spec(m17): multi-repo + usage dashboard + specialization + design loop + model UI #151

Merged

charles merged 1 commit from specs/m17-milestone into main

2026-04-20 14:36:52 +00:00

claude-desktop commented

2026-04-20 14:29:12 +00:00

Collaborator

Working document for the next milestone's five themes. Once merged, I'll dispatch the breakdown skill (A7, #147) against specs/m17-multi-repo-and-observability.md and have boss generate the user-story issues.

Themes

M17-1 Multi-repo — lift the hardcoded charles/claude-hooks scope so the fleet serves loom, proxmox-iac, ws-rpc-test alongside the current repo.
M17-2 Usage dashboard — Anthropic-console-style panel for Pro Max weekly token consumption + thresholds. (Pro Max = subscription, not per-token, so "cost budget" becomes "weekly usage vs configurable threshold".)
M17-3 Reviewer specialization — reviewer-security (opus, match area:security) + reviewer-fast (haiku, match type:chore). Pure config + tests.
M17-4 Design-reviewer end-to-end — wire the idle design-reviewer into a real feedback loop on area:dashboard PRs with diff-scoped skill.
M17-5 Model switch UI — model dropdown in the edit-agent modal + PATCH endpoint.

Review notes for the reviewer agent

This is a spec document, not implementation — only change is one new file under specs/. The skill's job on merge is to auto-decompose it into issues, so the review bar is:

Each story has a clear As a <role>, I want …, so that … intro.
Each has acceptance criteria grouped by sub-section (per the operator's global CLAUDE.md conventions).
Each has out-of-scope + references + dependencies.
Execution order at the end.

No code-correctness to check; flag only structural issues that would make the breakdown dispatch produce garbage tickets.

After merge

Dispatch POST /breakdown against specs/m17-multi-repo-and-observability.md — boss runs A7, files 5 issues (one per M17-N story), applies area:* + type:user-story labels, attaches them to the new milestone.
Operator creates milestone 17 on Forgejo (no MCP tool; UI action).
Normal dispatch + parallel implementation.

Working document for the next milestone's five themes. Once merged, I'll dispatch the breakdown skill (A7, #147) against `specs/m17-multi-repo-and-observability.md` and have boss generate the user-story issues. ## Themes - **M17-1 Multi-repo** — lift the hardcoded `charles/claude-hooks` scope so the fleet serves `loom`, `proxmox-iac`, `ws-rpc-test` alongside the current repo. - **M17-2 Usage dashboard** — Anthropic-console-style panel for Pro Max weekly token consumption + thresholds. (Pro Max = subscription, not per-token, so "cost budget" becomes "weekly usage vs configurable threshold".) - **M17-3 Reviewer specialization** — `reviewer-security` (opus, match `area:security`) + `reviewer-fast` (haiku, match `type:chore`). Pure config + tests. - **M17-4 Design-reviewer end-to-end** — wire the idle `design-reviewer` into a real feedback loop on `area:dashboard` PRs with diff-scoped skill. - **M17-5 Model switch UI** — model dropdown in the edit-agent modal + PATCH endpoint. ## Review notes for the reviewer agent This is a **spec document**, not implementation — only change is one new file under `specs/`. The skill's job on merge is to auto-decompose it into issues, so the review bar is: - Each story has a clear `As a <role>, I want …, so that …` intro. - Each has acceptance criteria grouped by sub-section (per the operator's global CLAUDE.md conventions). - Each has out-of-scope + references + dependencies. - Execution order at the end. No code-correctness to check; flag only structural issues that would make the breakdown dispatch produce garbage tickets. ## After merge 1. Dispatch `POST /breakdown` against `specs/m17-multi-repo-and-observability.md` — boss runs A7, files 5 issues (one per M17-N story), applies `area:*` + `type:user-story` labels, attaches them to the new milestone. 2. Operator creates milestone 17 on Forgejo (no MCP tool; UI action). 3. Normal dispatch + parallel implementation.

claude-desktop added 1 commit

2026-04-20 14:29:12 +00:00

spec(m17): multi-repo + usage dashboard + specialization + design loop + model UI

qa / qa (pull_request) Successful in 3m4s

Details

qa / dockerfile (pull_request) Successful in 12s

Details

361b0c3e6b

Working document for the next milestone's five themes:

- M17-1 Multi-repo — lift the hardcoded charles/claude-hooks scope so the
  fleet can serve loom, proxmox-iac, ws-rpc-test alongside claude-hooks
- M17-2 Usage dashboard — Anthropic-console-style panel for Pro Max
  weekly token consumption + thresholds (color states)
- M17-3 Reviewer specialization — put A3's match_labels to use with
  reviewer-security (opus) + reviewer-fast (haiku) instances
- M17-4 Design-reviewer end-to-end — wire the idle design-reviewer into
  an actual feedback loop on area:dashboard PRs
- M17-5 Model switch UI — dropdown in the edit-agent modal, no SQLite
  surgery needed

Execution order + per-story acceptance criteria / out-of-scope / tests
spelled out so the breakdown skill (#147 / A7) has enough structure to
turn each section into an issue on dispatch.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

charles merged commit 13a7322d09 into main

2026-04-20 14:36:52 +00:00

charles deleted branch specs/m17-milestone

2026-04-20 14:36:52 +00:00

charles referenced this pull request from a commit

2026-04-20 14:36:52 +00:00

spec(m17): multi-repo + usage dashboard + specialization + design loop + model UI (#151)