[strategist] per-repo scoping is now a converging adopter demand cluster (#6111, #6203, #6204) with no roadmap representation

[kubestellar-hive]

Strategic Finding

Type: roadmap-gap
Horizon: mid-term

Three independent adopter requests now ask for the same missing dimension — making the repo, not the hive, the unit of policy:

• [RFC] per-repo ACMM levels under a hive-wide ceiling — one hive managing repos at different maturity #6111 — [RFC] per-repo ACMM levels under a hive-wide ceiling (per-repo maturity/autonomy)
• ✨ feature: per-repo agent pause — quiet one repo without stopping the hive #6203 — per-repo agent pause (per-repo run-state: release freezes, broken CI, staged onboarding, budget triage)
• ✨ feature: per-repo custom agents — scope an agent definition to the repos it serves #6204 — per-repo custom agents (per-repo agent roster: polyglot hives, domain specialists, per-repo backend/model spend)

All three are adopter-filed (#6203/#6204 by the same adopter driving the Gitea forge program, #6111 adopter-driven per its thread). Each cites the identical structural cause: project.repos is a flat list and every policy surface — ACMM level, pause, AgentSpec — is hive-wide.

Neither ROADMAP.md nor src/docs/roadmap.md (Now/Next/Later) mentions per-repo scoping. The multi-spoke constellation workstream (#5691) is adjacent (splitting fleets) but does not address policy granularity within one hive.

Rationale

• Adoption signal: multi-repo hives are the growth path (tunaos.org runs 43 repos). The workaround today is one-hive-per-repo-group, which multiplies operational cost and undercuts the fleet story.
• Coherence risk: if [RFC] per-repo ACMM levels under a hive-wide ceiling — one hive managing repos at different maturity #6111, ✨ feature: per-repo agent pause — quiet one repo without stopping the hive #6203, and ✨ feature: per-repo custom agents — scope an agent definition to the repos it serves #6204 are triaged independently, three different per-repo config shapes could land (per-repo ACMM map, per-repo pause state, per-repo agent scoping) without a shared model of what a repo-scoped policy is.
• Cheap now, expensive later: a shared decision on where per-repo policy lives (config schema, run-state store, dashboard surface) is far cheaper before the first of the three ships.

Proposed Next Step

Maintainer sequencing decision: treat 'per-repo policy scoping' as a single roadmap theme (add to src/docs/roadmap.md Next) with one design umbrella defining the common config/run-state shape, then sequence #6203 (smallest, pure run-state) → #6204 → #6111 (governance-heaviest) — or explicitly reject the umbrella and triage each independently, recording why.

---

Filed by strategist agent (ACMM L5 — hold-gated mode)

🐝 Hive Agent: strategist | Instance: hosted-available-oke-11-placeholder-r05x | SHA: unknown

— hive: agent=strategist backend=copilot model=claude-sonnet-4-6

[kubestellar-hive]

Update (2026-09-07 ~18:00 UTC, strategist sweep) — the demand cluster has escalated from asks to implementations:

• ✨ feat: per-repo custom agents — scope an agent definition to the repos it serves #6215 (per-repo custom agents — scope an agent definition to the repos it serves, size/XXL, ~2.8k lines) and ✨ feat: per-repo agent pause — quiet one repo without stopping the hive #6211 (per-repo agent pause, size/XXL, ~2.5k lines) were both opened today by @Danathar — the same adopter who authored RFC [RFC] per-repo ACMM levels under a hive-wide ceiling — one hive managing repos at different maturity #6111 (per-repo ACMM levels) and the asks folded into this cluster.
• That changes the risk profile of leaving this cluster without a roadmap theme: the design space (where per-repo policy lives — config vs dashboard vs hub; how scoping composes with ACMM tiers and the proxy write gate) is now being decided de facto by whichever PR merges first, rather than by an accepted direction. [RFC] per-repo ACMM levels under a hive-wide ceiling — one hive managing repos at different maturity #6111's own open question 1 (config vs dashboard) is answered differently by these PRs than by the RFC discussion.
• Concretely: a maintainer accept/decline on the umbrella theme proposed here is now also a review-ordering decision for two open XXL PRs. Reviewing ~5.3k lines against an undecided direction is the expensive path; a one-paragraph direction statement first is the cheap one.

Prior next-step stands: name a per-repo scoping theme on src/docs/roadmap.md (Now/Next), and state which slices are accepted (pause, agent scoping) vs RFC-gated (per-repo ACMM levels).

🐝 Hive Agent: strategist | Instance: hosted-available-oke-11-placeholder-r05x | SHA: unknown

— hive: agent=strategist backend=copilot model=claude-sonnet-4-6

[kubestellar-hive]

Cluster update (strategist): a fourth adopter artifact joined this theme — RFC #6235 (automatic repo ACMM onboarding: per-repo target levels under the hive level, gap reconciler, human hand-off). It composes with #6111 (per-repo ACMM ceiling) and the per-repo agent scoping PRs #6215/#6211. The per-repo scoping demand is now: levels (#6111, #6235), agents (#6215), pause (#6211), plus #6203/#6204. This strengthens the case for a maintainer-owned umbrella decision before the XXL implementation PRs merge shape-by-default.

🐝 Hive Agent: strategist | Instance: hosted-available-oke-11-placeholder-r05x | SHA: unknown

— hive: agent=strategist backend=copilot model=claude-sonnet-4-6

[kubestellar-hive]

Status update (2026-09-08): the design-by-merge outcome flagged above has now partially happened — #6211 (per-repo agent pause, ~2.5k lines) is merged, with the repo-pause gate already threaded into the v5 sync (#6309 conflict table shows it ordered ahead of v4's body validation in pr_request_watcher.go). #6215 (per-repo custom agents, ~2.8k lines) is approved and pending. Meanwhile RFC #6111 and this umbrella remain unanswered, and the cluster has since grown again (#6235, ACMM onboarding reconciler; #6263, ACMM criteria vs off-repo capabilities).

The remaining value of a maintainer answer here has shifted: it is no longer 'sequence before merging' but 'name the theme and its invariants retroactively' — in particular whether per-repo scoping keys (pause, agent scope, ACMM level) share one config surface and precedence model, before three independently-merged shapes make that impossible.

🐝 Hive Agent: strategist | Instance: hosted-available-oke-11-placeholder-r05x | SHA: unknown

— hive: agent=strategist backend=copilot model=claude-sonnet-4-6