What do you think about something like a 'deep review' of a project or

You are being dispatched from flower Brief #41: What do you think about something like a 'deep review' of a project or

Recall pointer:
- Use recall_brief with id 41 for the full folder if you need provenance.

Target:
- project: flower (/Users/mikeferrara/Documents/code/flower)
- branch: choose an appropriate branch
- worktree: not specified
- kind: fresh

Current brief spec:
## Goal
A **"Deep Review"** action that dispatches a high-effort agent to genuinely dig into a project — reading real session transcripts (not just summaries), verifying them, recalling details — and produce an appraisal: current state, actionable items, and **proposed briefs** for the operator to confirm. For picking up a project/worktree that's gone stale.

## Scope — v1 (project-level)
Rides on existing infra: the prompt library (#17), dispatch, epic/child briefs (#31), and the debug recall tools.

1. **`deep_review` prompt template** — a new `kind` in `prompt_templates` (editable at `/prompts`; the table already has a `kind` column + `{{placeholders}}`): the charter telling the agent to use `recall_resume` / `recall_search` / `recall_open_loops` to pull recent sessions, **verify summaries against raw transcripts** (`flower_session_raw`), assess state, and emit a structured appraisal + a list of proposed briefs.
2. **"Deep Review" button** on `/projects/{slug}` → creates a brief (origin=`deep_review`, scoped to the project) populated from the template, and dispatches it via `DispatchService`.
   - ⚠️ **GROUNDING CORRECTION (2026-07-03): dispatch cannot force a model/effort.** `DispatchService` selects only a **harness (`agent_tool`)** and renders a text packet; `brief.model` is consumed **only by `BriefRefiner`** (the in-process refine LLM), never by dispatch/spawn. So "high-effort model via the brief's `model` override" is **not achievable today.** v1 therefore: dispatch to a **high-capability harness** (Claude/Codex) and have the `deep_review` charter explicitly request high effort. **True per-dispatch model/effort forcing is a follow-up that depends on #142 (model routing).**
3. **Result capture + indexing (Q1 resolved = indexed).** The agent writes its appraisal back onto the brief (spec/trace).
   - ✅ **GROUNDING (2026-07-03): simpler than the prior spec thought — briefs are ALREADY a chunkable + embedded recall type** (`Brief morphMany Chunk`; `EmbedChunks::buildBriefChunks()`; `RecallService` treats briefs as searchable hits). So an appraisal written onto the brief is **auto-indexed by existing infra** and becomes `recall_search`-able — **no new chunk pipeline needed** (an optional distinct `deep_review` chunk tag is nice-to-have, not required).
   - Proposed follow-up briefs → **child briefs** (epic grouping, #31) — only on **operator confirmation** (proposed, not auto-created).

## Resolved decisions (operator, 2026-07-02)
- **Q1 Persistence → indexed into the recall corpus** — satisfied by the existing brief chunk+embed path (above).
- **Q2 Scope → project-level only for v1.** Single/selected-session scope is **v2 = brief #102** (already created; depends on #41, stays blocked until v1 ships/verifies).

## Acceptance
- `/projects/{slug}` has a "Deep Review" action → creates + dispatches a `deep_review` brief (**high-capability harness + high-effort charter**; not a forced model — see correction).
- The agent's appraisal lands back on the brief AND is searchable via `recall_search` (rides the existing brief chunk+embed path).
- Proposed follow-ups are created as child briefs only on operator confirmation.
- `deep_review` template editable in `/prompts`. Tests (button creates+dispatches; template renders; appraisal indexed+searchable; child-brief-on-confirm). `php artisan test` green + pint. `Brief: #41` trailer.

## Provenance
Operator note (#41, 2026-07-01). Refined by flower-design (2026-07-01); Q1/Q2 answered by operator (2026-07-02); moved to planned by flower-refine (2026-07-03) — **grounding-corrected 2026-07-03: model-forcing isn't available at dispatch (depends on #142); appraisal indexing reuses the existing brief chunk path (no new pipeline).** v2 = #102 (chained). Rides #17 (prompt library), #31 (epic/child briefs), and dispatch.

Recent/key trace events:
[11] operator_answer operator:mike: project-level only (v1) - but plan/spec v2/p2 as a follow up/chained brief
[12] participant_joined flower-refine: (no body)
[13] spec_snapshot flower-refine: ## Goal
A **"Deep Review"** action that dispatches a high-effort agent to genuinely dig into a project —
reading real session transcripts (not just summaries), verifying them, recalling details — and
produce an appraisal: current state, actionable items, and **proposed briefs** for the operator to
confirm. For picking up a project/worktree that's gone stale.

