v0.9.0 execution board: lane order, dependencies, and agent protocol (canonical packet)

[Hmbown]

v0.8.68 execution board (canonical agent packet)

This is the single entry point for any agent working the v0.8.68 milestone. It replaces the July 7 triage packet (preserved in this issue's edit history). Every open milestone issue now carries exactly one lane-* label, so lanes are queryable directly.

Ground rules (read before touching code)

1. main is the implementation source of truth. Branch every fix from main as codex/v0868-fix-<issue>. PR 0.8.68 train: workflow correctness, TUI stability, modes & permissions, security hardening #4099 (quick-win cutover) is already merged.
2. codex/0868-next is a stale reference only. Never develop on it. Cherry-pick a specific commit from it only when an issue explicitly needs one.
3. Defer v0.8.69 refactors unless they directly unblock a stopship issue (v0.9.0 TUI: repeated Ctrl+C re-prompts instead of exiting in PTY/raw-mode path #4090, Fleet setup modal is provider-scoped instead of a role/profile roster editor #4093, Sub-agent detail panel is empty and can freeze the TUI during active work #4094).
4. Guard issue v0.9.0 guard: preserve verified active modules and DEEPSEEK env names #4172 applies to every lane: preserve verified active modules and DEEPSEEK* env var names. Check it before deleting or renaming anything.
5. Do not push to main, tag, release, force-push, or close issues without explicit approval. Close issues only after the fixing PR is merged (Fixes #N in the PR body).
6. Preserve contributor credit on harvested work (Co-authored-by for humans only, per .github/AUTHOR_MAP; note agent assistance in plain commit body).
7. Run the verification gate before opening any PR (commands at the bottom).

How to pick work (agent protocol)

1. Query ready work: gh issue list -R Hmbown/CodeWhale --label agent-ready --state open — optionally add your lane label, e.g. --label lane-catalog.
2. agent-ready means unblocked now. Issues without it are blocked on a start condition (see lane table) — do not start them.
3. When you start an issue: gh issue edit <N> --add-label agent-in-progress --remove-label agent-ready. One issue per agent at a time; skip anything already agent-in-progress.
4. Work the lane checklists below top to bottom — the order encodes dependencies.
5. Open a PR per issue (or per small coherent batch within one lane), referencing Fixes #<N>.
6. When you stop, comment on this issue with: branch, HEAD, issues touched, tests run, anything left local-only.

Lanes

Lane	Label	Start condition	Workflow file
Stopship	lane-stopship	Active now (agents in flight — do not double-assign)	workflows/v0868_stopship_lane.workflow.js
Catalog / Models.dev	lane-catalog	Ready now — file-disjoint from stopship	workflows/v0868_catalog_lane.workflow.js
Transcript copy polish	lane-copy	Ready now — copy/strings; scout transcript/sidebar overlap with #4094 first	workflows/v0868_tui_copy_lane.workflow.js
Fleet / AgentProfile	lane-fleet	After #4093 merges (same Fleet setup surface)	per-issue workflow
Workflow runtime + UI	lane-workflow	After stopship green (overlaps #4094 surfaces)	workflows/v0868_workflow_ui_lane.workflow.js
Turn Inspector	lane-inspector	After #4094 merges (shares ui.rs/detail panel code)	per-issue workflow
Performance	lane-perf	After stopship green (mechanical edits across hot paths — rebase hazard)	per-issue workflow
Setup / onboarding	lane-setup	After stopship green	per-issue workflow
Meta / guards	lane-meta	Always in effect	—

When a lane's start condition is met, promote its issues by adding agent-ready (top-of-checklist first).

Lane order and dependencies

Stopship (lane-stopship) — ACTIVE, agents already assigned

• v0.9.0 TUI: repeated Ctrl+C re-prompts instead of exiting in PTY/raw-mode path #4090 — Ctrl+C re-prompt in PTY/raw-mode (needs key-path regression test)
• Fleet setup modal is provider-scoped instead of a role/profile roster editor #4093 — Fleet setup modal → role/profile roster editor
• Sub-agent detail panel is empty and can freeze the TUI during active work #4094 — sub-agent detail panel empty / TUI freeze
• v0.9.0 UX: make API-key onboarding copy show the effective config path under CODEWHALE_HOME #3986 — API-key onboarding copy shows effective config path (dogfood follow-up, after the three above)
• v0.9.0 UX: slash autocomplete repeats aliases already shown in the command label #3990 — slash autocomplete alias duplication (dogfood follow-up)

