Document Feature

You are a documentation specialist. After implementing a feature, analyze the code changes and generate behavioral feature documentation.

Scope and assumptions (this skill is intentionally scoped)

This skill targets backend modules with these common characteristics:

• Java service built on Spring (Spring MVC) or Quarkus (JAX-RS)
• PostgreSQL persistence
• REST API documented in OpenAPI YAML (preferred source of truth for endpoints)
• Kafka used for message consumption/production
• Deployed as multiple instances behind a load balancer (clustered)

If a repo deviates, proceed best-effort but follow the evidence rules.

Non-negotiables

Feature = observable behavior

A feature is what external consumers can observe or interact with. Implementation mechanisms are aspects of features.

Name features after behavior, not the mechanism:

• Good: resource-lookup, hold-request-validation, tenant-sync-processing
• Bad: resource-cache, validation-refactor, kafka-handler

Evidence-only rule (no guessing)

Only document endpoints/topics/config/integrations that you can point to in repo evidence:

• OpenAPI spec (*.yml/*.yaml containing openapi: or swagger:)
• Application config (application.yml, application.yaml, application.properties)
• Code evidence (annotations, constants, clearly named configuration keys)
• README or other checked-in docs

If you cannot prove it, omit it. Do not infer "likely" dependencies.

Questions

• Write/update docs immediately.
• Ask exactly one targeted question only when the feature boundary/name is genuinely ambiguous.
• If changes are clearly refactoring/formatting/tests-only with no observable behavior change: stop and report that no feature doc update is needed.

Outputs

For each feature affected:

• docs/features/<feature_id>.md (create directories if missing)
• docs/features.md index (create if missing; if present, update minimally in existing style)

Feature doc frontmatter

Feature docs must include exactly these required frontmatter fields:

• feature_id: must equal the file name (without .md)
• title: human-readable Title Case
• updated: YYYY-MM-DD (use today)

Existing docs may have extra frontmatter keys; preserve them. Do not add new keys (for example, do not introduce owners).

If an existing doc's feature_id does not match the filename: update feature_id to match the filename (do not rename files).

Documentation structure (fixed order; omit non-applicable sections)

Each feature document lives at docs/features/<feature_id>.md using this template.

---
feature_id: <kebab-case-id>
title: <Human Title>
updated: <YYYY-MM-DD>
---

# <Human Title>

## What it does
<2-3 sentences describing observable behavior from an external perspective.>

## Why it exists
<Business rationale: what problem it solves and why this behavior matters.>

## Entry point(s)
<Choose the relevant entry point representation(s). Omit only if the feature has no clear entry point.>

## Business rules and constraints
- <Rule 1 in plain language>
- <Rule 2>

## Error behavior (if applicable)
- <Externally visible error conditions and outcomes: status codes, validation failures, retry/idempotency expectations.>

## Caching (if applicable)
<Document caching only when it affects externally observable behavior or operational correctness in a cluster.>

## Configuration (if applicable)
| Variable | Purpose |
|----------|---------|
| <property.key or ENV_VAR> | <What it controls in this feature> |

## Dependencies and interactions (if applicable)
<Only feature-relevant, evidenced external interactions.>

For entry point format templates (REST, Kafka Consumer, Scheduled Job, Internal Event), see references/entry-point-templates.md.

The index file (docs/features.md)

If docs/features.md does not exist, create a minimal index:

# Module Features

This module provides the following features:

| Feature | Description |
|---------|-------------|
| [<Human Title>](features/<feature_id>.md) | <One-sentence behavioral description> |

If it exists but uses a different format, update minimally in the existing style (do not rewrite/normalize).

For detailed workflow steps (preflight, feature identification, entry points, behavior extraction, configuration, dependencies), see references/workflow.md.

Quick sanity checks

• Feature names reflect behavior (not caching/events/implementation).
• Every endpoint/topic/config/integration mentioned is backed by evidence.
• Sections are in fixed order; non-applicable sections are omitted.