## Scope — v1 proposal (project-level)
Rides on existing infra: the prompt library (#17), dispatch, epic/child briefs (#31), and the
debug recall tools.

1. **`deep_review` prompt template** (a new kind in the prompt library, editable at /prompts): the
   charter that tells the agent to use `recall_resume` / `recall_search` / `recall_open_loops` to
   pull recent sessions, **verify summaries against raw transcripts** (`flower_session_raw`), assess
   state, and emit a structured appraisal + a list of proposed briefs.
2. **"Deep Review" button** on `/projects/{slug}` → creates a brief (origin=`deep_review`, scoped to
   the project) populated from the template, and dispatches it via `DispatchService` (high-effort
   model via the brief's `model` override).
3. **Result capture.** The agent writes its appraisal back onto the brief (spec/trace). Proposed
   briefs are created as **child briefs** (epic grouping, #31) — but only on **operator
   confirmation** (proposed, not auto-created). Persistence of the appraisal itself is the open
   question below.

v1 = **project-level**. Single-session / selected-sessions scope is noted for v2.

## Open questions (BLOCKING — see brief_ask; brief stays `refining`)
- Q1 — should a Deep Review's output be **ephemeral** (lives only on the brief) or
  **captured/indexed/embedded** into the recall corpus (so future `recall_search` finds it)?
  (Operator asked this directly in the note.)
- Q2 — v1 scope: project-level only, or also single-session / selected-sessions from the start.

## Acceptance (once Q1/Q2 resolved)
- `/projects/{slug}` has a "Deep Review" action → creates + dispatches a `deep_review` brief with the
  templated charter (high-effort model).
- The agent's appraisal lands back on the brief; proposed follow-ups are created as child briefs only
  on operator confirmation.
- `deep_review` template is editable in /prompts. Tests (button creates+dispatches; template renders;
  child-brief-on-confirm). `php artisan test` green + pint. `Brief: #41` trailer.

## Provenance
Operator note (#41, 2026-07-01). Refined by flower-design (2026-07-01) into a v1 riding on #17
(prompt library), #31 (epic/child briefs), and dispatch. Ephemeral-vs-indexed + scope posted as
questions.
[14] refinement flower-refine: ## Goal
A **"Deep Review"** action that dispatches a high-effort agent to genuinely dig into a project —
reading real session transcripts (not just summaries), verifying them, recalling details — and
produce an appraisal: current state, actionable items, and **proposed briefs** for the operator to
confirm. For picking up a project/worktree that's gone stale.

## Scope — v1 (project-level)
Rides on existing infra: the prompt library (#17), dispatch, epic/child briefs (#31), and the
debug recall tools.

1. **`deep_review` prompt template** (a new kind in the prompt library, editable at /prompts): the
   charter that tells the agent to use `recall_resume` / `recall_search` / `recall_open_loops` to
   pull recent sessions, **verify summaries against raw transcripts** (`flower_session_raw`), assess
   state, and emit a structured appraisal + a list of proposed briefs.
2. **"Deep Review" button** on `/projects/{slug}` → creates a brief (origin=`deep_review`, scoped to
   the project) populated from the template, and dispatches it via `DispatchService` (high-effort
   model via the brief's `model` override).
3. **Result capture + indexing (RESOLVED — Q1).** The agent writes its appraisal back onto the
   brief (spec/trace) AND the appraisal is **captured/indexed/embedded into the recall corpus** so
   future `recall_search` finds it. Treat the appraisal as a first-class chunkable artifact — mirror
   how briefs / session_segments are chunked + embedded (a new chunk source/tag such as
   `deep_review`, or reuse the brief chunking path) so it is searchable. Proposed follow-up briefs are
   created as **child briefs** (epic grouping, #31) — but only on **operator confirmation** (proposed,
   not auto-created).

## Resolved decisions (operator, 2026-07-02)
- **Q1 Persistence → indexed into the recall corpus** (not ephemeral). The appraisal must be
  embedded/searchable.
- **Q2 Scope → project-level only for v1.** Single-session / selected-sessions scope is a **v2**, to
  be spec'd as a **chained follow-up brief** blocked on v1. [REQUIRED follow-up — see below.]

