Reference · plugin fact sheet

`commercial-legal`

The ground-truth summary behind Module 3. Every line below was read from the plugin's actual files (version 1.0.2, checked 2026-06-22) — file paths cite the live repo.

Identity

From .claude-plugin/plugin.json: “Reviews vendor agreements, NDAs, and SaaS subscriptions against your sales-side or purchasing-side playbook, tracks renewals and cancel-by deadlines before they're missed, routes escalations to the right approver, and translates reviews into summaries business stakeholders will actually read.”

The 12 skills

All in skills/<name>/SKILL.md. Three are reference skills (user-invocable: false) — loaded by the front door, not called directly.

Skill	Role	What it does
`review`	The front door	Reads the document's titles (main agreement + exhibits) to identify what it is, routes to the right reference skill(s), confirms the routing with you (“Sound right?”), and integrates everything into one memo.
`vendor-agreement-review`	reference	Inbound vendor MSA / services agreement vs. your playbook: deviations, risk, specific redline language, named approver.
`nda-review`	reference	Fast triage of inbound NDAs into GREEN / YELLOW / RED so lawyer time goes only where needed.
`saas-msa-review`	reference	Subscription-specific overlay: auto-renewal mechanics, price escalation, data portability, uptime SLAs, subprocessor rights.
`cold-start-interview`	setup	The first-run interview that writes your practice profile (playbook, escalation matrix, house style).
`customize`	setup	Change one profile thing without re-running the whole interview.
`escalation-flagger`	routing	Routes an issue to the right approver per your escalation matrix and drafts the ask.
`renewal-tracker`	deadlines	Cancel-by deadlines from a maintained renewal register; warns before notice windows close.
`stakeholder-summary`	output	Translates a review into the two-minute “can I sign this?” answer a business owner will actually read.
`amendment-history`	analysis	Traces how a contract changed across its base agreement and amendments; provision-level evolution.
`review-proposals`	maintenance	Approve / reject the playbook-update proposals the `playbook-monitor` agent surfaces.
`matter-workspace`	firm-side	Matter management for multi-client practices (in-house users never see it).

The 3 subagents — a playbook-maintenance loop

From agents/*.md. Together they keep the playbook honest:

deal-debrief — weekly (Monday); surfaces recently signed agreements containing playbook deviations and prompts you to log the context while memory is fresh.
playbook-monitor — data-triggered; when the same clause position has been deviated from ~5 times in 12 months, it proposes a playbook update (you approve via review-proposals).
renewal-watcher — weekly; reads the renewal register and posts what's coming up. Its tool list is deliberately narrow: ["Read", "Write", "mcp__ironclad__*", "mcp__*__slack_send_message"].

The 7 bundled connectors

From .mcp.json: Ironclad (CLM), DocuSign (e-signature), iManage (DMS), Definely (contract structure), TopCounsel (outside-counsel network), Slack, Google Drive. Note: no research connector is bundled — add one yourself (Module 1, step 5) or citations are flagged unverified.

Files it keeps on your machine

~/.claude/plugins/config/claude-for-legal/commercial-legal/CLAUDE.md — your practice profile: who you are, integrations, the sales-side and purchasing-side playbooks (liability cap as four positions, indemnity, data protection, term, governing law, “the one thing”), the escalation matrix, house style.
…/matters/<slug>/ — one folder per matter (a slug is the matter's short, hyphenated folder name, e.g. acme-msa-2026; firm-side only).
…/verification-log.md — one line per verified citation/fact, so the next reviewer doesn't re-verify.

Output conventions (what a review will look like)

Work-product header — role-aware: lawyers get PRIVILEGED & CONFIDENTIAL — ATTORNEY WORK PRODUCT…; non-lawyers get RESEARCH NOTES — NOT LEGAL ADVICE…. The profile itself warns the header is a label, not a control, and that “work product” is a US doctrine.
⚠️ Reviewer note — one block above the deliverable: sources, what was actually read, items flagged [review], and “before relying.”
Dual severity — every finding gets legal risk (🔴🟠🟡🟢 can we be sued?) and business friction (does this cost revenue, trust, or time?).
Source tags — provenance, not confidence: [CourtListener] only if the tool actually returned it; otherwise [model knowledge — verify].
Decision tree — closes with options (draft the redline / escalate / get more facts / watch and wait), not a decision. Plus “one question I'd ask that isn't in my checklist.”

Source for this section: the plugin's own CLAUDE.md (shared-guardrails + Outputs sections) — shipped inside the plugin and readable on GitHub.

← Back to Module 3