Tutorials

SPARC 1.0 turns a methodology into an enforced state machine

SPARC was a way of working: specification, pseudocode, architecture, refinement, completion. Version 1.0 makes the five phases executable in TypeScript with deterministic gates, signed receipts, npm skills and both local and remote MCP transports.

GitHub activity: · Published:

What SPARC is

SPARC is an evidence gated engineering method. The difference in version 1.0 is that the five phases are enforced by code rather than suggested by a prompt: you cannot advance without passing the gate.

What changed

The August work packaged registered ChatGPT plugin apps and, immediately before it, shipped the deterministic SPARC metaharness and MCP plugins. Version 1.0 adds the TypeScript state machine, an npm CLI, local and remote MCP transports, shared Claude and OpenAI skills, principal isolation, exact idempotency and tamper evident receipts.

The established Python CLI stays available as a compatibility surface.

Get started

Node.js. Check the environment, then initialise:

npx --yes @ruvnet/sparc@1.0.0 doctor
npx --yes @ruvnet/sparc@1.0.0 init

Expected result: a doctor report and an initialised SPARC state store in the project.

Install the skills for your host:

npx --yes @ruvnet/sparc@1.0.0 skills install --host both --target .

Claude skills land in .claude/skills; Codex and ChatGPT Work skills land in .agents/skills.

npx --yes @ruvnet/sparc@1.0.0 mcp stdio

Installing a skill does not register an MCP server. The skill's JSON result names the exact launcher to register; do that explicitly, or the skill will have guidance but no tools.

Use it today

Practical case: a change that must not skip review. Start a run, submit the phase artefact, run the gate, advance only if it passes.

npx --yes @ruvnet/sparc@1.0.0 gate --run-id change-001

Acceptance test: try to advance a run whose gate failed. The state machine should refuse, and the refusal should be recorded in the receipt.

Push it further

Experimental commentary. Signed evidence makes agent work auditable in a way review comments never were. The dependency is key management: the attestation is only as meaningful as the verifier key map you configure in every process.

Falsifiable test: submit evidence signed with an unconfigured key. It should be rejected rather than quietly accepted.

Read the original on GitHub

feat: package registered ChatGPT plugin apps

SPARC repository

Back to the newsroom