## Required follow-up (operator-requested)
- Create a **chained v2 brief** — "Deep Review v2 — single-session / selected-sessions scope" — set to
  **depend on this brief (#41)** so it stays blocked until v1 ships/verifies. (Not yet created —
  pending operator/orchestrator decomposition; flower-refine flagged this to the operator.)

## Acceptance
- `/projects/{slug}` has a "Deep Review" action → creates + dispatches a `deep_review` brief with the
  templated charter (high-effort model).
- The agent's appraisal lands back on the brief AND is indexed/embedded into the recall corpus
  (searchable via `recall_search`).
- Proposed follow-ups are created as child briefs only on operator confirmation.
- `deep_review` template is editable in /prompts. Tests (button creates+dispatches; template renders;
  appraisal indexed+searchable; child-brief-on-confirm). `php artisan test` green + pint.
  `Brief: #41` trailer.

## Provenance
Operator note (#41, 2026-07-01). Refined by flower-design (2026-07-01); blocking Qs (Q1 persistence,
Q2 scope) answered by operator (2026-07-02); answers folded in + moved to planned by flower-refine
(2026-07-03). Rides on #17 (prompt library), #31 (epic/child briefs), and dispatch.
[15] status_change flower-refine: (no body)
[16] link_added flower-refine: (no body)
[17] participant_joined flower-orchestrator: (no body)
[18] status_change flower-orchestrator: (no body)
[19] link_added flower-orchestrator: (no body)
[20] spec_snapshot flower-refine: ## Goal
A **"Deep Review"** action that dispatches a high-effort agent to genuinely dig into a project —
reading real session transcripts (not just summaries), verifying them, recalling details — and
produce an appraisal: current state, actionable items, and **proposed briefs** for the operator to
confirm. For picking up a project/worktree that's gone stale.

## Scope — v1 (project-level)
Rides on existing infra: the prompt library (#17), dispatch, epic/child briefs (#31), and the
debug recall tools.

1. **`deep_review` prompt template** (a new kind in the prompt library, editable at /prompts): the
   charter that tells the agent to use `recall_resume` / `recall_search` / `recall_open_loops` to
   pull recent sessions, **verify summaries against raw transcripts** (`flower_session_raw`), assess
   state, and emit a structured appraisal + a list of proposed briefs.
2. **"Deep Review" button** on `/projects/{slug}` → creates a brief (origin=`deep_review`, scoped to
   the project) populated from the template, and dispatches it via `DispatchService` (high-effort
   model via the brief's `model` override).
3. **Result capture + indexing (RESOLVED — Q1).** The agent writes its appraisal back onto the
   brief (spec/trace) AND the appraisal is **captured/indexed/embedded into the recall corpus** so
   future `recall_search` finds it. Treat the appraisal as a first-class chunkable artifact — mirror
   how briefs / session_segments are chunked + embedded (a new chunk source/tag such as
   `deep_review`, or reuse the brief chunking path) so it is searchable. Proposed follow-up briefs are
   created as **child briefs** (epic grouping, #31) — but only on **operator confirmation** (proposed,
   not auto-created).

## Resolved decisions (operator, 2026-07-02)
- **Q1 Persistence → indexed into the recall corpus** (not ephemeral). The appraisal must be
  embedded/searchable.
- **Q2 Scope → project-level only for v1.** Single-session / selected-sessions scope is a **v2**, to
  be spec'd as a **chained follow-up brief** blocked on v1. [REQUIRED follow-up — see below.]

## Required follow-up (operator-requested)
- Create a **chained v2 brief** — "Deep Review v2 — single-session / selected-sessions scope" — set to
  **depend on this brief (#41)** so it stays blocked until v1 ships/verifies. (Not yet created —
  pending operator/orchestrator decomposition; flower-refine flagged this to the operator.)

## Acceptance
- `/projects/{slug}` has a "Deep Review" action → creates + dispatches a `deep_review` brief with the
  templated charter (high-effort model).
- The agent's appraisal lands back on the brief AND is indexed/embedded into the recall corpus
  (searchable via `recall_search`).
- Proposed follow-ups are created as child briefs only on operator confirmation.
- `deep_review` template is editable in /prompts. Tests (button creates+dispatches; template renders;
  appraisal indexed+searchable; child-brief-on-confirm). `php artisan test` green + pint.
  `Brief: #41` trailer.

