taskuary 0.3.3.0__tar.gz → 0.3.3.2__tar.gz

{taskuary-0.3.3.0/taskuary.egg-info → taskuary-0.3.3.2}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: taskuary
-Version: 0.3.3.0
+Version: 0.3.3.2
 Summary: Automate your job - the local-first operating system for your work: AI triage in, coding agents out.
 License: MIT
 Project-URL: Homepage, https://github.com/ldbumble/taskuary
@@ -62,22 +62,18 @@ Dynamic: license-file
 # Taskuary
 
 [![CI](https://github.com/ldbumble/taskuary/actions/workflows/ci.yml/badge.svg)](https://github.com/ldbumble/taskuary/actions/workflows/ci.yml)
-[![PyPI](https://img.shields.io/pypi/v/taskuary.svg?cacheSeconds=600)](https://pypi.org/project/taskuary/)
+[![PyPI](https://img.shields.io/pypi/v/taskuary.svg?cacheSeconds=300&release=0.3.3.2)](https://pypi.org/project/taskuary/)
 [![Python 3.10+](https://img.shields.io/badge/python-3.10%2B-3776ab.svg)](https://github.com/ldbumble/taskuary)
 [![License: MIT](https://img.shields.io/badge/license-MIT-blue.svg)](LICENSE)
 
-## Your inbox, staffed by AI agents.
+## Your inbox, staffed by AI agents
 
-Taskuary brings your mail, chats, issues, and scheduled reports into one local conversation:
-a live work pipe on the left and chat on the right. AI triage decides what is real work,
-your coding CLI works the tasks in your own repos, and replies wait for your approval.
-Nothing sends or ships without you.
+Taskuary turns incoming messages into organized work. It sorts what matters, hands tasks to
+your agents, and brings decisions back to you. Nothing sends or ships without your approval.
 
-![The Taskuary Studio: work arrives, triage decides what needs action, and agents take seats to work it.](https://raw.githubusercontent.com/ldbumble/taskuary/master/docs/hero.gif)
+![Taskuary: the pipe ranked by triage beside the assistant chat - a mail pulled in with its drafted reply, the Timeline in task view, the Reports and the Board.](https://raw.githubusercontent.com/ldbumble/taskuary/master/docs/hero.gif)
 
-Taskuary is early—currently **v0.3.3.0**—and moving fast. The funnel, review queue,
-agent sessions, and reports pipeline are in daily use; breaking changes are still possible
-before 1.0.
+Taskuary is early—currently **v0.3.3.2**—so breaking changes are still possible before 1.0.
 
 <p align="center">
   <a href="https://taskuary.com/demo/"><img
@@ -85,43 +81,61 @@ before 1.0.
     alt="Try Taskuary now, in your browser"></a>
 </p>
 
-<p align="center"><sub>The real app over invented work — the unified pipe and chat, triage,
-a coding session mid-run, and reports with charts. Nothing to install and nothing connects
-to anything.</sub></p>
+<p align="center"><sub>The real app with invented data. Nothing connects, sends, or runs.</sub></p>
 
-## One timeline, one chat
+## What Taskuary can do
 
-Outlook, Gmail, Teams, Slack, Telegram, WhatsApp, GitHub, Jira, alerts, and reports all
-arrive in the pipe beside the same conversation. **Unread** is the prioritized walkthrough;
-**All** is the recent timeline. Click any row—or name a sender, subject, or task in chat—to
-pull it into the conversation with its context, triage reasoning, draft, and available actions.
+**One Timeline and Assistant**
+
+Mail, chats, issues, alerts, and reports appear on one Timeline and in one Assistant chat.
+Open any item to see its context, status, and available actions.
 
 ![Taskuary's unified work pipe and chat, populated only with fictional demo data.](https://raw.githubusercontent.com/ldbumble/taskuary/master/docs/screenshot-chat.png)
 
