guidewire-policycenter
Skill3 starsChanged 7 months agoApache-2.0View on GitHub
Answer bounded underwriting questions with the Guidewire PolicyCenter MCP server's five implemented read-only tools. Use when reviewing an underwriter's submission queue, retrieving a submission, listing an account's policies, or checking cancellation and non-renewal history. Trigger with "PolicyCenter", "submissions waiting on me", "pull this submission", "policies for this insured", or "did we lose this account".
The file
Show its rules and text
We split this skill into the rules it gives an agent
13 rules. About 0 are common, 10 are distinctive, and 3 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈10
Distinctive · estimate
This author's own choices. Worth a look.
3
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 35
Use the current PolicyCenter MCP surface to perform five carrier-vocabulary reads.
- Very specific: line 101
Use identifiers already provided by the user or returned by a prior tool call.
- Very specific: line 102
Never infer an accountId or jobId from a company name, policy number, or submission number.
- Distinctive: line 112
Call exactly one of:
- Distinctive: line 120
Keep the request bounded. Do not add fields, filters, or execution modes that are not in the selected tool's input schema.
- Distinctive: line 135
Do not claim that contacts, locations, coverages, premium, or exposures were retrieved unless they are present in the actual response.
- Distinctive: line 159
Treat policy, account, contact, premium, and cancellation data as sensitive insurance information.
- Distinctive: line 162
Do not expose OAuth credentials, authorization headers, or raw diagnostic payloads containing secrets.
- Distinctive: line 164
Do not use a read result to initiate a write, renewal, cancellation, quote, payment, or claims action.
- Distinctive: line 166
Do not silently cross actors, tenants, or carrier profiles.
- Distinctive: line 167
Do not claim production validation for endpoint filters that the repository marks for tenant smoke testing.
- Very specific: line 176
Not found: Verify whether the tool expects accountId or jobId.
- Distinctive: line 187
Use the queue tool for a bounded morning review, the account-loss tool for a confirmed account ID, and the summary tool followed by the …
≈ “Common” is an estimate: at least 1 in 5 of the 372,429 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Guidewire PolicyCenter
Overview
Use the current PolicyCenter MCP surface to perform five carrier-vocabulary reads. Select one tool that answers the question, pass only the identifiers and pagination requested, and distinguish returned facts from underwriting interpretation.
This skill does not authorize writes. The current tools have read_only mode
and call PolicyCenter Job or Policy APIs. ClaimCenter, BillingCenter, drafting,
approval, and execute workflows are roadmap scope, not capabilities of this
skill.
Prerequisites
- Install and build the
guidewire-mcpClaude Code plugin on Node.js 22 or newer so itspolicycenterMCP server is connected. - Obtain an authorized Guidewire Hub OAuth client and tenant endpoints.
- Know the required Guidewire
accountIdor JobjobId, unless a prior tool response in the same workflow supplied it.
Authentication
The policycenter server uses Guidewire Hub OAuth client credentials. Its first
real tool call requires these runtime environment variables:
GUIDEWIRE_OAUTH_CLIENT_IDGUIDEWIRE_OAUTH_CLIENT_SECRETGUIDEWIRE_TOKEN_ENDPOINTGUIDEWIRE_PC_BASE_URL
Optional runtime settings include GUIDEWIRE_TENANT_ID,
GUIDEWIRE_OAUTH_SCOPES (default pc.read), GUIDEWIRE_ACTOR_ID, and
GUIDEWIRE_OBS_LOG_LEVEL.
The server can start without OAuth configuration, but calls then fail loudly; it does not substitute fixtures or mock data. Never ask the user to paste a client secret into chat, and never reproduce tokens or secrets in output.
Tool Selection
| User intent | Tool | Inputs | Returns |
| Review an underwriter's open queue | mcp__policycenter__find-submissions-waiting-on-me | Optional actorId; optional pageSize, pageOffset | Count and submission summaries |
| List policies for a known insured account | mcp__policycenter__show-policies-for-this-insured | accountId; optional pagination | Policy number, status, line, dates |
| Get an elevator-pitch submission view | mcp__policycenter__summarize-this-submission | jobId | Submission number, status, assignee, kind, optional account ID |
| Check cancellation, lapse, or non-renewal history | mcp__policycenter__did-we-lose-this-account | accountId; optional pagination | Lost policies and cancellation reasons |
| Retrieve the full attributes of one submission | mcp__policycenter__pull-this-submission | jobId | Core fields plus the complete returned attributes object |
accountId and jobId are Guidewire resource identifiers, not display names
or submission numbers. The current tool catalog has no account-name search or
identifier-resolution tool. If the user supplies only a name or business-facing
number and the needed resource ID is absent from conversation context, explain
the limitation instead of guessing.
