Golden Path

The golden path is the recommended road from idea to production, a term from Spotify that Netflix calls the paved road. You can also run any stage alone.

/golden-path

One command runs /product-manager:run (prd, prp) and then /sse:run (plan, dev, test, pr, monitor). Idea in, merged PR out. For the hands-off version with two approval gates and an intake stage, use /pipeline:run; see Autonomy.

flowchart LR
    idea([idea]) --> prd --> prp --> plan --> dev --> test --> pr --> merged([merged PR])
    subgraph PM["/product-manager:run"]
        prd
        prp
    end
    subgraph SSE["/sse:run"]
        plan
        dev
        test
        pr
    end

The five properties

It is opinionated: one pipeline and the repo's conventions. It is supported: sensors and evals gate every stage, so the harness catches drift. It is optional: you step off any time and run stages alone. It is self-serviceable: one command, no ticket to a platform team. It is transparent and extensible: every stage names what ran, and you override per repo through .claude/conventions/.

Start from a brief

The idea-brief builder at pierry.github.io/harness-kit/brief/ collects squad, problem, hypothesis, customers, and metric, checks them against the PRD conventions as you type, and emits a paste-ready /golden-path call. If you already know the idea, type /golden-path with the brief yourself.

Flags

Flags pass through to the SSE half.

Flag	Effect
`--local`	stop after test, no PR
`--sdd`	spec-driven loop: plan once, then dev, test, and eval until the PRP is met; local only
`--no-monitor`	open the PR, skip the merge watch

Stepping off the path

Detour	Command
PRD or PRP alone	`/product-manager:prd`, `/product-manager:prp`
one SSE stage	`/sse:plan`, `/sse:dev`, `/sse:test`, `/sse:pr`
dev and test, no PR	`/sse:run --local`
spec-driven loop	`/sse:sdd`
resume	`/pipeline:continue`
abandon the run	`/pipeline:reset`

Detours run the same sensors, evals, and artifacts. Only the one-command convenience goes away.

Per-discipline paving

The dev stage reads the repo's discipline conventions on top of the SSE defaults, and the project wins:

.claude/conventions/{backend,web,mobile,devops}.md

These files are feedforward the team controls; see Guides.

What runs behind the curtain

Every stage summary names the sensors, evals, guides, and refs that ran:

sensors: plan-structure ok
eval:    plan-quality 8.4/10 (attempts: 1)
guides:  pipeline.md, coding-style.md, skills/{area}/SKILL.md
refs:    prp/{feature_id}.md, conventions/{area}.md

The files behind them are plain markdown under .claude/agents/<agent>/sensors/, evals/, and guides/. Which judge scored the eval depends on /hk:eval; see Evals.

This page mirrors Golden Path in the wiki. Edit it there.