← ClaudeAtlas

start-tasklisted

Start a backend task through the Groundwork AI-SDD discovery/plan workflow. Use at the beginning of any non-trivial feature, bug fix, or API/schema/permission/workflow/integration change, before writing code.
YasMax91/groundwork · ★ 0 · AI & Automation · score 66
Install: claude install-skill YasMax91/groundwork
# Start task (discovery & plan) Stay in **Discovery mode**: inspect and plan only, do not edit files until the plan is approved. ## Steps 1. **Read the project `AGENTS.md`** for domain facts, invariants, permission rules, and runtime commands. 2. **Classify the task** (L0 tiny → L4 critical). State the level and what it implies. The level also decides **what you read**: at L0/L1 stay inside this skill — the per-level scale-downs below already state the whole contour (one question at most, a fail-first regression test, a live exercise of the one thing fixed), so opening the fan-out, clarify and blind-spot guidelines spends thousands of tokens to arrive back here. Open them from **L2** up, and only the ones that step actually needs. 3. **Inspect the directly-involved code.** Prefer Boost MCP tools — `application-info`, `database-schema`, `search-docs` — over recalling from memory. Read the actual routes, requests, controllers, services, models, resources, migrations, and tests the task names. - **Symbols: the `LSP` tool before grep**, when a PHP language server is active (`php-lsp`) — it follows imports, aliases and inheritance a name match misses. Grep still owns what it cannot see: string class names, `event('…')`, container bindings, config- and Blade-driven wiring. The `impact-mapper` subagent has no `LSP` tool, so resolve a symbol its map left open yourself. - **A production bug starts at the incident.** When the report comes fr