From software-factory
Build a full software project or website end-to-end from a single brief, by orchestrating the whole tool stack: superpowers discipline (brainstorm/plan/TDD/verify), karpathy code restraint, a ruflo agent swarm with shared memory, ECC agents + AgentShield security gate, and open-design for frontend. Use when the user wants to build, scaffold, or ship a complete app/website/service and wants one command to run the entire pipeline hands-free.
How this skill is triggered — by the user, by Claude, or both
Slash command
/software-factory:software-factoryThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
One invocation, full pipeline. Turns a brief into shipped software by chaining
One invocation, full pipeline. Turns a brief into shipped software by chaining superpowers (discipline) + karpathy (code restraint) + ruflo (swarm engine) + ECC (agent catalog + security) + open-design (frontend) + Obsidian (durable memory).
User says "build me X", "make a full website/app for Y", "/factory ...". For a 1-2 file fix, do NOT use this — just edit directly. This is for real multi-file builds.
This plugin ships its glue script at scripts/orchestrate.mjs. Resolve it as:
"$CLAUDE_PLUGIN_ROOT/scripts/orchestrate.mjs"
(Windows cmd/PowerShell: "%CLAUDE_PLUGIN_ROOT%\scripts\orchestrate.mjs" /
"$env:CLAUDE_PLUGIN_ROOT\scripts\orchestrate.mjs").scripts/factory/orchestrate.mjs.
Pick whichever exists. Below, <ORCH> means that resolved path.node "<ORCH>" doctor. It reports which tools are present (node/git/npx
required; ruflo + superpowers core; agentshield gate; obsidian + open-design
optional). If a required item is missing, stop and tell the user. Missing
core/optional tools are fine — note which phases will run degraded and continue.superpowers:brainstorming. Do NOT skip. Ask one question at a time,
propose 2-3 approaches, converge on a written design. karpathy rules stay
active on every turn: surface assumptions, simplest thing that works, no bloat.docs/superpowers/specs/<date>-<feature>-design.md.obsidian create path="projects/<feature>" template="spec" then
append the design body. Skip silently if Obsidian is not running.superpowers:writing-plans. Produce a task-broken plan at
docs/superpowers/plans/<date>-<feature>.md using ## Task N: headers,
**Files:** blocks, and - [ ] steps. No placeholders, exact paths.node "<ORCH>" plan docs/superpowers/plans/<date>-<feature>.md --seed-memory
It writes .swarm/dispatch/<slug>.json (tasks + assigned agent type + model
tier) and a .brief.md. Read the manifest JSON — it is your task queue.npx ruflo@latest swarm init
(or the mcp__ruflo__swarm_init tool with topology=hierarchical).superpowers:subagent-driven-development. For EACH task in the
manifest, in dependency order:
agent type (coder / frontend-dev /
backend-dev / tester / security-reviewer / architect / documenter), named
task-<N>, using the suggested model tier.superpowers:test-driven-development); obey
karpathy surgical-changes; before coding, recall prior decisions from ruflo
memory (mcp__ruflo__agentdb_hierarchical-recall query="<feature> <topic>");
after finishing, store what worked back to memory.superpowers:dispatching-parallel-agents), each in its own git worktree
(superpowers:using-git-worktrees) to avoid conflicts.DESIGN.md brand spec, then pull the
emitted HTML/components from .od/artifacts/ into the repo. Otherwise build
directly with the frontend-dev agent.node "<ORCH>" gate .
Deep adversarial pass: node "<ORCH>" gate . --opus.
Also run mcp__ruflo__aidefence_scan for leaked secrets/PII. Fix all
criticals, re-run until clean.superpowers:verification-before-completion. Actually run the app /
build / tests and observe real output. No "should work" — show evidence.npm run build && npm test (or the project's equivalent) must pass.superpowers:finishing-a-development-branch → PR. Write decisions/lessons back
to the Obsidian vault and store the run's patterns in ruflo memory.ecc-agentshield (security gate) — npm i -g ecc-agentshield.Guides creation, editing, and verification of skills for AI coding agents using test-driven development with subagent scenarios. Use when authoring or debugging skills.
npx claudepluginhub yasnoyas944-ops/claude-software-factory --plugin software-factory