Contract Evidence Analyzer
Search local contract files, compare amendments, and produce answers backed by verified citations.
The skill clearly restricts tool access, treats contract text as data rather than instructions, requires local scoping and user confirmation, and verifies citations; 7 points are deducted because the CLI path can still execute arbitrary code through restricted Node access, OCR performs an external download, and fine-grained sensitive-data isolation and rollback are not fully specified.
The workflow documents failure states, coverage tracking, quote retries, parameter validation, and diagnosable errors; 11 points are deducted because this static review did not execute the key paths, and Node, Claude CLI authentication, database, and dependency behavior remain unverified.
The audience, contract-question scenarios, accepted inputs, amendment handling, and unreadable-file boundaries are clearly described; 5 points are deducted because the skill depends on a local Claude Code filesystem environment, explicitly fits remote containers poorly, does not document Chinese-language support, and first-use OCR downloads may be unreliable on mainland-China networks.
The skill has a substantial workflow structure and README covering setup, dependencies, security, limitations, and usage; 6 points are deducted because license metadata is unknown and explicit versioning, changelog, maintenance ownership, update path, examples, and troubleshooting coverage are incomplete.
The intended output—cross-contract answers with verified citations—has clear practical value, with full-corpus reading and amendment precedence built into the design; 9 points are deducted because static review cannot validate result correctness, extraction may miss paraphrases or complex tables, rescue and human review may be required, and comparative cost-benefit evidence is limited.
The source includes input validation, citation checks, coverage records, and failure reporting, with benchmark claims documented in the README; 7 points are deducted because no tests were executed, independently reproducible third-party execution evidence is absent, and the performance and accuracy claims lack corroboration.
- Do not run from an external directory containing untrusted .claude configuration or agent definitions; copy only the contract files into a trusted directory.
- First-use OCR downloads language data; pre-stage dependencies and validate connectivity in mainland-China or air-gapped environments.
- Manually review complex tables, amendment chains, and rescue-read results; treat the documented accuracy and performance figures as unverified claims.
What it does & when to use it
This skill answers factual questions across a local collection of contract documents. It reads contracts and amendments, determines the relevant scope, accounts for superseding terms, and produces conclusions with citations that can be checked against the source text. It is intended for procurement, legal, and contract-analysis workflows involving multiple documents. It depends on the local filesystem and the contract-analysis service bundled with the plugin.
Reads contract documents from a local filesystem, defines a document set for the question, searches and extracts relevant clauses, evaluates amendment relationships, records unresolved or unreadable material, and returns a contract-analysis answer with verifiable quotations.
- A procurement lead needs to identify which supplier contracts contain a specific clause.
- A contract analyst needs to compare pricing, term, or rights across multiple agreements.
- A legal reviewer needs to determine how amendments changed the original contract terms.
- An operations team needs to locate and quote a particular obligation or restriction across local contracts.
Pros & cons
- Designed for questions spanning multiple contracts rather than only one document.
- Requires citations to be verifiable against the original document text.
- Explicitly addresses amendments, parse failures, and unresolved judgments.
- Contract files must be available on the local filesystem.
- Requires Node.js 22.13 or newer and the contract-analysis runtime.
- The workflow relies heavily on Claude CLI, named reader agents, and plugin MCP services, so portability requires adaptation.
- The source provides no test suite, performance benchmarks, or pricing information.
How to install
Install the repository as the Claude Healthcare plugin: /plugin marketplace add anthropics/healthcare, then run /plugin install healthcare@healthcare. The source does not document installation steps for other clients.
How to use
Place the contract files in a local folder, then ask a cross-document question such as: “Which contracts contain a termination right? Quote the relevant language and explain whether an amendment changed it.” The source defines the workflow but does not provide a separate parameterized command.