# Discover > Front-load pattern discovery before implementing an OpenSpec change. Reads existing similar implementations to surface conventions, wiring patterns, file-length risks, and parallel workstreams. Use when this capability is needed. - Skill: `tomevault-io/discover-5` (Agent Skill, multi-file: 2 files) - Install (CLI): `npx skillmds@latest add tomevault-io/discover-5` - Raw SKILL.md: https://api.skillmd.com/api/skills/tomevault-io/discover-5/raw - Safety review: pending - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: Product & Planning - Author: tomevault-io (https://skillmd.com/u/tomevault-io) - Updated: 2026-09-17 - Page: https://skillmd.com/skills/tomevault-io/discover-5 --- Prepare a pattern discovery report for OpenSpec change: $ARGUMENTS ## Steps 1. Read `openspec/changes/$ARGUMENTS/tasks.md` to understand the full scope of work. 2. Scan `openspec/adrs/` for existing ADRs whose topic overlaps with this change. Read `adr.md` for any that seem relevant. Note their numbers and titles for the report. 3. Identify the implementation type (new provider, SDK wiring, middleware, E2E suite, etc.) and find 1–2 existing implementations of the same type to use as a reference: - **New simple provider** → the only file to edit is the service's own `routes.py`. Add a `DESCRIPTOR = ServiceDescriptor(name=..., factory=...)` constant and the service is auto-discovered. Do NOT list `inprocess.py`, `_SERVICE_NAMES`, `_build_service_apps`, or `_ldk_providers_extended.py` as edit points. - New complex provider → read an existing provider's `routes.py`, unit tests, integration tests, and E2E suite (e.g. `lang/python/core/src/lws/providers/sqs/`) - **New cross-cutting concern** → add a field to `ProviderContext` in `lang/python/core/src/lws/providers/_shared/provider_context.py`. Do NOT modify individual `create_*_app` factory signatures. - SDK wiring → read `lang/python/sdk/src/lws_testing/_transport/inprocess.py` and `_extended_services.py` - Shared middleware → read an existing middleware file and its wiring across provider routes - E2E suite → read an existing suite's `given/`, `when/`, `then/`, `conftest.py`, `constants.py` 4. For every file that will be **modified** (not created), check its current line count with `wc -l`. Flag any file within 50 lines of the 500-line architecture limit. 5. Group the tasks from `tasks.md` into parallel workstreams — identify which groups have no dependencies on each other and can run concurrently. ## Output Produce a structured report with these sections: ### Relevant ADRs List any ADRs found in `openspec/adrs/` whose topic overlaps with this change. For each: number, title, one-sentence summary of the constraint or decision it records, and whether it directly affects any task in this change. If no ADRs exist yet, say so. ### Reference implementations used List the files you read and why they were chosen. ### Key patterns to follow - Naming conventions (files, classes, functions) - Import style - Required boilerplate (e.g. AAA comments, `# Arrange`/`# Act`/`# Assert`) - Any non-obvious conventions (e.g. how `X-Amz-Target` headers are parsed, boto3 parameter names) ### File-length risks Table of files to be modified with current line count. Mark files within 50 lines of 500 as ⚠️. Suggest extraction targets if a file will exceed the limit. ### Parallel workstreams Group task numbers from `tasks.md` into independent workstreams that can run concurrently. Example: - Workstream A (core provider): tasks 1–3, 6–9 - Workstream B (SDK wiring): tasks 4–5 - Workstream C (E2E suite): tasks 8 (depends on A) ### Gotchas Specific pitfalls found in the reference code that are easy to miss: - e.g. boto3 sends `X-Amz-Target` with full namespace prefix, not just the action name - e.g. `sqs.create_queue` raises if queue already exists — wrap in try/except in Given steps - e.g. `StartTime`/`EndTime` arrive as Unix timestamps, not ISO strings --- > Source: [local-web-services/local-web-services](https://github.com/local-web-services/local-web-services) — distributed by [TomeVault](https://tomevault.io).