What does this skill do, and when should you use it?
This skill drives the contracts engine’s local stdio MCP server to validate runtime behavior after changes. It covers type checking, JSON-RPC tool calls, SQLite schema behavior, concurrency, and document-processing flows. The workflow depends on Node.js, Bun, SQLite, shell access, and the local filesystem. It is intended for developers maintaining the contracts skill and its documents server within the Anthropic healthcare plugin.
Runs bun run check for type checking; invokes server tools through the Node.js CLI or stdio JSON-RPC; tests corpus preparation, writes, citation lookup, rejection of invalid citations, and coverage using scratch data; checks SQLite schema versions, views, and triggers; races at least 16 server processes across fresh and warm databases for concurrency changes; validates tool schemas against JSON Schema draft 2020-12; runs sweep.mjs --limit 2; and checks that parallel worker time windows overlap as expected.
- A developer maintaining the contracts MCP server needs end-to-end regression checks after an engine change.
- An engineer changing tool input schemas needs to catch invalid schemas that can terminate an agent before tool execution.
- A maintainer changing SQLite schema or triggers needs to verify compatibility with existing databases.
- An engineer investigating concurrent writes or worker scheduling needs fresh-database and warm-database race tests.
- A contributor revising verification prose or protocol behavior needs to confirm tool names, paths, and required calls remain consistent.
How do you install this skill?
- Resolve and verify the actual target of ../../servers/documents from the assessed SKILL.md directory before running commands; the documented path may be invalid there.
- Obtain explicit confirmation for database deletion and concurrency stress testing, isolate healthcare documents in a recoverable scratch directory, and define the data boundary first.
- Do not treat repository-level README information or official publisher identity as evidence that this skill has been execution-tested.
- Shell / CLI
- Local filesystem
- MCP Server
Node.jsBunSQLite
Install the healthcare plugin collection with /plugin marketplace add anthropics/healthcare, followed by /plugin install healthcare@healthcare. This skill is located at plugins/healthcare/skills/contracts/.claude/skills/verify/SKILL.md; the source does not document a separate installation procedure.
tmp="$(mktemp -d)"
git clone --depth 1 https://github.com/anthropics/healthcare.git "$tmp"
mkdir -p ~/.claude/skills
cp -R "$tmp/plugins/healthcare/skills/contracts/.claude/skills/verify" ~/.claude/skills/
rm -rf "$tmp"Generated from the source repository and skill path; it copies only this skill's folder. If the author's install steps above differ, follow those first. To scope it to one project, replace ~/.claude/skills with that project's .claude/skills.
How do you use this skill?
In a Claude Code environment with the healthcare plugin installed, request verification of contracts-engine changes. The documented commands include cd ../../servers/documents && bun run check and node ../../servers/documents/src/index.mjs <tool> '<json-args>' for a single CLI tool call; a bare invocation runs MCP stdio mode. For protocol smoke tests, send initialize, notifications/initialized, and then tools/list or tools/call as sequential JSON-RPC lines. Point throwaway tests at a scratch parent directory through CLAUDE_HEALTHCARE_DATA. The source does not fully specify the skill’s trigger syntax.
What are this skill's strengths and limitations?
- Covers type checking, protocol behavior, data integrity, schema changes, and concurrency.
- Explicitly directs throwaway runs to scratch data, reducing risk to the live database.
- Targets JSON Schema draft 2020-12 and real agent startup behavior, not only plain tool listing.
- The server source is the shipped artifact and requires no build step.
- It verifies the contracts engine rather than providing contract analysis or contract question answering itself.
- It depends on the healthcare repository’s relative layout, Node.js, Bun, and local SQLite data.
- Concurrency verification is expensive, requiring at least 16 servers and at least 60 trials for each database state.
- The source does not document complete test-suite coverage or a direct invocation method for non-Claude-Code clients.
How does this skill compare with similar options?
Side by side with related skills; every score comes from the same FSRS standard.
| Skill | FS score | Stars | Last updated | License |
|---|---|---|---|---|
| Contracts Engine Verifier this page ✓ Anthropic · Official | 44 · Not recommended | ★ 425 | 1mo ago | — |
| GitHub Copilot SDK Builder ✓ Microsoft · Official | 49 · Use with care | ★ 2.7k | 3mo ago | MIT |
| MCP Integration for Claude Plugins ✓ Anthropic · Official | 51 · Use with care | ★ 38k | 11d ago | Apache-2.0 |
| Claude Code MCP Integration ✓ Anthropic · Official | 49 · Use with care | ★ 151k | 11d ago | — |
| e2e Playground Verification Skill | 59 · Recommended | ★ 10k | 11d ago | Apache-2.0 |
How did FollowSkills review this skill?
The skill explicitly recommends scratch data directories, avoiding the live database, and read-only verification, making scope and some data-flow risks visible. However, it does not require user confirmation, suggests database deletion and high-concurrency process spawning, and lacks rollback guidance, sensitive healthcare-data handling, and dependency-security guidance, so points are deducted.
It covers type checking, CLI, MCP smoke tests, concurrency, schema validation, and abnormal schema behavior, with some expected failure symptoms. Static review cannot establish runnability; the relative path ../../servers/documents may not resolve from the assessed nested skill directory, and complete failure diagnostics or reproducible test evidence are absent. The score is also limited by the static calibration cap.
The target scenario—end-to-end verification of contracts-engine changes—is clear, with guidance for ordinary, schema, concurrency, and sweep changes. Inputs, output formats, non-fit boundaries, and user roles are not clearly declared, there is no Chinese-language guidance, and the workflow depends on Bun, Node, Claude CLI, and local paths with limited environment-fit evidence, so points are deducted.
The document is readable and progressively organized around type checking, data, CLI, schema, concurrency, MCP, and sweep verification, with concrete examples. It lacks skill-level versioning, changelog, maintenance ownership, update path, explicit license metadata, FAQ coverage, and systematic limitation disclosure; the README only states Anthropic terms rather than a clear standard license, so points are deducted.
The skill supplies a fairly concrete verification workflow covering the main tool chain, database isolation, schema checks, and concurrency checks, so it could complete much of the intended task. The path concern, lack of representative output samples and automated acceptance criteria, and absence of execution evidence limit confidence; therefore only a moderate static score is assigned.
It names commands, a test file, SQL queries, and expected observations, providing some auditability. The supplied material contains no actual CI results, test output, independent reproduction record, or corroborating evidence, and no commands were executed in this review. The score is therefore limited by static calibration and further reduced for thin evidence.
Open a dimension to read why it scored that way
Evidence confidence：Low — Mostly static review, author material or a limited demo; useful for discovery, not high-risk decisions.
See the full review method →