Tasks

Card edge: cloud (Claude) big local model small local model Drag a card between columns · click a card to open it

Scoping

Backlog

TASK-ERRAND-1 high

Email intake (READ-ONLY first): connect Gmail -> screening agent classifies + extracts action items into the inbox

errand email intake build_effort
TASK-FLEET-001 high

Fix the fleet mDNS/IPv6 bridging heat-loop (bridge0 duplicated IPv6 across ~12 en interfaces)

hardware fleet heat claude_cloud
TASK-Q240624-01 high

Agent: signal-screener — screen inbound Signal + draft a first-reply (propose-only, never auto-send)

agent comms signal build_effort
TASK-Q240624-06 high

Agent: travel-cascade — a moved flight re-plans everything it touches (2nd/3rd-order)

agent cascade northstar build_effort
TASK-R01-1 high

Define the inbox-item schema: kind (ack|question|warno), body, options, originating_task_id, answer

inbox schema build_effort
TASK-R01-2 high

Wire the unified Inbox (apex + Pixel + desk terminal) to read/write inbox items live; Principal answers from any device

inbox ui build_effort
TASK-R01-3 high

ACK flow: every capture gets an inbox ack (created + assumptions) with accept/correct -> training record

inbox feedback training-data build_effort
TASK-R02-1 high

Lock the STT transcript output-path contract (single source/dest)

swarm stt plumbing build_effort
TASK-R02-2 high

Make Ollama listen on the tailnet on zachmac1 (so other nodes can call it)

swarm ollama surface-a build_effort
TASK-R02-3 high

Serialize the todo-extractor (remove the race condition)

swarm extractor plumbing build_effort
TASK-R02-4 high

End-to-end smoke test: Pixel capture -> ledger entry

swarm test build_effort
TASK-R03-1 high

Define the Task envelope Pydantic schema (task_id, kind, inputs, model, prompt_version, params seed/top_p/top_k, upstream, downstream, per-field confidence, needs_approval)

swarm schema build_effort
TASK-R03-2 high

Make every agent read + write the envelope (no free-form I/O)

swarm schema build_effort
TASK-R04-1 high

Choose + stand up a SQLite-backed (or watched-dir) work queue

swarm queue build_effort
TASK-R04-2 high

Worker base class: claim -> transform -> emit envelope -> clear context

swarm queue context-discipline build_effort
TASK-R04-3 high

Enqueue-next-step handoff between workers

swarm queue handoff build_effort
TASK-R05-1 high

Classifier agent: todo vs event vs task (Pixel tag is the label/prior)

swarm agent classifier classifier
TASK-R05-2 high

Field-filler agent: extract + assume fields, emit per-field confidence

swarm agent extraction field_filler
TASK-R05-3 high

Validator agent (zachmac2): check filler vs transcript, flag low-confidence fields for review

swarm agent validator validator
TASK-R05-4 high

Wire the 3-agent chain end-to-end through the queue

swarm agents integration build_effort
TASK-R06-1 high

Swap all qwen2.5 defaults -> US-origin (Llama 3.x / Phi) in extractor + agent YAMLs

models policy build_effort
TASK-R06-2 high

Assign model per node by role (zachmac2 = STT + validator/composer; zachmac1/3/4 = narrow agents; homeserver1 = orchestration only)

models hardware build_effort
TASK-R07-1 high

Verify node-to-node tailnet reachability (queue, Ollama, STT) — not localhost

tailnet ops build_effort
TASK-R08-1 high

Every agent writes a real agent_runs + node_heartbeats row per run

swarm telemetry build_effort
TASK-R08-2 high

Swarm page reads live rows: model, node, tokens, latency, prompt_version, confidence, approved/corrected, current+next task

swarm telemetry ui build_effort
TASK-SHOP-1 high

Storage + view: tools_shopping.html + /ops/views/tools/shopping + shopping_items sqlite table

shopping ui db build_effort
TASK-SHOP-2 high

Voice/text intent: 'Witan, add X to my shopping list' → append to shopping_items (regex, no LLM)