-*Everything shown is generated by the built-in demo. Dana Whitfield, Northwind Facilities,
-the people, messages, repositories, tasks, and reports are all fictional.*
-
-The same conversation also speaks up when something falls between tasks: a reply went
-unanswered, a meeting needs context, a task went quiet, or a pattern appears across the
-week's work. Each observation carries its evidence and can be acted on, dismissed, snoozed,
-or taught away with **Not this**. These observations now appear in this same chat instead of
-as separate Assistant posts on the Timeline.
-
-These checks can watch the systems you run on, not only incoming messages. Point one at a
-Sage Intacct query, a database, a file, a REST or MCP tool, a cloud log group, or an agent
-skill—no saved report has to stand behind it—and Taskuary reads it silently on schedule.
-Describe what a check should watch and AI can write the source cards using only connected
-systems, after reading their real schema and asking rather than guessing a filter. The same
-help sits on every source card in the report builder. Voice, schedule, model, and thresholds
-remain yours to change; a check stays silent when there is nothing useful to add to the chat.
-
-Coding work can go straight to Claude Code, Codex, Gemini, Cursor, Copilot, or any CLI that
-accepts a prompt on stdin. You watch the live terminal, answer questions, review the diff,
-and approve what happens next. Research, marketing, and other general work belong in a
-conversational workspace built with assistant-ui and your existing Taskuary AI connections.
-Chat and terminal are two views of the same session: model, history, queued instructions,
-image attachments, and the side-by-side browser stay put when you switch. The same workspace
-also appears on the Wall.
+**AI triage**
+
+Triage files routine updates, turns real requests into tasks, and puts urgent decisions first.
+
+**Agents in your own repos**
+
+Send coding work to Claude Code, Codex, Gemini, Cursor, Copilot, or another CLI.
+
+**Live workspaces**
+
+Watch the terminal, answer questions, review changes, and use the built-in browser without
+losing the session.
+
+**Scheduled checks**
+
+Watch databases, files, APIs, cloud logs, reports, or agent skills. Quiet checks stay silent.
+
+**You stay in control**
+
+Replies and proposed actions wait in Review. You decide what sends, runs, closes, or gets dismissed.
+
+**Catches what slipped**
+
+Surface unanswered replies, upcoming meetings, quiet tasks, failed reports, and important changes.
+
+## How work flows
+
+### 1. Work arrives and gets sorted
+
+Taskuary reads connected inboxes, separates tasks from noise, and shows the result on one Timeline.
+
+![The Timeline in task view: every message with its verdict, and the drafted reply waiting on you](https://raw.githubusercontent.com/ldbumble/taskuary/master/docs/screenshot-timeline-crop.png)
+
+### 2. An agent works on it
+
+Tasks move from **Queued** to **Agent working**. The Board shows live progress and anything waiting on you.
+
+![Taskuary Board with queued, active, and waiting work](https://raw.githubusercontent.com/ldbumble/taskuary/master/docs/screenshot-board.png)
+
+### 3. The result comes back to you
+
+Finished work, draft replies, questions, and loose ends return for review. You decide what to send, change, snooze, or dismiss.
+
+![Taskuary assistant showing loose ends and suggested actions](https://raw.githubusercontent.com/ldbumble/taskuary/master/docs/screenshot-assistant.png)
 