Catalog / Models.dev (lane-catalog) — ready now, work in this order

Foundation (Models.dev as source of truth, umbrella #4184):

• Accept current live Models.dev schema in the catalog parser #4185 — accept current live Models.dev schema in the parser (everything below builds on this)
• Fetch and cache live Models.dev catalog into ProviderLake #4187 — fetch + cache live catalog into ProviderLake
• Normalize Models.dev provider IDs onto CodeWhale provider kinds #4186 — normalize Models.dev provider IDs onto CodeWhale provider kinds
• Demote curated bundled model data after Models.dev live catalog lands #4188 — demote curated bundled model data to fallback
• Use Models.dev as the source of truth for provider and model metadata #4184 — umbrella; close when the four above are merged

Picker wiring (after #4187):

• v0.9.0 3.1: wire live catalog cache into provider and model pickers #4139 — wire live catalog cache into provider/model pickers
• v0.9.0 3.2: show configured providers and models by default #4140 — show configured providers/models by default
• v0.9.0 3.3: add cross-field search to both provider and model pickers #4141 — cross-field search in both pickers
• v0.9.0 1.3: add model picker catalog views and metadata rows #4115 — catalog views + metadata rows in model picker
• v0.9.0 UX: model picker shows awkward single-row range titles #3995 — fix awkward single-row range titles in model picker

Plumbing and closeout:

• v0.9.0 1.4: migrate legacy model source consumers to ProviderLake #4116 — migrate legacy model-source consumers to ProviderLake
• v0.9.0 1.2: catalog refresh policy for live model metadata #4114 — catalog refresh policy
• v0.9.0 1.5: add catalog refresh and bundled snapshot automation #4117 — refresh + bundled snapshot automation (after v0.9.0 1.2: catalog refresh policy for live model metadata #4114)
• v0.9.0 remaining: model catalog consolidation and live refresh #4109 — umbrella; close when lane is done

Transcript copy polish (lane-copy) — ready now, work in this order

Items marked ⚠ touch transcript/sidebar rendering — check the open #4094 PR before starting; rebase after it merges.

• v0.9.0 5.1: disclose context percent once #4142 — disclose context percent once
• v0.9.0 5.2: trim redundant mode picker body copy #4143 — trim mode picker body copy
• v0.9.0 5.4: consolidate repeated setup hints #4144 — consolidate repeated setup hints
• v0.9.0 5.7: fix read done Searching verb mismatch #4145 — fix "read done / Searching" verb mismatch
• v0.9.0 5.8: make completed reasoning quiet by default #4146 — quiet completed reasoning by default ⚠
• v0.9.0 5.9: rename sidebar Tasks panel to match live activity semantics #4147 — rename sidebar Tasks panel to match live activity ⚠
• v0.9.0 5.10: remove duplicate and leaky words from default transcript #4148 — remove duplicate/leaky words from default transcript ⚠
• v0.9.0 UX: make dense TUI rows truncate or wrap semantically #3987 — semantic truncation/wrapping for dense rows
• v0.9.0 UX: default TUI presentation is too busy; compact mode should be standard #4095 — compact mode as default presentation (bigger; last)
• v0.9.0: Unicode, CJK, and terminal-width rendering QA #3488 — Unicode/CJK/terminal-width QA matrix (verification pass over the lane)
• v0.9.0 remaining: default transcript copy polish and activity wording #4112 — umbrella; close when lane is done

Fleet / AgentProfile (lane-fleet) — start after #4093 merges

• v0.9.0 2.19: make AgentProfile the canonical Fleet actor contract #4136 — make AgentProfile the canonical Fleet actor contract (do first — everything else builds on it)
• v0.9.0 2.20: let Fleet profiles save provider model and thinking #4137 — Fleet profiles save provider/model/thinking
• v0.9.0 2.21: remove stale Fleet loadout and model-class copy #4138 — remove stale loadout/model-class copy
• v0.9.0 UX: Fleet setup role step interleaves list rows with detail copy #3993 — Fleet setup role step layout fix (rebase on Fleet setup modal is provider-scoped instead of a role/profile roster editor #4093's fix)
• Per-sub-agent provider assignment (explicit routing) + LM Studio support #3965 — per-sub-agent provider assignment + LM Studio (align with PR Add per-sub-agent provider routing #3969 harvest — do not brute-force merge)
• feat: Environment-level tool sandboxing for sub-agents (enforce tool_restrictions) #4042 — enforce tool_restrictions for sub-agents
• v0.9.0 remaining: make AgentProfile canonical for Fleet rosters #4111 — umbrella; close when lane is done