shopping voice intent build_effort
TASK-SKEL-A high

Step A: Seed the recurring skeleton blocks (incl. R&R sleep block)

calendar skeleton seed build_effort
TASK-SKEL-B high

Step B: Auto-assigner — real-time, sticky, same-domain

calendar assigner build_effort
TASK-SKEL-C high

Step C: Empty-block-as-signal surfaces in brief + dashboard

calendar brief signal build_effort
TASK-SKEL-D high

Step D: Apex 'Current Focus' card on mywitan.xyz (Done/Not-Done/Else)

apex ui focus-card build_effort
TASK-SKEL-E high

Step E: block_outcomes table + TodoRow telemetry fields

calendar schema telemetry build_effort
TASK-SWARMCOLOR-001 high

Add a `model_class` field (cloud|big_local|small_local) to each agent/node in the registry (config/agents/*.yaml, nodes.yaml)

swarm config claude_cloud
TASK-SWARMCOLOR-002 high

Render colour-coded agent cards on the Swarm page; add the bug agent's card

swarm ui claude_cloud
TASK-SYS-003 high

Add a `tags` column (JSON) + migration; support a 'bug' tag + filter chips

schema tags bug build_effort
TASK-SYS-006 high

Load this backlog seed into the LIVE homeserver1 tasks DB

ops seed deploy build_effort
task_20260629_b9127f20 high

BUG: PIPE ALARM (n8n): workflow "Witan Social Scheduler" failed: The connection was a

✎
TASK-APEX2-4

Move 'submit photo' into Tools (a photo tool tile)

apex tools photo build_effort
TASK-BRIEF-2

Soft-content sections (quote / learn / kids) wired per-section: local model + curated banks, NO external API

brief soft-content build_effort
TASK-BRIEF-3

RSS/news in the brief — GATED behind prompt-injection containment (curated allow-list + local fetch + sanitize)

brief rss injection build_effort
TASK-CLAUDE-3

Swarm page renders live Claude cards (current+next task, model, session id); supports 0..N instances

claude swarm ui build_effort
TASK-ERRAND-3

Outbound-to-a-person (text the wife): confirm-gated message-out asking a question + capturing the reply

errand outbound signal build_effort
TASK-ERRAND-5

Meeting action-items -> schedule: from a meeting transcript, extract action items -> to-dos/events

errand meeting intake build_effort
TASK-ERRAND-6

Priority-gated 'it's all ready' inbox notify (WARNO): surface the summary only once the Principal is free

errand inbox warno build_effort
TASK-FLEET-002

Stand up EXO across the Mac mini fleet (convene-on-demand for big models)

hardware fleet exo claude_cloud
TASK-FUNC-005

Shared action client (actions_client.py) so every agent + the MCP server call the same functions; refactor the extractor onto it

actions client mcp claude_cloud
TASK-INTERVIEW-2

Turn-based conversation surface (desk terminal + Pixel): ask -> wait -> follow up; no streaming

interview voice ui build_effort
TASK-ORCH-2

Write the n8n-vs-internal decision note (per-flow): which errand steps, if any, justify n8n

n8n decision injection claude_cloud
TASK-Q240624-02

Agent: translation-normalizer — any-language inbound -> canonical English for downstream

agent intake translate build_effort
TASK-Q240624-03

Agent: transcription-cleaner — repair voice disfluencies/punctuation pre-scope

agent intake voice build_effort
TASK-Q240624-07

Agent: dependency-cascade — finishing X unblocks Y, renegotiate Z

agent cascade northstar build_effort
TASK-R01-5

WARNO flow: priority-gated notifications (never interrupt higher-priority work with a lesser notice)

inbox warno priority-gated build_effort
TASK-R03-3

Use the envelope as the Swarm-row + training-record source

swarm telemetry training-data build_effort
TASK-R06-3

Pin seed/top_p/top_k per agent (determinism lever)

models determinism build_effort
TASK-R06-5

Fix the training-data logger path (services/research_log.py writes to a deleted research/ path)

data logger training-set claude_cloud
TASK-R07-3

Pixel + desk terminal RECEIVE inbox items / acks (ties to R01)

