feat: dashboard Bundles section for Fable-orchestrated multi-task sessions

[Jammy2211]

Overview

The PyAutoMind dashboard lists every task individually, which is right for one-at-a-time work. As more work moves to a model where a Fable session orchestrates several related tasks with Opus subagents, the dashboard needs an additional Bundles view: sets of independent prompts that make sense in one orchestrated session. Bundles are distinct from epics (ordered, phase-gated); every task still appears in its usual section, and bundles refresh on the existing nightly dashboard_refresh.yml.

Plan

• Add PyAutoMind/bundles.md registry (human-pinned bundles only, epics.md-style schema) and document the optional Bundle: <slug> prompt header in REFERENCE.md.
• Parse bundles.md + Bundle: headers in _intake.py; add deterministic render-time auto-bundling of the draft pool (same Target, exclusions, size cap, min 2 members).
• Render a new "Bundles" H2 in dashboard.md and dashboard.html between Backlog and Recent — pinned first, then auto — each with a members table, total size and a one-tap Fable orchestration prompt (copy button in HTML). Existing sections untouched.
• Add a start_bundle skill body in PyAutoBrain/skills describing the orchestration contract (one issue per member, shared worktree per repo, per-task PRs so /prm works unchanged, Fable plans / Opus executes) plus a WORKFLOW.md paragraph.
• Regenerate the dashboard via pyauto-brain intake --apply dashboard; the nightly needs no change beyond adding bundles.md to its path triggers.

Detailed implementation plan

Affected Repositories

• PyAutoBrain (primary)
• PyAutoMind

Branch Survey

Repository	Current Branch	Dirty?
./PyAutoBrain	main	clean
./PyAutoMind	main	clean (prompt file only)

Suggested branch: feature/dashboard-bundles

Implementation Steps

1. PyAutoBrain/agents/conductors/intake/_intake.py
   • Extend the light-header regex (~L453) with Bundle.
   • Add parse_bundles(mind) beside parse_epics (~L569): ## <slug> then - title: / - members: (prompt paths) / - rationale: / - status:.
   • Add auto_bundles(census): group draft/ prompts by Target; drop Epic: members, unresolved Blocked-by:, Autonomy: human-required, Difficulty: too-large, and members of pinned bundles; pack greedily under a size cap (≤ 1 large + 3 small, or ≤ 4 medium); minimum 2 members; deterministic ordering (priority desc, path asc). Returns {slug, title, members, total, origin: "auto"} — render-only, never written back.
   • Add bundle_prompt(bundle) producing the Fable orchestration prompt: read each member prompt, run start_dev per member, one shared worktree per repo, delegate implementation to Opus subagents, per-task PRs.
   • Render in render_dashboard() (~L1111, after Backlog, before Recent) and render_dashboard_html() (~L1348) with copy buttons; pinned before auto. Warn on a Bundle: slug absent from bundles.md, mirroring the epic warning (~L1277).
2. PyAutoMind: create bundles.md (intro + schema, no entries initially); add Bundle: to the header table in REFERENCE.md (~L287-351); add bundles.md to .github/workflows/dashboard_refresh.yml path triggers (L28-49).
3. PyAutoBrain/skills/start_bundle/ (start_bundle.md + SKILL.md, symlinked into ~/.claude/commands like siblings) + a paragraph in skills/WORKFLOW.md under the delegation ladder.
4. Tests (PyAutoBrain intake tests): fixture Mind with one pinned bundle, auto-bundle-able drafts and each excluded case; assert grouping, exclusions, size cap, determinism, unknown-slug warning, dashboard --check idempotence.
5. Regenerate dashboard.md/dashboard.html and commit to Mind.

Key Files

• PyAutoBrain/agents/conductors/intake/_intake.py — census, epics parsing, both renderers
• PyAutoMind/epics.md — schema to mirror
• PyAutoMind/REFERENCE.md — prompt header documentation
• PyAutoMind/.github/workflows/dashboard_refresh.yml — nightly regen + self-heal
• PyAutoBrain/skills/WORKFLOW.md — delegation ladder

Trade-offs

• Render-only auto bundles keep nightly bot commits noise-free; only human pins are persisted.
• Deterministic rules over LLM grouping for reproducibility; an optional Claude refinement pass is deferred to a phase 2.