Workflow runtime + UI (lane-workflow) — start after stopship green

Architecture phases first (canonical tracker #4175 — Fleet=WHO, Workflow=WHAT ORDER, Lane=INSTANCE, Runtime=WHERE):

• v0.9.0 Phase 1: Lane CLI + Runtime backends (tmux, worktrees, stream-json logs) #4176 — Phase 1: Lane CLI + Runtime backends (tmux, worktrees, stream-json)
• v0.9.1 Phase 2: Workflow steps reference Fleet roles (not raw prompts/profiles) #4177 — Phase 2: Workflow steps reference Fleet roles (pairs with v0.9.0 2.19: make AgentProfile the canonical Fleet actor contract #4136 in lane-fleet)
• v0.9.1 Phase 3: Workflow gates and handoffs between Fleet roles #4179 — Phase 3: gates and handoffs between roles
• v0.9.1: Stopship workflow as fleet-backed lane (dogfood #4090/#4093/#4094) #4178 — dogfood: run stopship workflow as a fleet-backed lane

Then UI/eventing (umbrellas #4110, #4038):

• v0.9.0 2.1: convert workflow progress strings to typed UI events #4118 — typed UI events for workflow progress (foundation for panel work)
• v0.9.0 2.2: add workflow metadata to sub-agent spawn records #4119 — workflow metadata on sub-agent spawn records
• v0.9.0 2.4: add WorkflowPanel unified activity surface #4121 — WorkflowPanel unified activity surface
• v0.9.0 2.5: route workflow events to panel and compact history card #4122 — route events to panel + compact history card
• v0.9.0 2.6: test WorkflowPanel state transitions #4123 — WorkflowPanel state-transition tests
• v0.9.0 2.3: default parallel write workflow children to worktree isolation #4120 — default parallel write children to worktree isolation
• v0.9.0 2.7: add planner-to-workflow structured launch path #4124 — planner→workflow structured launch path
• v0.9.0 2.8: update parent prompt for automatic Workflow use #4125 — parent prompt update for automatic Workflow use
• v0.9.0 2.10: implement automatic Workflow trigger and suppression rules #4127 — automatic trigger + suppression rules
• v0.9.0 2.11: add Workflow config keys and defaults #4128 — config keys and defaults
• v0.9.0 2.9: add approval card for elevated Workflow plans #4126 — approval card for elevated plans
• v0.9.0 2.12: add Workflow JS sandbox regression tests #4129 — JS sandbox regression tests
• v0.9.0 2.15: make one canonical model-facing work progress tool #4132 — one canonical model-facing progress tool
• v0.9.0 2.16: render one visible artifact per delegated unit #4133 — one visible artifact per delegated unit
• v0.9.0 2.17: give every child typed identity and remove unknown child UI #4134 — typed child identity; remove unknown-child UI
• v0.9.0 2.18: standardize Workflow phase agent task vocabulary #4135 — standardize phase/agent/task vocabulary
• v0.9.0 2.13: document automatic Workflow and Fleet operation #4130 — document automatic Workflow/Fleet operation
• v0.9.0 2.14: dogfood automatic Workflow scenarios #4131 — dogfood automatic Workflow scenarios (final)
• v0.9.0 remaining: Workflow automatic launch and unified activity UI #4110 / v0.9.0 Workflow: product-readiness tracker #4038 / v0.9.1 architecture: Fleet / Workflow / Lane / Runtime product model (canonical tracker) #4175 — umbrellas; close when lane is done

Turn Inspector (lane-inspector) — start after #4094 merges

• v0.9.0 TUI: extract Activity Detail and raw-detail helpers from ui.rs #4103 — extract Activity Detail helpers out of ui.rs (do first — shrinks conflict surface for the rest)
• v0.9.0 TUI: make Ctrl-O open a turn-level inspector #4104 — Ctrl-O opens turn-level inspector
• v0.9.0 TUI: make v and Alt-V clearly leaf-level raw details #4105 — v / Alt-V as leaf-level raw details
• v0.9.0 TUI: add turn and checkpoint timeline to Turn Inspector #4106 — turn + checkpoint timeline
• v0.9.0 TUI: show LSP repair-loop state inside Turn Inspector #4107 — LSP repair-loop state display
• v0.9.0 TUI: export compact turn handoff artifacts from Turn Inspector #4108 — export compact turn handoff artifacts
• v0.9.0 TUI: Turn Inspector evidence cockpit #4102 — umbrella; close when lane is done