 ## Install
 
@@ -164,25 +178,15 @@ connection before it goes live.
 taskuary --demo                    # or: docker compose --profile demo up
 ```
 
-A full Taskuary over invented work — a morning's mail and chats already triaged with the
-reasons attached, drafts waiting in Review, agents on the board with their transcripts
-playing, live task handoffs on the Board, and the durable, high-signal company Hub.
-Every Hub post names its human or agent author and originating task; Board handoffs are short-lived
-checkout coordination that rolls up nightly. The AI answers from a script and every door
-to the outside world is shut at the API layer (`taskuary/demo.py`): nothing sends, no
-connection can be made or edited, no tool runs, and no CLI starts. Nobody's real data is in
-it — Dana Whitfield and Northwind Facilities are invented.
+The demo is the real interface with fictional work and scripted replies. It cannot connect to
+outside systems, send messages, run tools, or start agents. Its changes reset when you reload.
 
 ## Installs
 
 ![Daily installs of taskuary from PyPI, mirror traffic excluded](https://raw.githubusercontent.com/ldbumble/taskuary/master/docs/downloads.svg)
 
-Redrawn every morning by [`downloads.yml`](.github/workflows/downloads.yml) from PyPI's own
-numbers, with mirror traffic excluded — a full PyPI mirror pulls every release, so counting it
-would let a handful of real users read as hundreds. What it does **not** claim is that CI
-installs have been removed: the field that would separate `pip install` in a build runner from
-one on somebody's laptop is in PyPI's BigQuery dataset, not in the public API. The daily series
-is [docs/downloads.csv](docs/downloads.csv).
+Updated daily from PyPI with mirror traffic excluded. The raw series is
+[docs/downloads.csv](docs/downloads.csv).
 
 ## Documentation
 

taskuary-0.3.3.2/README.md

@@ -0,0 +1,140 @@
+# Taskuary
+
+[![CI](https://github.com/ldbumble/taskuary/actions/workflows/ci.yml/badge.svg)](https://github.com/ldbumble/taskuary/actions/workflows/ci.yml)
+[![PyPI](https://img.shields.io/pypi/v/taskuary.svg?cacheSeconds=300&release=0.3.3.2)](https://pypi.org/project/taskuary/)
+[![Python 3.10+](https://img.shields.io/badge/python-3.10%2B-3776ab.svg)](https://github.com/ldbumble/taskuary)
+[![License: MIT](https://img.shields.io/badge/license-MIT-blue.svg)](LICENSE)
+
+## Your inbox, staffed by AI agents
+
+Taskuary turns incoming messages into organized work. It sorts what matters, hands tasks to
+your agents, and brings decisions back to you. Nothing sends or ships without your approval.
+
+![Taskuary: the pipe ranked by triage beside the assistant chat - a mail pulled in with its drafted reply, the Timeline in task view, the Reports and the Board.](https://raw.githubusercontent.com/ldbumble/taskuary/master/docs/hero.gif)
+
+Taskuary is early—currently **v0.3.3.2**—so breaking changes are still possible before 1.0.
+
+<p align="center">
+  <a href="https://taskuary.com/demo/"><img
+    src="https://img.shields.io/badge/%E2%96%B6%20Try%20it%20now-no%20install%2C%20in%20your%20browser-2f4858?style=for-the-badge&labelColor=1f2a22"
+    alt="Try Taskuary now, in your browser"></a>
+</p>
+
+<p align="center"><sub>The real app with invented data. Nothing connects, sends, or runs.</sub></p>
+
+## What Taskuary can do
+
+**One Timeline and Assistant**
+
+Mail, chats, issues, alerts, and reports appear on one Timeline and in one Assistant chat.
+Open any item to see its context, status, and available actions.
+
+![Taskuary's unified work pipe and chat, populated only with fictional demo data.](https://raw.githubusercontent.com/ldbumble/taskuary/master/docs/screenshot-chat.png)
+
+**AI triage**
+
+Triage files routine updates, turns real requests into tasks, and puts urgent decisions first.
+
+**Agents in your own repos**
+
+Send coding work to Claude Code, Codex, Gemini, Cursor, Copilot, or another CLI.
+
+**Live workspaces**
+
+Watch the terminal, answer questions, review changes, and use the built-in browser without
+losing the session.
+
+**Scheduled checks**
+
+Watch databases, files, APIs, cloud logs, reports, or agent skills. Quiet checks stay silent.
+
+**You stay in control**
+
+Replies and proposed actions wait in Review. You decide what sends, runs, closes, or gets dismissed.
+
+**Catches what slipped**
+
+Surface unanswered replies, upcoming meetings, quiet tasks, failed reports, and important changes.
+
+## How work flows
+
+### 1. Work arrives and gets sorted
+
+Taskuary reads connected inboxes, separates tasks from noise, and shows the result on one Timeline.
+
+![The Timeline in task view: every message with its verdict, and the drafted reply waiting on you](https://raw.githubusercontent.com/ldbumble/taskuary/master/docs/screenshot-timeline-crop.png)
+
+### 2. An agent works on it
+
+Tasks move from **Queued** to **Agent working**. The Board shows live progress and anything waiting on you.
+
+![Taskuary Board with queued, active, and waiting work](https://raw.githubusercontent.com/ldbumble/taskuary/master/docs/screenshot-board.png)
