api-development

name	api-development
description	REST API development guide for the HMIS project. Use when creating or extending any REST API endpoint — covers file structure, auth pattern, response format, ApplicationConfig registration, CapabilityStatementResource update, AnthropicApiService integration (system prompt module + tool handler), and the full post-implementation checklist.
user-invocable	true

HMIS REST API Development Guide

When This Applies

Any time a new @Path class is created under com.divudi.ws.*, or an existing API gets new endpoints.

If you instead need to call an existing endpoint (verify data via API, write a debug script, check a request/response shape), use the api-usage skill instead — this one has no consumer-facing reference.

---

Mandatory Checklist — Every New API

1. File placement

• REST class: src/main/java/com/divudi/ws/<module>/<Name>Api.java
• @Path("<name>"), @RequestScoped
• No @Stateless on the REST class itself

2. Auth — always Finance header (unless module uses a different scheme)

String key = requestContext.getHeader("Finance");
if (validateApiKey(key) == null) return errorResponse("Not a valid key", 401);

Copy validateApiKey, errorResponse, successResponse verbatim from DepartmentApi.java (lines 437–488). Do not reinvent.

3. Response envelope — always this shape

{"status":"success","code":200,"data":{...}}
{"status":"error","code":400,"message":"..."}

For POST duplicate-detection: {"status":"already_exists","id":...,"name":...} (no wrapping envelope).

4. Gson

private static final Gson gson = new GsonBuilder().setDateFormat("yyyy-MM-dd HH:mm:ss").create();

5. Register in ApplicationConfig

src/main/java/com/divudi/ws/common/ApplicationConfig.java — add one line to addRestResourceClasses():

resources.add(com.divudi.ws.<module>.<Name>Api.class);

6. Document in CapabilityStatementResource

src/main/java/com/divudi/ws/common/CapabilityStatementResource.java — add a resource(...) entry to buildResources().

7. Expose to AI Chat (AnthropicApiService) — TWO places

Both are required every time:

a) buildSystemPrompt — add an appendModule(...) block so the system prompt tells Claude what the endpoint does:

appendModule(sb, "My Module", "/my_module",
    "Description of what this module manages.",
    null,  // or githubUrl(branch, "developer_docs/api/using-apis/API_MY_MODULE.md")
    new String[][]{
        {"GET",    "/my_module",      "List entries. Supports query, page, size"},
        {"POST",   "/my_module",      "Create a new entry. Body: {name, code}"},
        {"PUT",    "/my_module/{id}", "Update an entry by ID"},
        {"DELETE", "/my_module/{id}", "Soft-delete an entry by ID"}
    });

b) buildToolsArray + executeToolCall — add a tool so Claude can actually call the API:

• Add a JsonObject myModuleTool in buildToolsArray() and include it in the returned array.
• Add a case "my_tool_name": in executeToolCall(...) that calls a private callMyModuleApi(...) method.
• The callMyModuleApi method uses hmisBaseUrl + hmisApiKey (both available as params to executeToolCall).
• AiChatController.sendMessage(...) already passes hmisApiBaseUrl and userHmisApiKey through to executeToolCall — use them.

See ClinicalMetadataApi + manage_clinical_metadata tool in AnthropicApiService as a reference implementation.

8. Write a developer_docs API file (optional but recommended)

developer_docs/api/using-apis/API_<MODULE>.md — list all endpoints, params, example request/response. Reference it in the appendModule(...) call via githubUrl(branch, "developer_docs/api/using-apis/API_<MODULE>.md").

---

JPQL Patterns for Simple CRUD APIs

List (with optional text search + limit)

Map<String, Object> params = new HashMap<>();
params.put("t", symanticType);  // if filtering by type
String jpql;
if (query != null && !query.isEmpty()) {
    jpql = "select c from MyEntity c where c.retired=false and c.symanticType=:t"
         + " and upper(c.name) like :n order by c.name";
    params.put("n", "%" + query.trim().toUpperCase() + "%");
} else {
    jpql = "select c from MyEntity c where c.retired=false and c.symanticType=:t order by c.name";
}
List<MyEntity> items = facade.findByJpql(jpql, params, size);  // 3-arg: (jpql, map, maxRecords)

Duplicate check before POST

MyEntity existing = facade.findFirstByJpql(
    "select c from MyEntity c where c.retired=false and upper(c.name)=:n",
    Map.of("n", name.toUpperCase()));
if (existing != null) {
    // return already_exists response
}

Soft delete

entity.setRetired(true);
entity.setRetiredAt(new Date());
facade.edit(entity);

---

No-service-layer Rule

For thin CRUD over a single entity (like ClinicalMetadataApi), no @Stateless service EJB is needed — inject facades directly into the REST class. Only add a service layer when there is real business logic, multiple entity coordination, or the method is reused elsewhere.

---

Testing After Implementation

# List
curl -s -H "Finance: <key>" "http://localhost:9090/rh/api/<path>?type=X"

# Create
curl -s -H "Finance: <key>" -H "Content-Type: application/json" \
  -X POST "http://localhost:9090/rh/api/<path>?type=X" \
  -d '{"name":"Test"}' | python -m json.tool

# Duplicate check (POST same name again — must return already_exists)
# same command as above

# Update
curl -s -H "Finance: <key>" -H "Content-Type: application/json" \
  -X PUT "http://localhost:9090/rh/api/<path>/<id>" \
  -d '{"name":"Updated"}' | python -m json.tool

# Delete
curl -s -H "Finance: <key>" -X DELETE "http://localhost:9090/rh/api/<path>/<id>" | python -m json.tool

# Capabilities (confirm new entry appears)
curl -s http://localhost:9090/rh/api/capabilities | python -m json.tool

For complete reference: developer_docs/api/building-apis/rest-api-development-guide.md