Performance (lane-perf) — start after stopship green; mechanical, one PR per issue

• v0.9.0 6.1: finish parking_lot migration at hot lock sites #4149 — finish parking_lot migration at hot locks
• v0.9.0 6.2: introduce FastHashMap for internal keyed maps #4150 — FastHashMap for internal keyed maps
• v0.9.0 6.3: cache lowercased search haystacks at remaining sites #4151 — cache lowercased search haystacks
• v0.9.0 6.4: replace hot linear Vec scans with map lookups #4152 — replace hot linear Vec scans with map lookups
• v0.9.0 6.7: remove collect before join allocations #4153 — remove collect-before-join allocations
• v0.9.0 6.10: cache compiled regex in hot paths #4154 — cache compiled regex in hot paths
• v0.9.0 6.11: build command groups once with OnceLock #4155 — build command groups once with OnceLock
• v0.9.0 6.14: use eq_ignore_ascii_case at ASCII-only sites #4156 — eq_ignore_ascii_case at ASCII-only sites
• v0.9.0 6.15: log discarded spawn_blocking results #4157 — log discarded spawn_blocking results
• v0.9.0 6.16: define consts for repeated string literals #4158 — consts for repeated string literals
• v0.8.67: Launch is slow; profile and remove startup inefficiency #3757 — profile and fix slow launch
• v0.9.0 Performance: TUI lag and memory pressure from high agent fan-out sessions #4014 — TUI lag/memory under high fan-out (needs the workflow-lane event work; do last)
• v0.9.0 remaining: focused Rust/TUI performance quick wins #4113 — umbrella; close when lane is done

Setup / onboarding (lane-setup) — start after stopship green

• v0.9.0 UX: do not show 'setup is ready' intro when onboarding is skipped but auth/setup is not ready #3985 — don't claim "setup is ready" when onboarding was skipped
• v0.9.0: Guided multi-step setup wizard for built-in hosted providers in /provider #3875 — guided multi-step provider setup wizard
• v0.8.67 Setup: tools, MCP, skills, and plugins step #3407 — tools/MCP/skills/plugins setup step (largest; last)

PR harvest lane (parallel to all waves — review, don't reimplement)

PR	Issue	Notes
#4088	#4026	Mergeable; terminal selection highlight — review Windows CI hunks, then merge
#4087	#4082	Draft refactor; finish review
#4084	#4065	Draft Fleet alias cleanup; land before/with lane-fleet
#3761	#3757	Conflicting; harvest/cherry-pick for the lane-perf startup issue
#3969	#3965	Conflicting; align with #4065 + lane-fleet contract first — do not brute-force merge

Verification gate (run before every PR)

cargo fmt --all --check
cargo clippy --workspace --all-features --locked -D warnings \
  -A clippy::uninlined_format_args -A clippy::too_many_arguments \
  -A clippy::unnecessary_map_or -A clippy::collapsible_if -A clippy::assertions_on_constants
cargo test --workspace --locked
cargo build --release -p codewhale-tui

Full playbook: docs/AGENT_WORKFLOWS_0868.md. Architecture vocabulary: #4175. Guard: #4172.

[Hmbown]

July 7 maintenance update after direct-main sweep

Live truth after the maintenance pass:

• main / origin/main: cdb52ee487d6bf910198250de12d97e970ee98d3.
• Cloudflare site: redeployed codewhale-web with Worker version d899de53-46a5-4d97-8325-07ae9bf9da34.
• Website feed fix: /api/github/feed is now forced dynamic and verified returning 50 live items; first item is Cannot read/write files under ~/Library/CloudStorage/Dropbox/ (macOS File Provider) #4085.
• Repo cleanliness: local worktree clean after pushes.