## Provenance
Operator note (#41, 2026-07-01). Refined by flower-design (2026-07-01); blocking Qs (Q1 persistence,
Q2 scope) answered by operator (2026-07-02); answers folded in + moved to planned by flower-refine
(2026-07-03). Rides on #17 (prompt library), #31 (epic/child briefs), and dispatch.
[21] refinement flower-refine: ## Goal
A **"Deep Review"** action that dispatches a high-effort agent to genuinely dig into a project — reading real session transcripts (not just summaries), verifying them, recalling details — and produce an appraisal: current state, actionable items, and **proposed briefs** for the operator to confirm. For picking up a project/worktree that's gone stale.

## Scope — v1 (project-level)
Rides on existing infra: the prompt library (#17), dispatch, epic/child briefs (#31), and the debug recall tools.

1. **`deep_review` prompt template** — a new `kind` in `prompt_templates` (editable at `/prompts`; the table already has a `kind` column + `{{placeholders}}`): the charter telling the agent to use `recall_resume` / `recall_search` / `recall_open_loops` to pull recent sessions, **verify summaries against raw transcripts** (`flower_session_raw`), assess state, and emit a structured appraisal + a list of proposed briefs.
2. **"Deep Review" button** on `/projects/{slug}` → creates a brief (origin=`deep_review`, scoped to the project) populated from the template, and dispatches it via `DispatchService`.
   - ⚠️ **GROUNDING CORRECTION (2026-07-03): dispatch cannot force a model/effort.** `DispatchService` selects only a **harness (`agent_tool`)** and renders a text packet; `brief.model` is consumed **only by `BriefRefiner`** (the in-process refine LLM), never by dispatch/spawn. So "high-effort model via the brief's `model` override" is **not achievable today.** v1 therefore: dispatch to a **high-capability harness** (Claude/Codex) and have the `deep_review` charter explicitly request high effort. **True per-dispatch model/effort forcing is a follow-up that depends on #142 (model routing).**
3. **Result capture + indexing (Q1 resolved = indexed).** The agent writes its appraisal back onto the brief (spec/trace).
   - ✅ **GROUNDING (2026-07-03): simpler than the prior spec thought — briefs are ALREADY a chunkable + embedded recall type** (`Brief morphMany Chunk`; `EmbedChunks::buildBriefChunks()`; `RecallService` treats briefs as searchable hits). So an appraisal written onto the brief is **auto-indexed by existing infra** and becomes `recall_search`-able — **no new chunk pipeline needed** (an optional distinct `deep_review` chunk tag is nice-to-have, not required).
   - Proposed follow-up briefs → **child briefs** (epic grouping, #31) — only on **operator confirmation** (proposed, not auto-created).

## Resolved decisions (operator, 2026-07-02)
- **Q1 Persistence → indexed into the recall corpus** — satisfied by the existing brief chunk+embed path (above).
- **Q2 Scope → project-level only for v1.** Single/selected-session scope is **v2 = brief #102** (already created; depends on #41, stays blocked until v1 ships/verifies).

## Acceptance
- `/projects/{slug}` has a "Deep Review" action → creates + dispatches a `deep_review` brief (**high-capability harness + high-effort charter**; not a forced model — see correction).
- The agent's appraisal lands back on the brief AND is searchable via `recall_search` (rides the existing brief chunk+embed path).
- Proposed follow-ups are created as child briefs only on operator confirmation.
- `deep_review` template editable in `/prompts`. Tests (button creates+dispatches; template renders; appraisal indexed+searchable; child-brief-on-confirm). `php artisan test` green + pint. `Brief: #41` trailer.

## Provenance
Operator note (#41, 2026-07-01). Refined by flower-design (2026-07-01); Q1/Q2 answered by operator (2026-07-02); moved to planned by flower-refine (2026-07-03) — **grounding-corrected 2026-07-03: model-forcing isn't available at dispatch (depends on #142); appraisal indexing reuses the existing brief chunk path (no new pipeline).** v2 = #102 (chained). Rides #17 (prompt library), #31 (epic/child briefs), and dispatch.
[22] status_change flower-refine: (no body)

Recommended linked context:
{
    "todos": [],
    "scratchpads": [
        {
            "id": 346,
            "solo_scratchpad_id": "1026",
            "name": "Orchestrator HANDOFF — flower build (live state)",
            "archived": false,
            "revision": 84
        },
        {
            "id": 364,
            "solo_scratchpad_id": "1055",
            "name": "flower-refine — reset handoff (2026-07-03)",
            "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: #41` to every commit for this brief so flower can exact-link commits back to the brief.