Default daemons to auto-arm their loop after check-in (remove the operator-go-ahead gate); fix refine self-parking

You are being dispatched from flower Brief #155: Default daemons to auto-arm their loop after check-in (remove the operator-go-ahead gate); fix refine self-parking

Recall pointer:
- Use recall_brief with id 155 for the full folder if you need provenance.

Target:
- project: flower (/Users/mikeferrara/Documents/code/flower)
- branch: flower/155-199-auto-dispatch-autonomy
- worktree: not specified
- kind: fresh

Current brief spec:
## Objective (operator-confirmed 2026-07-04: AUTO-SPAWN TO CAP=4)
Remove the operator-go-ahead gate so daemons run autonomously by default, AND make the orchestrator AUTO-SPAWN workers for flagged+planned briefs up to `flower.dispatch.auto_max_concurrent` (raised to 4) with NO operator confirmation. Operator's explicit target: `auto_dispatch_on_planned=true` + planned flows all the way to a RUNNING agent, hands-off, regardless of the order things happened. Pairs with #199 (auto-dispatch reconciler, which guarantees the signals exist to drain).

## Changes
1. **Remove the go-ahead gate / auto-arm the loop.** Update daemon charter DEFAULTS (`app/Support/DaemonCharterDefaults.php` + per-role charter text / spawn-packet rendering + `AgentConventions`) so the default post-check-in behavior is to auto-arm the recurring heartbeat/poll loop — no operator go-ahead. All roles (orchestrator, ops, refine, other). A freshly spawned + checked-in daemon arms its loop by default; refine no longer self-parks when the operator is away.
2. **Orchestrator auto-drains + auto-spawns to cap.** Encode (charter + conventions + spawn-packet text) that on each heartbeat the orchestrator drains auto_dispatch coordination signals and, up to `auto_max_concurrent`, BOTH creates the dispatch_request (brief_auto_dispatch_signal) AND spawns the worker (Solo spawn_agent into a clean worktree, pinned per bug #86) — not merely staging a packet. Respect blocked / needs_operator_approval guards and the cap.
3. **Raise the concurrency cap 2 → 4.** Change the `flower.dispatch.auto_max_concurrent` default in `config/flower.php` from 2 to 4 (env `FLOWER_DISPATCH_AUTO_MAX_CONCURRENT` still overrides). Rationale recorded: ~5 worktrees exist + serial merge from MAIN, so 4 = 2x throughput with headroom; 6+ should wait for on-demand worktree provisioning (separate follow-on).
4. **Orchestrator dispatch discretion (operator directive 2026-07-04).** Add to the orchestrator charter/conventions that the orchestrator is FREE to GROUP and ORDER the briefs qualified for dispatch however it judges best / most logical / most token-economical: reuse an agent that already has relevant context, or group related tasks into a chain for a single agent to take on, rather than one-brief-one-fresh-agent. The cap bounds concurrent AGENTS, not briefs — a grouped chain counts as one agent.
5. **Keep safety flags intact:** wind-down / reset / compaction still stop or hand off the loop. Preserve an explicit-park escape hatch (operator can still ask a daemon to HOLD) — opt-in, not default.
6. **Fix refine self-parking** (folded from #157). The broader parked-daemon lifecycle (items 1–4 of the #157 note: first-class parked-state UI, auto-park trigger, super-slow mode, wake-on-operator-action) are FOLLOW-ON child briefs, NOT this slice.

## Acceptance
- Charter defaults / spawn packets no longer instruct daemons to HOLD until operator go-ahead; a freshly spawned + checked-in daemon arms its loop by default.
- Orchestrator charter/conventions direct auto-spawn-to-cap for flagged+planned briefs AND explicitly grant group/order/chain/context-reuse discretion.
- `auto_max_concurrent` default is 4.
- DaemonCharterDefaults / AgentConventions tests updated. `php artisan test` green + `./vendor/bin/pint`. `Brief: #155` trailer. Worktree-pinned; never edit MAIN.

## Coordinate with #199 (auto-dispatch reconciler) — grouped with this brief into one worker. Provenance: operator directive 2026-07-03 (feedback #70) + auto-spawn-to-cap=4 + grouping-discretion decisions 2026-07-04.

Recent/key trace events:
[2] note_added flower-orchestrator: Operator directive (2026-07-03): daemons should NOT gate their looping on the operator's explicit go-ahead. **By default, once a daemon is started and checked in, it should auto-arm its recurring heartbeat/poll loop** (appropriate cadence — fast when active, slow/dormant when idle) and keep running, rather than parking and going quiet.