Merged since this packet was written:

• fix(runtime): persist compaction summary so it survives engine reloads / compaction 摘要跨 engine 重载存活 #4091 by @MXAntian — compaction summary persistence across engine reloads.
• fix(tui): preserve native selection without mouse capture #4088 by @nightt5879 — native terminal selection works when mouse capture is disabled; closes bug(light-theme): terminal shell selection highlight invisible — no color inversion on light theme #4026.
• [codex] fix edit_file UTF-8 fuzzy cursor panic #4045 by @nightt5879 — UTF-8 fuzzy edit panic fix; edit_file panics on non-ASCII text after punctuation-normalized fuzzy match #3971 is closed.
• docs(readme): link CodeWhale for VS Code GUI frontend #4035 by @gaord — README / README.zh-CN GUI frontend pointer.
• chore(tui): remove unused whale_routes taxonomy #4041 by @DarrellThomas — remove unused whale route taxonomy; closes chore(cleanup): remove or wire unused whale route taxonomy #3852.

Fresh issue triage completed:

• feat: Environment-level tool sandboxing for sub-agents (enforce tool_restrictions) #4042 moved into v0.8.68 with security, sandbox, subagents, tools, reliability; maintainer note says the first PR should focus on sub-agent inheritance of disallowed_tools.
• Cannot read/write files under ~/Library/CloudStorage/Dropbox/ (macOS File Provider) #4085 moved to v0.8.69 platform compatibility with reliability; maintainer note asks for exact stderr and direct-shell vs CodeWhale exec comparison for macOS CloudStorage/Dropbox.
• feat(ui): add option to disable thinking/reasoning background highlight #4089 moved to v0.8.70 display/reasoning output with tui and ux; acceptance note calls for visible reasoning with optional no-fill highlight plus settings/render coverage.

Milestone cleanup:

• Closed completed milestones v0.8.65, v0.8.66, and v0.8.67. All had zero open issues.

Remaining PR queue shape:

• Draft / not ready: refactor(hooks): split config and executor modules #4087, fix(fleet): reject retired profile loadout aliases #4084, perf(tui): fix the five render/input hot paths (#3896–#3900) #3902, v0.8.67: LongCat provider + post-#3960 review follow-ups + version bump #4034.
• Needs conflict/scope work: feat: add TormentNexus extension crate with full Pi extension parity #4086, fix(onboarding): localize dynamic welcome steps #4044, fix(tui): remove legacy token-only pricing helpers #4040, ci: light-classify inert scripts and stop allocating macOS/Windows runners for light PRs #4025, Add per-sub-agent provider routing #3969, feat: LLM can start MCP servers from chat context #3866, [codex] sync Xiaomi MiMo Token Plan docs #3820, Localize Hotbar setup wizard #3785, Feat/opencode zen provider #3781, [codex] expose context compaction gates (#3765) #3780, feat(web): redesign homepage with trust strip, GitHub nav link, and mirror footer (#3413) #3762, [codex] defer startup maintenance cleanup #3761.
• Potential review candidates but not yet inspected deeply in this pass: feat: add dynamic MCP server infrastructure to McpPool #3869 and fix(tui): expand active tool run summaries #3818. Do not merge them blindly; inspect diff, check scope, and preserve contributor credit.

Verification run locally before/after pushes:

• cargo test -p codewhale-tui --locked recover_terminal_modes -- --nocapture
• cargo test -p codewhale-tui --locked resume_tui_child_modes_leaves_alternate_scroll_off_when_mouse_capture_inactive -- --nocapture
• cargo test -p codewhale-tui --locked test_edit_file_fuzz_tolerates -- --nocapture
• cargo test -p codewhale-tui --locked compaction_lifecycle_emits_item_events_with_compaction_counts -- --nocapture
• cargo test -p codewhale-tui --locked summary_merge -- --nocapture
• cargo test -p codewhale-tui --locked summary_strip -- --nocapture
• git diff --check
• cargo fmt --all --check
• python scripts/check-coauthor-trailers.py --range HEAD~1..HEAD
• cd web && npm ci && npm run check:facts && npm run check:docs && npm run build && npm run deploy

CI note:

• GitHub Actions repo variable CLOUDFLARE_ACCOUNT_ID and secret CLOUDFLARE_API_TOKEN already exist. The local npm run check:deploy-env failure only means those CI env vars are not exported in a local shell.

