You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Update — 2026-06-16 (maintainer). The two strategic calls in this issue are now decided and recorded as ADRs. Move 1 is amended below; moves 2 and 3 are resolved.
ADR-0005 — split the persona register: gates stay terse, the thinking surfaces (brainstorming/debug/SMARTS) go conversational. (Resolves move 2.)
ADR-0006 — broad-adoption OSS posture: codeArbiter stays broad OSS optimized for adoption and declines re-targeting to a regulated/commercial vertical. (Overrules move 3.)
Source: 2026-06 multi-agent market-readiness evaluation (objective grader). Market-readiness scored 4/10, and that number is about distribution, not code. Engineering is genuinely strong; the gap is that the product has never met a second human.
Signal: ~5 weeks old, 3 stars, 0 forks, 1 watcher; every issue and PR self-authored. Zero external validation that anyone but the author wants this.
The trajectory-changing moves (above any feature):
Get 5 strangers to install it cold and watch them.Amended. Watching 5 strangers is not feasible: the maintainer does not have 5 external coders to observe, and a survey from them is off the table. The maintainer has watched 5 cold installs on his own team, which surfaced real friction (the gates feel hostile until you see what each one catches; /ca:btw discoverability; the slash-command habit takes time) but is not stranger validation. New move 1: get found, and make the repo do the convincing. Since no one will be in the room, the README and the demo are the entire onboarding. Optimize discovery (where the audience already is) and the first five minutes on the repo page.
✅ Decided — see ADR-0005. Split the register: keep gates terse; make brainstorming/debug/SMARTS conversational. (Original eval note: "Terse by design" structurally caps the back-and-forth that makes assistants feel alive, which is why daily use feels flat.)
✅ Decided — see ADR-0006 (declined as written). The eval recommended re-targeting the ICP to regulated / audited / multi-dev-on-agents teams. The maintainer declined: this is broad OSS, the goal is public adoption of a tool with demonstrated value, not a commercial vertical. The audit-trail / SMARTS machinery stays as a broad-audience quality feature.
Source: 2026-06 multi-agent market-readiness evaluation (objective grader). Market-readiness scored 4/10, and that number is about distribution, not code. Engineering is genuinely strong; the gap is that the product has never met a second human.
Signal: ~5 weeks old, 3 stars, 0 forks, 1 watcher; every issue and PR self-authored. Zero external validation that anyone but the author wants this.
The trajectory-changing moves (above any feature):
Get 5 strangers to install it cold and watch them.Amended. Watching 5 strangers is not feasible: the maintainer does not have 5 external coders to observe, and a survey from them is off the table. The maintainer has watched 5 cold installs on his own team, which surfaced real friction (the gates feel hostile until you see what each one catches;/ca:btwdiscoverability; the slash-command habit takes time) but is not stranger validation. New move 1: get found, and make the repo do the convincing. Since no one will be in the room, the README and the demo are the entire onboarding. Optimize discovery (where the audience already is) and the first five minutes on the repo page.✅ Decided — see ADR-0005. Split the register: keep gates terse; make
brainstorming/debug/SMARTS conversational. (Original eval note: "Terse by design" structurally caps the back-and-forth that makes assistants feel alive, which is why daily use feels flat.)✅ Decided — see ADR-0006 (declined as written). The eval recommended re-targeting the ICP to regulated / audited / multi-dev-on-agents teams. The maintainer declined: this is broad OSS, the goal is public adoption of a tool with demonstrated value, not a commercial vertical. The audit-trail / SMARTS machinery stays as a broad-audience quality feature.
Radically shorten time-to-first-value (see the
/ca:preview+ lite-mode issues). Now the top active workstream, since amended move 1 makes the repo the only onboarding: [positioning] Name the ICP + pre-empt the overhead objection (small README edits) #72 (README proof + positioning), [positioning] Record the demo GIF and place it above the fold #71 (demo GIF), [feature] /ca:preview — zero-onboarding dry-run of the reviewer fleet #81 (/ca:preview).This is the umbrella issue. The dimension issues below are the concrete work; this one holds the strategic call.