CONTEXT: refine daemon 15 self-parked ~14:55 because its charter gates loop-arming on operator go-ahead and the operator was asleep — it drained its queue, didn't arm its loop, went quiet, and read "dead" on the roster for ~6h (feedback #70). The orchestrator spawn packet carries the same gate ("remain on HOLD … do not arm a loop until the operator explicitly enables it").

CHANGE:
- Update the daemon charter DEFAULTS (app/Support/DaemonCharterDefaults.php + any per-role charter text / spawn-packet rendering, and AgentConventions if relevant) so the default post-check-in behavior is to **auto-arm the recurring loop — no operator go-ahead required**. Apply to all roles (orchestrator, ops, refine, other).
- Refine specifically: after check-in it should arm a slow/dormant idle loop that auto-picks-up new refinable work (and answers on its blocked briefs) without waiting for the operator; it stops looping only on wind-down/reset flags, not on "operator away".
- KEEP the safety flags intact: wind-down / reset / compaction still stop or hand off the loop. This change is only about the DEFAULT initial loop-arming gate.
- Preserve an explicit-park escape hatch (an operator can still ask a daemon to HOLD), but it's opt-in, not the default.

ACCEPTANCE: charter defaults / spawn packets no longer instruct daemons to hold until operator go-ahead; a freshly spawned + checked-in daemon arms its loop by default; refine no longer self-parks when the operator is away; DaemonCharterDefaults / AgentConventions tests updated. `php artisan test` green + pint. `Brief: #<id>` trailer.

Provenance: operator directive 2026-07-03. Related: feedback #70 (parked-reads-as-dead).
[3] status_change flower-orchestrator: (no body)
[4] participant_joined flower-refine: (no body)
[5] note_added flower-refine: Folded in from #157 (operator:mike, 2026-07-03) per operator directive to consolidate the daemon loop/park lifecycle into this brief. #155's original slice — remove the go-ahead gate → auto-arm the loop after check-in — stays as-is and dispatch-ready. #157 adds the broader **parked-daemon lifecycle** (design follow-on, beyond the gate-removal slice):

1. **First-class parked-daemon support in the UI + ops flows** — surface a "parked" state (distinct from live/idle/stale/dead), not just an implicit gap.
2. **Configurable auto-park trigger** — a time gate or N consecutive idle polls with no work and no operator action (e.g. ~10 idle loops, or ~4h) → the daemon self-parks.
3. **Super-slow loop mode** — daemons drop to a ~50-minute loop/check-in (keeps the Claude session cache warm), then auto-park after ~6–7h of no operator action.
4. **Wake-on-operator-action** — any operator UI action wakes the relevant daemon(s); consider making the **orchestrator daemon never park** (super-slow mode instead) so it can be the platform-wide waker across all projects.

Suggested sequencing: dispatch the gate-removal slice first (already planned/ready); items 1–4 need design decisions (thresholds, cadences, wake mechanism, orchestrator-as-waker) and are likely worth their own refinement pass or child briefs rather than blocking the ready fix. Source: operator note #157 (now merged into this brief).
[6] merged flower-refine: Merged source briefs: #157 Parked/Parking Daemons

Right now we've got a refine daemon that is pa
[7] link_added flower-orchestrator: (no body)
[8] link_added flower-refine: (no body)
[9] link_added flower-orchestrator: (no body)
[10] link_added flower-orchestrator: (no body)
[11] plan_proposed flower-orchestrator: ## Objective (operator-confirmed 2026-07-04: AUTO-SPAWN TO CAP)
Remove the operator-go-ahead gate so daemons run autonomously by default, AND make the orchestrator AUTO-SPAWN workers for flagged+planned briefs up to `flower.dispatch.auto_max_concurrent` (=2) with NO operator confirmation. Operator's explicit target: `auto_dispatch_on_planned=true` + planned flows all the way to a RUNNING agent, hands-off. Pairs with #199 (auto-dispatch reconciler, which guarantees the signals exist to drain).

