spec-kit-performance

A spec-driven-development (SDD) methodology for performance / load testing, built on GitHub spec-kit. It encodes generic load-testing methodology patterns as a ratified constitution and ships them as a spec-kit bundle — so a load-testing engagement runs as a governed, traceable, gated pipeline instead of a pile of ad-hoc scripts.

Status: research & design. No production code yet. The design lives in docs/. Methodology content is generic, retrospective inspiration — no specific platform, tool, or org is assumed or integrated.

The pipeline — 4 phases, 2 front-door actors

A · bootstrap (load engineer)  →  B · order (customer skill)  →  C · spec & plan (load engineers)   →  D · build & run
constitution + SUT                plain-language order           NFR prose → clarify → EARS → numeric   scripts → run → gate → report

• Customer files a plain-language order through a wiki-aware skill — no numbers required ("survive Black Friday" is enough).
• Load engineers bootstrap the repo (methodology + system-under-test), take the order, write the charter with NFRs in natural language, then quantify them into EARS SLOs compiled to a machine-readable nfr.yml — giving numeric, traceable, real pass/fail acceptance.
• A ratified constitution governs everything.

Packaging

Shipped as a spec-kit bundle:

Layer	What it does
preset	Reshapes core spec-kit phases (constitution / specify / clarify / plan / tasks / implement) for the perf domain
extension	Adds new phases: perf.order, perf.sut, perf.run, perf.analyze, perf.report
workflow	Gated automation chaining the phases end to end

Everything through the spec-kit CLI

Every capability is discoverable, installable, and runnable only via specify — no bespoke installers, no manual file copying as a supported path.

# Discover
specify preset search perf
specify extension search perf

# Install
specify init --preset perf
specify extension add <id>

# Run
specify workflow run        # or the /speckit.* agent commands

Distribution is the spec-kit catalog (public repo + GitHub Release zip + a PR to catalog.community.json), surfaced by specify ... search.

Repository layout

docs/            Design docs and ADRs (start here)
.specify/        spec-kit scaffolding: constitution, templates, scripts, extensions, workflows
.claude/skills/  speckit-* command skills
CLAUDE.md        Guidance for Claude Code working in this repo

Documentation — read in order

1. 01 — Spec-Kit, cover to cover — how spec-kit works: layers, init flow, command pipeline, presets/extensions/bundles/workflows.
2. 02 — SDD landscape & artifacts — spec-kit vs Kiro / BMAD / Agent-OS / Tessl / Roo; ADR / PRD / RFC / context-file vocabulary.
3. 03 — Prior art & inspiration — generic, retrospective catalogue of load-testing methodology patterns worth borrowing.
4. 04 — spec-kit-performance design — the proposed 4-phase / 2-actor flow, the order skill, NFR → EARS → numeric gate.
5. 06 — Decision: stack choice + harness — own vs embed vs other; the campaign-vs-fleet harness model.
6. ADR-0001 — packaging approach — decision to ship as a bundle (preset + extension), automated by a workflow.

Full index: docs/README.md.

License

Apache-2.0. See NOTICE for attribution.