Pagination rules are pageSize from 1 through 100 and a non-negative
pageOffset. Default to the server's page size of 20 unless the user requests a
different bounded result set.
Instructions
Step 1: Classify the question
Choose the single closest intent from the tool table. Do not call all five tools for a question one call can answer. If the request includes several distinct questions, state the minimal call sequence before proceeding.
Step 2: Confirm identifiers
Use identifiers already provided by the user or returned by a prior tool call.
Never infer an accountId or jobId from a company name, policy number, or
submission number.
For mcp__policycenter__find-submissions-waiting-on-me, omit actorId to use
the server's configured actor. Only provide another actor when the user
explicitly requests that person's queue and the deployment's access policy
allows it.
Step 3: Invoke the read
Call exactly one of:
mcp__policycenter__find-submissions-waiting-on-memcp__policycenter__show-policies-for-this-insuredmcp__policycenter__summarize-this-submissionmcp__policycenter__did-we-lose-this-accountmcp__policycenter__pull-this-submission
Keep the request bounded. Do not add fields, filters, or execution modes that are not in the selected tool's input schema.
Step 4: Interpret conservatively
Preserve PolicyCenter facts such as identifiers, statuses, dates, counts, and cancellation reasons. Apply these rules:
- An empty list means no matching records were returned; it does not prove the account or submission never existed.
- A raw typelist code or blank label may indicate that the carrier profile is incomplete.
- The default in-memory profile is not carrier-specific. A supplied profile can map field aliases, lines of business, roles, and typelist labels.
summarize-this-submissioncurrently projects fields from one Job resource. Do not claim that contacts, locations, coverages, premium, or exposures were retrieved unless they are present in the actual response.pull-this-submissionexposes the returned attributes object; summarize only fields relevant to the user's question and avoid unnecessary personal data.
Step 5: Report with provenance
Lead with the direct answer, followed by the relevant identifiers and returned facts. State which PolicyCenter tool supplied the result. Clearly label any inference, and never present a business recommendation as a Guidewire fact.
The current server emits structured start, completion, and failure audit events through observability. Do not claim that a hash-chained persistence layer was used unless the running deployment independently proves that integration.
Output
Return the direct answer first, then a compact fact list containing the relevant Guidewire identifiers, statuses, dates, counts, and source tool. Label inferences explicitly. When the call fails or data is incomplete, return the failure boundary and the safe next check instead of fabricating a result.
Safety and Data Handling
- Treat policy, account, contact, premium, and cancellation data as sensitive insurance information.
- Return only fields necessary to answer the question.
- Do not expose OAuth credentials, authorization headers, or raw diagnostic payloads containing secrets.
- Do not use a read result to initiate a write, renewal, cancellation, quote, payment, or claims action.
- Do not silently cross actors, tenants, or carrier profiles.
- Do not claim production validation for endpoint filters that the repository marks for tenant smoke testing.
Error Handling
- OAuth not configured: Name the missing environment variable category and stop. Do not request secret values in chat.
- Unauthorized or forbidden: Confirm tenant, scopes, actor, and carrier policy outside the response body before retrying.
- Not found: Verify whether the tool expects
accountIdorjobId; do not substitute a display number. - Profile load failure: Report the named profile file and validation path from the error. Do not fall back silently to the default profile.
- Rate limit or transient service failure: Report that the result is incomplete. Retry only when safe, honoring provider guidance.
- Unknown field or typelist code: Preserve the raw value, label it as unresolved, and recommend updating the carrier profile.
Examples
Use the queue tool for a bounded morning review, the account-loss tool for a confirmed account ID, and the summary tool followed by the pull tool only when the user requests a deeper drill-down. See the exact sequences and response boundaries in the tool contracts.
Success Criteria
A successful invocation uses a real Guidewire resource identifier, performs only the smallest necessary read, reports the response accurately with tool provenance, protects sensitive insurance data and credentials, and makes no claim that exceeds the current server implementation.
Resources
- PolicyCenter tool contracts and examples
- Repository sources:
.mcp.json,servers/policycenter-mcp/src/index.ts, andservers/policycenter-mcp/src/tools/
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

