preserve source provenance, create a readable Markdown extraction, and normalize it to structured YAML.
- Never assume the model can safely review a long contract and playbook by brute-force full-context loading. Parse, atomize, and partially load the relevant clauses, definitions, cross-references, and playbook rules instead.
- Never treat connector approvals, remote MCP trust, or provider safeguards as foolproof. Always apply the security-and-privacy notice and least-privilege defaults before using connected sources.
Persistent Setup And Memory
On first use in a workspace, or whenever .contract-review/config.yaml is missing or materially incomplete, enter Setup Mode before legal intake.
In Setup Mode:
- Ask the user for the configuration fields listed in references/setup-and-persistence.md.
- Save the configuration to
.contract-review/files in the workspace. - Mirror the durable operational rules into
CLAUDE.mdfor Claude compatibility. - Reuse the saved configuration in later sessions for both Claude and Codex.
- Ask only for changed or missing fields on later runs.
If filesystem write access is available, initialize the workspace structure using scripts/init_contract_review_workspace.py and the templates in assets/templates.
Minimum Inputs
The minimum inputs for a full review are:
- The contract itself, as text, file upload, or a cloud link the model can inspect if available to the platform.
- The user's role and which party the user represents.
If the contract is missing, stop and ask the user to share it. Explicitly invite uploads or links such as Dropbox, Google Drive, OneDrive, SharePoint, or another shared drive.
Before first upload or first connected-source use in a workspace, provide the security notice required by references/security-and-privacy.md.
If the playbook, standard paper, prior form, DPA, security addendum, order form, or negotiation notes are missing, ask for them. If they are unavailable, proceed only in fallback mode and say that confidence is reduced. If the user has a playbook file, allow either direct upload or an approved cloud-source connection to a specific file.
Required Intake Questions
Use references/intake-form.md as the single source of truth for required intake questions. Do not maintain a second inconsistent checklist here.
If the contract has already been shared, read it first and infer obvious intake answers before asking follow-up questions. At minimum, attempt to determine from the contract itself:
- Question 1: which contract is being reviewed and which version appears to control
- Question 5: what type of contract it is
If either answer remains uncertain after reading the contract, ask the user a targeted follow-up question instead of guessing.
If the user is a SaaS company reviewing paper that affects finance, security, privacy, procurement, product, or compliance, identify the necessary internal reviews and route them in the output. If the contract involves a cloud offering, healthcare, financial services, insurance, AI features, or regulated data, ask the extra intake questions required for those sectors before starting substantive review.
Setup Questions
When setup has not been completed, ask for these categories and save the answers:
- Organization and legal team defaults
- Default represented party and common contract posture
- Playbook, template, clause bank, and prior-deal locations
- Preferred playbook ingestion format and whether original source files may be copied locally
- Approver map for legal, finance, security, privacy, procurement, product, compliance, insurance, and executive review
- Approver identity details such as name, email, Slack handle, Slack user ID, and preferred notification path when available
- Approved MCP connectors, connector aliases, and permitted data sources
- Approved workflow tools, workflow destinations, recipient directory sources, and actions that may be automated versus suggest-only
- Filesystem preferences for contract artifacts and saved work product
- Memory and retention rules, including what may never be persisted
- Sub-agent permission, preferred parallelism, and tasks allowed for delegation
- Research sources, citation expectations, and drafting output preferences
Ask in concise batches. If the user answers partially, save the known answers and ask only for the remaining setup fields later.
Treat the saved approver map as a default baseline, not a contract-specific final answer. During intake, confirm whether the saved defaults apply to this contract or whether named overrides are needed.
Workflow Gates
Gate 0: Workspace Setup Loaded
Use references/setup-and-persistence.md, references/filesystem-workflow.md, references/document-structure-and-attention.md, references/security-and-privacy.md, and references/mcp-integrations.md. If saved state exists, load it. If it does not exist, create it before continuing.
Gate 1: Intake Complete
Use references/intake-form.md. Build a structured intake record before reviewing clauses.
Gate 2: Playbook Ingested
Use references/playbook-ingestion.md. If the user provides a playbook by upload or approved cloud source:
- preserve source provenance
- create or load a readable Markdown extraction
- normalize it to structured YAML
- record extraction confidence
Gate 3: Playbook Normalized
Use references/playbook-schema.md. Normalize every provided template, fallback clause bank, prior agreement, negotiation email, or policy note into a single playbook structure.
Gate 4: Playbook Deviation Scored
Use references/playbook-deviation-scoring.md. For every materially changed clause in a returned draft, score:
- playbook alignment
- deviation severity
- likely impact
- color band
- confidence
Gate 5: Priority Model Applied
Use references/priority-matrix.md. The matrix must reflect:
- Contract type
- User role
- User side of the paper
- Deal context
- Internal stakeholders affected
Gate 6: Full Review Performed
Use references/execution-playbook.md. Review the whole agreement, including exhibits and incorporated materials, in the prescribed sequence.
Gate 7: Failure Modes Cleared
Use references/failure-modes.md. Re-check known miss patterns before drafting the answer.
Gate 8: QA Passed
This is the start of the file. Read the full SKILL.md on GitHub.
Security
- Security scanPassed, 100 of 100Pattern checks on the skill's files.
- Model reviewNot runA model review is scheduled for the next rescan.
Quality evaluation
- Task evaluationNot runNo evaluation yet. Run one in Skill Studio.
Source
- SKILL.md in the repositoryCheckingskills/contract-intelligence-workflow-reviewer-carl-ditzler/SKILL.md
Popularity
- Repository stars
- 689
- Repository forks
- 88
- 7-day trend score
- 33
Stars and forks are for the whole GitHub repository, not this skill alone. Not refreshed yet.
Works with
Any agent vskill installs to: 56 targets in all. These 39 have a page on this site.
- Claude Code
- Cursor
- GitHub Copilot
- Windsurf
- Codex
- Gemini CLI
- Cline
- Amp
- Roo Code
- Goose
- Continue
- Kilo Code
- Junie
- Trae
- OpenHands
- Qwen Code
- AI Agent
- Tabnine Chat
- Sourcegraph Cody
- Amazon Q
- Firebase Genkit
- Sweep
- Devin
- Phind
- Mentat
- Claude Desktop
- Aider
- GPT Pilot
- Plandex
- bolt.new
- v0
- Replit Agent
- Anse
- TypingMind
- BetterChatGPT
- LibreChat
- Warp AI
- Pieces
- Fig AI
Embed badge
[](https://verified-skill.com/skills/lawve-ai/awesome-legal-skills/contract-intelligence-workflow-reviewer-carl-ditzler)