Original Prompt

Click to expand starting prompt

Dashboard "Bundles" — grouping tasks for Fable-orchestrated multi-task sessions

Type: feature
Target: PyAutoMind
Repos:

• PyAutoMind
• PyAutoBrain
  Difficulty: medium
  Autonomy: supervised
  Priority: high
  Status: formalised
  Filed: 2026-08-27

Original request (verbatim)

I am slowly moving to a model where Fable orchestrates larger tasks with opus.
I therefore think we need a grouping mechanism on the pyautomind dashboard,
where similar tasks which make sense to be done in on go with fable as an
orchestrator make sense. This should not change the dashboard as it is, seeing
all tasks listed individually is still highly valueable and makes sense when I
want to do one at a time. I think we need a group section, and a sensible
mechanism which will update the grouped tasks regularly (every night?).

Approved design (2026-08-27)

A bundle is a set of independent prompts that make sense in one Fable
session, with Opus subagents executing the members. Distinct from an epic
(ordered, phase-gated, one phase at a time). Every task still appears in its
usual dashboard section — a bundle is an additional view, never a replacement.

1. PyAutoMind/bundles.md registry — holds only human-pinned bundles,
   schema like epics.md: ## <slug> then - title: / - members: (prompt
   paths) / - rationale: / - status:. Optional Bundle: <slug> prompt
   header mirrors Epic: so a human can force membership. Nothing ever
   rewrites prompt files automatically.
2. Render-only auto-bundling in
   PyAutoBrain/agents/conductors/intake/_intake.py — deterministic, computed
   at dashboard-render time, never written to the registry. Rules: same
   Target repo; exclude Epic: members, unresolved Blocked-by:,
   Autonomy: human-required, Difficulty: too-large; size cap (≤ 1 large +
   3 small, or ≤ 4 medium); minimum 2 members; pinned members leave the auto
   pool. Refresh rides the existing nightly dashboard_refresh.yml — no new
   schedule.
3. New "Bundles" H2 in dashboard.md + dashboard.html, placed between
   Backlog and Recent. One card per bundle (pinned first, then auto): members
   table (size / priority / status), total size, and a one-tap Fable
   orchestration prompt with copy button (HTML), analogous to the epic
   resume prompt. Existing sections unchanged.
4. Brain orchestration contract — start_bundle skill (or
   start_dev --bundle <slug>) describing: one issue per member (keeps the
   no-bulk-issue rule), shared worktree per repo, per-task PRs so /prm works
   unchanged, Fable plans / Opus executes. Plus a paragraph in
   PyAutoBrain/skills/WORKFLOW.md.

Deferred (phase 2, only if auto bundles feel dumb): an optional Claude
refinement pass over the deterministic proposals.

[Jammy2211]

Library PR Created

• PyAutoBrain: feat(intake): dashboard Bundles section + start_bundle skill #310 (renderer, start_bundle skill, 22 tests — 84/84 intake, 601/601 full suite)
• PyAutoMind: mind: bundles.md registry + dashboard Bundles section PyAutoMind#365 (bundles.md, REFERENCE, workflow triggers, regenerated dashboard — 276/276)

Merge order: Brain #310 first — the Mind PR's dashboard freshness check renders with Brain main, so it stays red until the renderer lands.

API changes: none. Prompt headers gain an optional Bundle: <slug> key.

Heart was RED for unrelated reasons (release integrate failure, shared_preloads.py timeout, hook-manifest drift, stale PyAutoFit PR); shipped on explicit human acknowledgement. Next: /prm per PR.

[Jammy2211]

Shipped

• PyAutoBrain feat(intake): dashboard Bundles section + start_bundle skill #310 merged (8b60255a) — Bundles renderer, start_bundle skill, 22 tests.
• PyAutoMind #365 merged (efdca45b) — bundles.md registry, docs, workflow triggers, regenerated dashboard.

The dashboard now carries a Bundles section (top 8 of 20 auto proposals; pin any in bundles.md). Phase 2 (optional Claude refinement of auto proposals) deferred until the rules-based bundles have been lived with.