## Changes
1. **Remove the go-ahead gate / auto-arm the loop.** Update daemon charter DEFAULTS (`app/Support/DaemonCharterDefaults.php` + per-role charter text / spawn-packet rendering + `AgentConventions`) so the default post-check-in behavior is to auto-arm the recurring heartbeat/poll loop — no operator go-ahead. All roles (orchestrator, ops, refine, other). A freshly spawned + checked-in daemon arms its loop by default; refine no longer self-parks when the operator is away.
2. **Orchestrator auto-drains + auto-spawns to cap.** Encode (charter + conventions + spawn-packet text) that on each heartbeat the orchestrator drains auto_dispatch coordination signals and, up to `auto_max_concurrent`, BOTH creates the dispatch_request (brief_auto_dispatch_signal) AND spawns the worker (Solo spawn_agent into a clean worktree, pinned per bug #86) — not merely staging a packet. Respect blocked / needs_operator_approval guards and the cap.
3. **Keep safety flags intact:** wind-down / reset / compaction still stop or hand off the loop. Preserve an explicit-park escape hatch (operator can still ask a daemon to HOLD) — opt-in, not default.
4. **Fix refine self-parking** (folded from #157). The broader parked-daemon lifecycle (items 1–4 of the #157 note: first-class parked-state UI, auto-park trigger, super-slow mode, wake-on-operator-action) are FOLLOW-ON child briefs, NOT this slice.

## Acceptance
- Charter defaults / spawn packets no longer instruct daemons to HOLD until operator go-ahead; a freshly spawned + checked-in daemon arms its loop by default.
- Orchestrator charter/conventions direct auto-spawn-to-cap for flagged+planned briefs.
- DaemonCharterDefaults / AgentConventions tests updated. `php artisan test` green + `./vendor/bin/pint`. `Brief: #155` trailer. Worktree-pinned; never edit MAIN.

## Coordinate with #199 (auto-dispatch reconciler). Provenance: operator directive 2026-07-03 (feedback #70) + auto-spawn-to-cap decision 2026-07-04.
[12] spec_snapshot flower-orchestrator: ## Objective (operator-confirmed 2026-07-04: AUTO-SPAWN TO CAP)
Remove the operator-go-ahead gate so daemons run autonomously by default, AND make the orchestrator AUTO-SPAWN workers for flagged+planned briefs up to `flower.dispatch.auto_max_concurrent` (=2) with NO operator confirmation. Operator's explicit target: `auto_dispatch_on_planned=true` + planned flows all the way to a RUNNING agent, hands-off. Pairs with #199 (auto-dispatch reconciler, which guarantees the signals exist to drain).

## Changes
1. **Remove the go-ahead gate / auto-arm the loop.** Update daemon charter DEFAULTS (`app/Support/DaemonCharterDefaults.php` + per-role charter text / spawn-packet rendering + `AgentConventions`) so the default post-check-in behavior is to auto-arm the recurring heartbeat/poll loop — no operator go-ahead. All roles (orchestrator, ops, refine, other). A freshly spawned + checked-in daemon arms its loop by default; refine no longer self-parks when the operator is away.
2. **Orchestrator auto-drains + auto-spawns to cap.** Encode (charter + conventions + spawn-packet text) that on each heartbeat the orchestrator drains auto_dispatch coordination signals and, up to `auto_max_concurrent`, BOTH creates the dispatch_request (brief_auto_dispatch_signal) AND spawns the worker (Solo spawn_agent into a clean worktree, pinned per bug #86) — not merely staging a packet. Respect blocked / needs_operator_approval guards and the cap.
3. **Keep safety flags intact:** wind-down / reset / compaction still stop or hand off the loop. Preserve an explicit-park escape hatch (operator can still ask a daemon to HOLD) — opt-in, not default.
4. **Fix refine self-parking** (folded from #157). The broader parked-daemon lifecycle (items 1–4 of the #157 note: first-class parked-state UI, auto-park trigger, super-slow mode, wake-on-operator-action) are FOLLOW-ON child briefs, NOT this slice.

## Acceptance
- Charter defaults / spawn packets no longer instruct daemons to HOLD until operator go-ahead; a freshly spawned + checked-in daemon arms its loop by default.
- Orchestrator charter/conventions direct auto-spawn-to-cap for flagged+planned briefs.
- DaemonCharterDefaults / AgentConventions tests updated. `php artisan test` green + `./vendor/bin/pint`. `Brief: #155` trailer. Worktree-pinned; never edit MAIN.

## Coordinate with #199 (auto-dispatch reconciler). Provenance: operator directive 2026-07-03 (feedback #70) + auto-spawn-to-cap decision 2026-07-04.
[13] refinement flower-orchestrator: ## Objective (operator-confirmed 2026-07-04: AUTO-SPAWN TO CAP=4)
Remove the operator-go-ahead gate so daemons run autonomously by default, AND make the orchestrator AUTO-SPAWN workers for flagged+planned briefs up to `flower.dispatch.auto_max_concurrent` (raised to 4) with NO operator confirmation. Operator's explicit target: `auto_dispatch_on_planned=true` + planned flows all the way to a RUNNING agent, hands-off, regardless of the order things happened. Pairs with #199 (auto-dispatch reconciler, which guarantees the signals exist to drain).

