Repository navigation
[Idea]: AI-assisted idea-to-system pipeline tool #3
Description
Activity
Maintainer acknowledgement (UNDER REVIEW) — acknowledged within the 7-day SLA. Decision: proceed to Research. This is a real feasibility pass, not a formality: the orgs own pipeline is the use case, so the bar is "does the tool make the pipeline more accessible and more honest, and can we build it in-scope?".
RESEARCH — feasibility & scope write-up (go/no-go: GO, explored, not assumed)
Feasibility (could we build it in-scope?)
- Scope control is achievable only by building a CLI with a narrow surface and NO GUI in v0. The moment a web UI enters the picture this becomes a month-long yak-shave; the CLI keeps v0 at 2–3 months part-time.
- The pipeline is well-bounded: IDEA → CONCEPT → ARCHITECTURE → SYSTEM → AUTOMATION → REALITY. Each stage has a natural artifact (feasibility notes, architecture sketch, scaffolded repo). That artifact-speech is the core product: a prompt chain + spec templates + an eval harness that grades the outputs.
- The org has a real first-contribution surface here: eval harness construction, spec-template drafting, and prompt-engineering are good-first-issue sized.
- The risk is not technical but honest-scoring: the eval harness can silently become theater. Mitigation (binding): the harness runs offline, publishes raw outputs, and the go/no-go decision stays human — no automated "ship" verdict, ever. Aligns with §21.
Key open questions found (to resolve during Architecture, not now):
- How are a generated spec and a human spec merged when a human takes over? (v0: generated spec is a starting draft, clearly labeled.)
- Which model/provider? (v0: provider-agnostic prompt files; no API keys in the org.)
Verdict: GO to Architecture. Two explicit constraints recorded: CLI-only v0, and human-in-the-loop go/no-go everywhere.
Project home created (M9): this idea now has its flagship repository — idea-forge (status
concept). The entry bar for that repo: an accepted Architecture note for v0. This issue stays open and tracked on the Idea Lifecycle board until the Architecture stage hands off.
One-line idea
An AI-assisted idea-to-system tool that walks a rough idea through the orgs own stages — feasibility notes, architecture sketching, system scaffolding — so that the pipeline this laboratory exists to run is itself the first product it runs on.
Problem it solves
The orgs founding premise: most ideas fail between concept and execution. A tool that makes the bridge explicit, reproducible, and inspectable closes exactly that gap for everyone who wants to ship a small real system but stalls at where do I start. Today that gap is closed only by human mentorship; this product would close it at scale, publicly, and with an auditable paper trail.
Why Hidden Alchemy
It is the pipeline, dogfooded: IDEA -> CONCEPT -> ARCHITECTURE -> SYSTEM -> AUTOMATION -> REALITY is precisely the workflow this tool would automate. It sits in the orgs domain alignment (AI, automation, dev tools) and gives newcomers a concrete, meaningful first-contribution surface (prompt engineering, spec templates, eval harness) rather than abstract process docs.
Rough scope
A minimal CLI fallback first (no GUI): staged spec-generation templates, a feasibility/risk checklist, an architecture-sketching prompt chain, an eval harness to grade output quality, and full docs. ~2-3 months part-time; realistically one maintainer plus a coat of good-first-issues. Not a chatbot wrapper a weekend dash-and-dash course — a disciplined, recorded pipeline.