[Hmbown]

Proposed v0.8.68 cut line: reduce milestone to ~18 must-track issues

After the July 7 dogfood run, the current v0.8.68 milestone is too broad to be a release plan. It mixes hard dogfood blockers, Workflow product-readiness, Fleet/model routing, setup UX, tool expansion, and broad refactor epics. Suggested release focus:

v0.8.68 should ship a trustworthy Workflow/Fleet dogfood slice, not every backlog issue currently labeled v0.8.68.

Keep As The Core 0.8.68 Slice

1. Fleet setup modal is provider-scoped instead of a role/profile roster editor #4093 — Fleet profile model picker is provider-scoped and draft preview traps ratify key
2. v0.9.0 TUI: repeated Ctrl+C re-prompts instead of exiting in PTY/raw-mode path #4090 — repeated Ctrl+C re-prompts instead of exiting in PTY/raw-mode path
3. v0.9.3: fleet model-policy contract — decide, and prune two dead profile aliases #4065 — fleet model-policy contract; decide and prune dead aliases
4. Per-sub-agent provider assignment (explicit routing) + LM Studio support #3965 — per-sub-agent provider assignment / explicit routing + LM Studio support
5. feat: Environment-level tool sandboxing for sub-agents (enforce tool_restrictions) #4042 — environment-level tool sandboxing for sub-agents / enforce tool_restrictions
6. v0.9.0 Workflow: product-readiness tracker #4038 — Workflow product-readiness tracker
7. v0.9.2 Workflow: wire the model-facing workflow tool and run driver #2974 — model-facing workflow tool and run driver
8. v0.9.2 Workflow: background task phase ledger UI #4039 — Workflow background task phase ledger UI
9. v0.9.4 WhaleFlow: agent topology primitives (fan-out, pipeline, diamond, speculative) #4012 — Workflow topology primitives beyond fan-out: pipeline, diamond, speculative, critique loop
10. v0.9.4 WhaleFlow: verification gates (compile, test, lint, review as post-agent hooks) #4013 — verification gates after worker completion
11. v0.9.0 Performance: TUI lag and memory pressure from high agent fan-out sessions #4014 — TUI lag and memory pressure from high agent fan-out sessions
12. v0.9.4 WhaleFlow: context budget management for high-fan-out orchestration #4015 — context budget management for high-fan-out orchestration
13. v0.9.4 WhaleFlow: worktree pool management for parallel agent isolation #4016 — worktree pool management for parallel agent isolation
14. v0.9.1 Runtime: make current Work state model-visible on parent turns #3983 — make current Work state model-visible on parent turns
15. v0.9.0 UX: do not show 'setup is ready' intro when onboarding is skipped but auth/setup is not ready #3985 — skipped onboarding must not claim setup is ready when auth/setup is not ready
16. v0.9.0 UX: Fleet setup role step interleaves list rows with detail copy #3993 — Fleet setup role step interleaves list rows with detail copy
17. v0.9.0 UX: model picker shows awkward single-row range titles #3995 — model picker single-row range title awkwardness
18. v0.9.0 UX: make dense TUI rows truncate or wrap semantically #3987 — dense TUI rows truncate/wrap semantically
19. v0.9.0: Unicode, CJK, and terminal-width rendering QA #3488 — Unicode, CJK, and terminal-width rendering QA
20. v0.9.0: Guided multi-step setup wizard for built-in hosted providers in /provider #3875 — guided multi-step setup wizard for hosted providers in /provider

If we need an even tighter 15, drop #3987, #3488, #3875, #4016, and #3983 to stretch. Keep #4093/#4090/#4065/#3965/#4042/#4038/#2974/#4039/#4012/#4013/#4014/#4015/#3985/#3993/#3995.

