← ClaudeAtlas

task-flowlisted

Use when implementing a feature or multi-step task end-to-end — orchestrates spec intake, adaptive analysis, planning, and a per-task implement/verify/review loop over superpowers, with a single human gate after the plan; includes a human-pulled respec gate to correct the spec mid-build.
Endika/eskills · ★ 0 · AI & Automation · score 72
Install: claude install-skill Endika/eskills
# task-flow ## Overview My end-to-end flow for building a feature with trust. It is a **thin orchestrator** over `superpowers` — it sequences existing skills, it does not re-implement them. One human gate (after the plan); everything else runs automatically with per-task gates. The keystone is an adversarial verifier that re-earns "done" in fresh context. ## When to use - Implementing a feature or non-trivial task from a spec, ticket, or request. - NOT for a one-line fix or a quick question — the ceremony isn't worth it there. ## Phase 1 — understand & plan (auto, adaptive) 1. **`eskills:spec-intake`** — normalize the task into objective, acceptance criteria, constraints, definition of done. It also classifies **greenfield vs brownfield** and flags a **hard-tech challenge** (complex algorithm / performance bottleneck). 2. **If brownfield →** dispatch the **`code-explorer` agent** (Agent tool, from the optional `feature-dev` plugin) to trace the existing code before planning. It ships **agents, not skills** — there is nothing to invoke with the Skill tool. Without the plugin, use the built-in `Explore` agent; either way the step is not optional, only the tool is. 3. **If hard-tech challenge → SPIKE sub-phase:** explore 2–3 approaches, benchmark candidates if needed (apply `eskills:perf-bar`), and pick the approach **before** committing the plan. The **`code-architect` agent** (same plugin, same caveat) helps here when present; it is not require