/composure:blueprintPROTrial · 3/day

Pre-Work Assessment

Graph-powered planning with progressive refinement. Classifies work, scans code graph for related files, confirms scope with user, analyzes impact, writes implementation spec.

What it does

6 queries replace 76 agent calls (6.9x more efficient)

Progressive checkpoints shape the plan through conversation

Template forces per-file specs — catches design flaws during planning

Try this

Real Composure runs with measured metrics — compare to your current workflow before you invest the time to switch.

Blueprint vs Plan Mode (dual signup flow)

Next.js 16 · TypeScript · Stripe · OAuth 2.0

Adding a second signup path that reverses OAuth → Stripe → Dashboard into Website → Stripe → OAuth → Dashboard. Cross-cuts auth, billing, webhooks, checkout UI, and marketing pages.

2.75x

Fewer tokens

2.6x

Faster (3m 24s vs 8m 50s)

39/40

Quality score

OAuth CLI auth — plan + build in 28 minutes

Node.js · OAuth 2.1 PKCE · shell hooks · Next.js API

OAuth 2.1 PKCE flow, token management, slash commands, shell hooks, and a curl installer — planned + implemented across 2 repos in one session.

28 min

Plan + build (Opus)

97.7%

Prompt cache hit rate

807

Lines shipped, 8 files

Unified plugin catalog (this codebase)

Next.js · TypeScript · pnpm workspaces · Supabase

Catalog drift across 10 hand-curated locations (plugin-data, feature-data, plugin-catalog, etc.) consolidated into one canonical source in @composure/shared with nested Plugin → Feature → Skill layers and a sync pipeline from composure-pro.

3 phases

Shippable independently

44 ops

Across 5 PRs

47 skills

Now auto-synced

Graph narrows, agent deepens

Tree-sitter graph · MCP tools · Sub-agents

Code-graph queries replace broad Explore-agent fanout. Graph pinpoints the 8 files that matter; an agent then deep-reads just those instead of 30+ files of speculative exploration.

30x

Fewer tokens for discovery

<5s

Graph query vs 45+s of agent search

9,000→500

Words of context (avg)