+
+### 3. The result comes back to you
+
+Finished work, draft replies, questions, and loose ends return for review. You decide what to send, change, snooze, or dismiss.
+
+![Taskuary assistant showing loose ends and suggested actions](https://raw.githubusercontent.com/ldbumble/taskuary/master/docs/screenshot-assistant.png)
+
+## Install
+
+### Windows app
+
+Download the latest single-file
+[Taskuary.exe](https://github.com/ldbumble/taskuary/releases/latest/download/Taskuary.exe)
+and open it. No Python or installer is required.
+
+### Python
+
+Python 3.10 or newer works on Windows, macOS, and Linux:
+
+```bash
+pip install taskuary
+taskuary
+```
+
+Taskuary opens at [http://127.0.0.1:7787](http://127.0.0.1:7787). For a native desktop
+window instead, install `pip install "taskuary[desktop]"` and run `taskuary-desktop`.
+
+### Docker
+
+```bash
+git clone https://github.com/ldbumble/taskuary
+cd taskuary
+docker compose up
+```
+
+Then open [http://127.0.0.1:7787](http://127.0.0.1:7787). Docker runs the web app;
+coding CLIs and the optional WhatsApp bridge remain on the host.
+
+On first run, connect an AI provider or local Ollama model, add at least one inbound
+channel, then choose the coding CLI that should receive tasks. The setup wizards test each
+connection before it goes live.
+
+## Try it without installing anything
+
+```bash
+taskuary --demo                    # or: docker compose --profile demo up
+```
+
+The demo is the real interface with fictional work and scripted replies. It cannot connect to
+outside systems, send messages, run tools, or start agents. Its changes reset when you reload.
+
+## Installs
+
+![Daily installs of taskuary from PyPI, mirror traffic excluded](https://raw.githubusercontent.com/ldbumble/taskuary/master/docs/downloads.svg)
+
+Updated daily from PyPI with mirror traffic excluded. The raw series is
+[docs/downloads.csv](docs/downloads.csv).
+
+## Documentation
+
+- [Getting started](docs/getting-started.md)—installation, first-run setup, Docker, and data
+- [Product guide](docs/product-guide.md)—the workflow, learning loop, agents, and operator documents
+- [Integrations](docs/integrations.md)—channels, AI providers, work systems, and report sources
+- [Reports and proactive checks](docs/reports-and-assistant.md)—the report pipeline, AI-written source cards, and what Taskuary watches
+- [Status and roadmap](docs/roadmap.md)—what works today and what is next
+- [Contributing](CONTRIBUTING.md)—development setup and contribution guide
+
+Taskuary is free and open source under the [MIT License](LICENSE). Issues and pull requests
+are welcome; security reports belong in [SECURITY.md](SECURITY.md).

{taskuary-0.3.3.0 → taskuary-0.3.3.2}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "taskuary"
-version = "0.3.3.0"
+version = "0.3.3.2"
 description = "Automate your job - the local-first operating system for your work: AI triage in, coding agents out."
 readme = "README.md"
 license = { text = "MIT" }

{taskuary-0.3.3.0 → taskuary-0.3.3.2}/taskuary/concierge.py

@@ -1005,23 +1005,35 @@ def _sender_for(store, words: list) -> str:
     return max(tally, key=tally.get) if tally else ''
 
 
+# "skip all the fyi from Chana and Dovid" names a LANE, not a subject. Read as subject words it
+# wrote "fyi from chana@ours.com" and "dovid from dovid@ours.com" - one rule about a word nobody
+# writes in a subject line, one about a man's own name (the 2026-09-03 break test).
+LANE_RULE_WORDS = {'fyi': 'fyi', 'fyis': 'fyi', 'report': 'report', 'reports': 'report'}
+
 def _rules_from(swept: list) -> list:
-    """One rule per sender, carrying the words that hit THEIR mail: [{'sender', 'words'}]. A sender
-    who was swept on their name alone gets a rule about them; everyone else needs their subject
-    words, so a rule can never quietly grow into "everything from this person"."""
-    by = {}
+    """One rule per sender: [{'sender', 'words'}] or [{'sender', 'lane'}]. A sender swept on their
+    name alone gets a rule about them; a sweep that named a LANE gets that lane from that sender;
+    everyone else needs their subject words, so a rule can never quietly grow into "everything from
+    this person"."""
+    by, lanes = {}, {}
     for x in swept:
         key = x['email'] or (x['who'] or '').lower()
         if not key: continue
         by.setdefault(key, set()).update(x['words'])
+        for w in x['words']:
+            if w in LANE_RULE_WORDS: lanes.setdefault(key, set()).add(LANE_RULE_WORDS[w])
     out = []
     for key, words in by.items():
-        subject = sorted(w for w in words if not any(w in part for part in re.split(r'[@.\s]+', key)))
+        if lanes.get(key):
+            out.append({'sender': key, 'lane': sorted(lanes[key])[0], 'words': []}); continue
+        own = set(re.split(r'[@.\s]+', key))
+        subject = sorted(w for w in words if w not in LANE_RULE_WORDS and not any(w in part for part in own))
         out.append({'sender': key, 'words': subject or sorted(words)})
     return out
 
 