pixel inbox feedback build_effort
TASK-R08-3

Waterfall/swimlane view from parent_run_id chains

swarm waterfall ui build_effort
TASK-R09-1

Compute rolling approval rate + completed-count per agent (or task-type)

swarm autonomy metrics build_effort
TASK-R09-2

Show progress toward 50@99% on each Swarm card

swarm autonomy ui build_effort
TASK-R09-3

Gate policy: below threshold -> requires_validation by a larger model

swarm autonomy policy build_effort
TASK-R10-1

On a 100%-approved task, store its envelope as a reusable plan

determinism store build_effort
TASK-R10-2

Match an incoming near-identical task (same kind, new variables)

determinism matching build_effort
TASK-R10-3

Variable-substitution replay path that skips re-reasoning

determinism replay build_effort
TASK-SKEL-F

Step F: Reconciliation strip + auto roll-forward (next same-domain block)

calendar reconcile build_effort
TASK-SKEL-G

Step G: /ops/views/focus reports page (4 headline metrics + Not-Done analysis)

reports focus ui build_effort
TASK-SKEL-H

Step H: Voice-driven rapid replan (e.g. 'Army called me in for 2 hours')

voice intake replan build_effort
TASK-SOCIAL-1

Capture: a thought / win / link -> a post-draft request (voice or text)

social capture build_effort
TASK-SOCIAL-2

Per-platform drafting agent (LinkedIn long-form vs X short) in the Principal's voice — local model

social drafting build_effort
TASK-SOCIAL-3

Review + approve drafts in the Inbox before anything posts; (optional, later, gated) API post to X/LinkedIn

social approval gated build_effort
TASK-SYS-005

Task create/edit modal + recurring-task fields (cron expr) for the 'today' view

tasks ui recurring build_effort
task_20260617_89da8371

When Witan suggests changes to my schedule, I need a more in-depth assessment be

output should feedback ✎
task_20260623_4cb76af3

Scoper just died with my to do I wanted made for my daughters school paperwork

bug should feedback ✎
task_20260711_18c81cfa

STATE-DELTA: ui 2026-07-11 — RTR-43 E1-3 UNBLOCKED (adds to STATE msg_e924e004)

output should feedback ✎
task_20260711_686d9bdf

Brief Claude: FIX TASK (routed from gap-critic msg_9561bb32, for your relaunch q

output should feedback ✎
task_20260711_6f780fe4

UI Claude: STATE-DELTA 2: ui 2026-07-11 — RTR-43 E1 render is FULLY UNBLOCKED (a

output should feedback ✎
task_20260711_aa03c942

PRIORITY note — #1561 (spine: lane-plumbing never becomes a task + emitters card

output should feedback ✎
task_20260711_e4ea5af0

capture-door scope note for the fix: my 10:44:18Z REPLY on msg_fb663225 minted Z

output should feedback ✎
task_20260711_ef9b9125

Agent Claude: LOOP ESCALATION 2 (msg_4aacacd2 still unread ~20min)

output should feedback ✎
TASK-BUG-003

Bug-fix agent (DEFERRED): autonomous fixer — not now

bug agent deferred claude_cli
TASK-ERRAND-4

Product find + add-to-cart (DEFERRED, most-gated): right color/size -> cart, NEVER auto-purchase

errand shopping browse build_effort
TASK-Q240624-04

Agent: whoop-health — recovery/strain -> schedule load signal

agent domain whoop build_effort
TASK-Q240624-05

Agent: finance-categorizer — Plaid txns -> domains/budgets

agent domain plaid build_effort
TASK-SKEL-I

Step I: Sunday weekly-review paper handout in morning brief (companion to Day-view cycle)

brief sunday-review paper build_effort

Ready

Active

Reads the agent task ledger (tasks table). Drag cards between columns to move them; click a card to open it, add notes for the agents, or send it to a hidden column (Blocked / Review / Done). Seeded from tasks_backlog_seed.yaml.

Tasks

Scoping

Backlog

Next

Ready

Active