← ClaudeAtlas

prompt-architectlisted

Classifies a task against a 10-bucket taxonomy, decides output architecture (single-turn vs. chained multi-stage), and produces a technique-grounded v1 draft prompt. Verdicts are SINGLE-TURN, SINGLE-TURN (CEILING) with a strain note, or CHAIN producing a README + Project orchestrator + per-stage prompts. Use when you describe a non-trivial task but have no prompt written yet. Trigger phrases include "help me prompt this", "set up a prompt for", "draft a prompt for", "how should I ask Claude to", "architect a prompt for", "I'm about to start work on X — help me frame it". Also trigger for multi-step analytical tasks (research, knowledge document, design, decision support) where no draft exists. Do NOT trigger when a draft prompt already exists — that's prompt-optimizer. Do NOT trigger for tasks already covered by a dedicated domain-specific skill — route to that skill instead.
viktor-milev/claude-skills · ★ 0 · AI & Automation · score 72
Install: claude install-skill viktor-milev/claude-skills
# Prompt Architect (State-of-the-Art Edition) ## CHANGELOG **Validated for Opus 4.8 — June 2026.** Re-tested end to end against Claude Opus 4.8 with no changes required; the architecture-decision module, the 10-bucket taxonomy, and the technique stacks all hold. Published to the public repo at v2. **v2 — May 2026.** Introduces architecture decision as a first-class step, in lockstep with the prompt-optimizer v4 upgrade. The v1 skill had the same structural alignment problem the optimizer v3 had: it optimized a proxy (a well-formed single-turn v1 draft) rather than the actual target (state-of-the-art Claude output on the user's task). For genuinely complex work, those objectives diverge — forcing a single-turn structure compresses output the user could have had as a chain. The v1 Bucket 9 "two-phase handling" was a narrow workaround for one specific case (design/architecture); v2 generalizes it into an explicit architecture decision that applies to every IN-PURVIEW task. v2 adds a new Step 2 (Architecture decision, between Step 1 Classify and Step 3 Technique selection). The decision drives branched behavior in Steps 3–4 and produces one of two distinct output formats: - **Single-turn template** — the v1 output, refined. Includes a new SINGLE-TURN (CEILING) sub-band that makes the "fits but at the edge" cases visible to the user (previously silent). - **Chain template** (new) — produces a README/setup guide, a Project orchestrator (custom instructions), and per-stage pro