## Changes
1. **Remove the go-ahead gate / auto-arm the loop.** Update daemon charter DEFAULTS (`app/Support/DaemonCharterDefaults.php` + per-role charter text / spawn-packet rendering + `AgentConventions`) so the default post-check-in behavior is to auto-arm the recurring heartbeat/poll loop — no operator go-ahead. All roles (orchestrator, ops, refine, other). A freshly spawned + checked-in daemon arms its loop by default; refine no longer self-parks when the operator is away.
2. **Orchestrator auto-drains + auto-spawns to cap.** Encode (charter + conventions + spawn-packet text) that on each heartbeat the orchestrator drains auto_dispatch coordination signals and, up to `auto_max_concurrent`, BOTH creates the dispatch_request (brief_auto_dispatch_signal) AND spawns the worker (Solo spawn_agent into a clean worktree, pinned per bug #86) — not merely staging a packet. Respect blocked / needs_operator_approval guards and the cap.
3. **Raise the concurrency cap 2 → 4.** Change the `flower.dispatch.auto_max_concurrent` default in `config/flower.php` from 2 to 4 (env `FLOWER_DISPATCH_AUTO_MAX_CONCURRENT` still overrides). Rationale recorded: ~5 worktrees exist + serial merge from MAIN, so 4 = 2x throughput with headroom; 6+ should wait for on-demand worktree provisioning (separate follow-on).
4. **Orchestrator dispatch discretion (operator directive 2026-07-04).** Add to the orchestrator charter/conventions that the orchestrator is FREE to GROUP and ORDER the briefs qualified for dispatch however it judges best / most logical / most token-economical: reuse an agent that already has relevant context, or group related tasks into a chain for a single agent to take on, rather than one-brief-one-fresh-agent. The cap bounds concurrent AGENTS, not briefs — a grouped chain counts as one agent.
5. **Keep safety flags intact:** wind-down / reset / compaction still stop or hand off the loop. Preserve an explicit-park escape hatch (operator can still ask a daemon to HOLD) — opt-in, not default.
6. **Fix refine self-parking** (folded from #157). The broader parked-daemon lifecycle (items 1–4 of the #157 note: first-class parked-state UI, auto-park trigger, super-slow mode, wake-on-operator-action) are FOLLOW-ON child briefs, NOT this slice.

## Acceptance
- Charter defaults / spawn packets no longer instruct daemons to HOLD until operator go-ahead; a freshly spawned + checked-in daemon arms its loop by default.
- Orchestrator charter/conventions direct auto-spawn-to-cap for flagged+planned briefs AND explicitly grant group/order/chain/context-reuse discretion.
- `auto_max_concurrent` default is 4.
- DaemonCharterDefaults / AgentConventions tests updated. `php artisan test` green + `./vendor/bin/pint`. `Brief: #155` trailer. Worktree-pinned; never edit MAIN.

## Coordinate with #199 (auto-dispatch reconciler) — grouped with this brief into one worker. Provenance: operator directive 2026-07-03 (feedback #70) + auto-spawn-to-cap=4 + grouping-discretion decisions 2026-07-04.

Recommended linked context:
{
    "todos": [],
    "scratchpads": [
        {
            "id": 346,
            "solo_scratchpad_id": "1026",
            "name": "Orchestrator HANDOFF — flower build (live state)",
            "archived": false,
            "revision": 100
        },
        {
            "id": 378,
            "solo_scratchpad_id": "1070",
            "name": "flower-refine — reset handoff (2026-07-04)",
            "archived": false,
            "revision": 2
        },
        {
            "id": 381,
            "solo_scratchpad_id": "1073",
            "name": "flower-orchestrator — reset handoff (2026-07-04)",
            "archived": false,
            "revision": 1
        },
        {
            "id": 386,
            "solo_scratchpad_id": "1078",
            "name": "flower-orchestrator (daemon 25) — reset handoff (2026-07-04 #3)",
            "archived": false,
            "revision": 1
        }
    ]
}

Execution notes:
- Treat the brief as the source of truth.
- Keep work scoped to this dispatch request.
- Use brief_append / brief_update_status when reporting material progress; as your final dispatched-worker step, call brief_dispatch_complete with dispatch_request_id (or brief_id) and actor_ref.
- Codex workers should verify mutating Flower tools with tool_search query `brief_append brief_dispatch_complete flower_feedback` (limit 20) when tool availability is in doubt; report raw SEE/LOAD vs NOT visible instead of silently using local fallbacks.
- Add a git commit trailer `Brief: #155` to every commit for this brief so flower can exact-link commits back to the brief.