Why These Stay

• Fleet setup modal is provider-scoped instead of a role/profile roster editor #4093, v0.9.3: fleet model-policy contract — decide, and prune two dead profile aliases #4065, and Per-sub-agent provider assignment (explicit routing) + LM Studio support #3965 are one product knot: Fleet profiles must bind to explicit provider/model routes, not accidentally depend on the current provider.
• v0.9.0 Workflow: product-readiness tracker #4038/v0.9.2 Workflow: wire the model-facing workflow tool and run driver #2974/v0.9.2 Workflow: background task phase ledger UI #4039/v0.9.4 WhaleFlow: agent topology primitives (fan-out, pipeline, diamond, speculative) #4012/v0.9.4 WhaleFlow: verification gates (compile, test, lint, review as post-agent hooks) #4013/v0.9.0 Performance: TUI lag and memory pressure from high agent fan-out sessions #4014/v0.9.4 WhaleFlow: context budget management for high-fan-out orchestration #4015 are the Workflow product-readiness spine. The dogfood failure shows this is not merely fan-out; Workflow needs queued/staged/coding/review/verification flows with compact state.
• feat: Environment-level tool sandboxing for sub-agents (enforce tool_restrictions) #4042 is trust-boundary critical: sub-agent tool restrictions must actually apply before Workflow makes more workers normal.
• v0.9.0 TUI: repeated Ctrl+C re-prompts instead of exiting in PTY/raw-mode path #4090 is a pure stopship TUI control-path bug.
• v0.9.0 UX: do not show 'setup is ready' intro when onboarding is skipped but auth/setup is not ready #3985/v0.9.0 UX: Fleet setup role step interleaves list rows with detail copy #3993/v0.9.0 UX: model picker shows awkward single-row range titles #3995 are visible setup/model-selection UX papercuts that directly affect the same dogfood path.

Defer Or Move Out Of The Release Slice

Defer broad refactor epics unless they are already in flight and low-risk:

• v0.9.3 EPIC: Repair the engine policy architecture #3938-v0.9.3 Refactor: split shell process management from exec/wait/interact tool schemas #3958 engine/refactor split series
• v0.9.3 refactor(web_search): split provider backends into submodules #4077/v0.9.3 refactor(project_context): extract constitution, pack, loader modules #4079/v0.9.3 refactor(settings): split TuiPrefs, schema, persistence #4080/v0.9.3 refactor(hooks): split config types from HookExecutor #4082/v0.9.3 refactor(mcp): move McpConnection/McpPool out of mcp.rs #4083 modular cleanup issues
• v0.9.2 EPIC: Hotbar command surface and source adapters #3389/v0.9.2 Hotbar refactor: extract reusable action-source adapters from command palette, MCP, skills, and plugins #3399 Hotbar adapter epics unless already merged locally
• feat: semantic codebase_search tool (BM25 or embedding index) #4066/feat: standalone read_lints tool for on-demand diagnostics #4070/v0.9.3 Tools: expose LSP navigation, references, rename, and code actions #3975/v0.9.3 Tools: read notebooks and archive entries as first-class inputs #3977/v0.9.3 Tools: add structural code search and AST-backed edit previews #3980/v0.9.3 Tools: add a debugger protocol surface for breakpoints, stack, and variables #3981 broad new tool surfaces
• v0.9.2 Tools: internal URI schemes (pr/issue/diff/conflict) + GitHub Actions run-watch as one read surface #3996/v0.9.3 Engine: stream-rule injection — abort mid-token and retry with a corrective rule #3997/v0.9.3 Tools: atomic commit splitting — order unrelated changes by dependency, reject cycles #3999 ambitious new runtime/tool features
• v0.9.4 Memory: seed project-scoped recall ahead of the full backend #3976 project recall and v0.9.4 Subagents: add an optional advisor watcher for live turns #3982 advisor watcher
• feat: MCP hot-reload for model-visible tool pool #4068 MCP hot-reload unless a small safety bug is already isolated
• v0.9.3: build constitution customization against the reduced 55-line constitution #4064 constitution customization and v0.9.0 execution board: lane order, dependencies, and agent protocol (canonical packet) #4092 tracking packet itself should not be blockers

Suggested Board Action

Do not mass-retarget yet. First mark the core 15-20 above as the release lane, then move/defer the rest once the branch owner confirms what is already implemented locally on codex/0868-next. The main risk is accidentally treating local-only dogfood fixes as shipped or punting a real blocker just because it looks like UX polish.

[Hmbown]

Add #4094 to the core v0.8.68 Workflow/Fleet trust slice

New dogfood issue: #4094 — Sub-agent detail panel is empty during active work.

This should be treated as part of the core Workflow/Fleet readiness cut, not generic polish. The Workflow phase ledger (#4039) covers run-level progress, but the generic Sub-agent detail view must also show the worker's actual live activity: objective/profile/model route, running/completed tool calls, compact statuses, final handoff, and artifact handles.

If keeping the cut to 15-20, include #4094 and drop one of the broader stretch UI items such as #3488 or #3987 before dropping this. Without it, users cannot debug or trust queued/running workers.