 def _rule_words(rule: dict) -> str:
+    if rule.get('lane'): return f"every {rule['lane']} from {rule['sender']}"
     return f"{' '.join(rule['words'][:4])} from {rule['sender']}" if rule.get('sender') else ' '.join(rule['words'][:4])
 
 
@@ -1153,7 +1165,10 @@ def _carry_out(store, tid: int, text: str, words: dict, item0: dict | None, acto
             return {'say': say_, 'options': [], 'decision': None}
         say_ = (f"Split: {out['ref']} keeps \"{out['kept']}\" and {out['newRef']} is \"{out['title']}\". "
                 'Each is its own job now, so an agent sent at one only gets that one.'
-                if out.get('newRef') else "There is only one ask in that one - nothing to split.")
+                if out.get('newRef') else
+                ('I can only see one ask in that one' + (f" - {out['why']}" if out.get('why') else '')
+                 + '. Tell me the second job in your own words and I will make it, or open '
+                 + f"{out.get('ref') or 'the task'} and split it there. It stays on the table either way."))
         record(store, tid, 'assistant', say_)
         return {'say': say_, 'options': [], 'decision': {'verb': 'split', **out}}
     if words['verb'] == 'remember':
@@ -1248,8 +1263,12 @@ def split_item(store, item: dict, text: str, actor: str = 'owner') -> dict:
         new = ingest.split_message(store, item['mid'], actor)
         t = store.get_task(new) or {}
         return {'ref': None, 'kept': '', 'newRef': task_ref(new), 'taskId': new, 'title': t.get('Title') or ''}
-    prop = reshape.propose_split(store, tid, brain(store, fast=True))
-    if not prop.get('two') or not (prop.get('second') or {}).get('title'): return {'ref': task_ref(tid), 'kept': (store.get_task(tid) or {}).get('Title') or ''}
+    # NOT the fast brain: reading two jobs out of one mail is a judgement, and the two-sentence
+    # gear missed a mail that plainly carried two ("add Priya to the list" + "remove the Bulk
+    # Approve button") and answered "only one ask" (the 2026-09-03 break test)
+    prop = reshape.propose_split(store, tid, brain(store))
+    if not prop.get('two') or not (prop.get('second') or {}).get('title'):
+        return {'ref': task_ref(tid), 'kept': (store.get_task(tid) or {}).get('Title') or '', 'why': prop.get('why') or ''}
     new = reshape.split_task(store, tid, prop['second'], prop.get('first'), prop.get('move_message_ids') or [], actor)
     return {'ref': task_ref(tid), 'kept': (store.get_task(tid) or {}).get('Title') or '',
             'newRef': task_ref(new), 'taskId': new, 'title': (store.get_task(new) or {}).get('Title') or ''}

{taskuary-0.3.3.0 → taskuary-0.3.3.2}/taskuary/funnel.py

@@ -44,6 +44,7 @@ LANE_WORDS = {'blocked': ('agent waiting', 'you'), 'time': ('coming up', 'workin
               'asked': ('asked you', 'working'), 'forgotten': ('slipped', 'info'), 'report': ('landed', 'info'), 'fyi': ('fyi', None),
               'working': ('agent working', 'working')}   # in hand: at the very top, nothing to do until the agent stops or asks
 SOON_MIN, ALERT_MIN = 120, 15     # a meeting inside two hours is time-sensitive; inside fifteen it interrupts
+SETUP_GRACE_MIN = 15              # a walk-through the owner is still in does not raise its own hand
 LATER_HOURS = 3                   # "not now" - it comes back this much later
 FEED_DAYS = 7
 # the owner's two knobs (Settings > Assistant): how far back the pipe reaches for ordinary mail - a
@@ -96,6 +97,9 @@ def muted(rule: dict, i: dict) -> bool:
     from .routing import tokens
     key = str(rule.get('sender') or '').lower()
     if key and key not in (str(i.get('email') or '').lower(), str(i.get('who') or '').lower()): return False
+    # a rule can name a LANE rather than words: "skip all the fyi from Chana" is every fyi she sends,
+    # not the mails with 'fyi' in the subject (2026-09-03)
+    if rule.get('lane'): return i.get('lane') == rule['lane'] and bool(key)
     words = [w for w in (rule.get('words') or []) if w]
     if not words: return bool(key)
     return like(words, set(tokens(f"{i.get('who') or ''} {i.get('email') or ''} {i.get('title') or ''}"))) >= min(2, len(words))
@@ -113,23 +117,41 @@ def _item(key, kind, lane, title, *, who='', when='', since='', why='', mid=None
 
 
 # ── the producers: each reads one thing the hub holds ─────────────────────────────────────────
+def _feed_skip(r: dict) -> bool:
+    """Rows that are nobody asking anything: our own sends, withdrawn lines, an auto-reply, and
+    anything on a task that is over (unless a draft on it still waits for a yes)."""
+    if r.get('Direction') == 'out' or r.get('MsgStatus') == 'withdrawn' or r.get('Channel') == 'assistant': return True
+    if r.get('TaskStatus') in ('done', 'dropped') and r.get('ReviewStatus') != 'pending': return True
+    return bool(_OOO.match(str(r.get('Subject') or '')))
+
+
+def thread_speaker(rows: list) -> dict:
+    """Which row speaks for each conversation: {cid: MessageId}. A draft waiting for a yes speaks
+    whatever its age, then an agent parked on a question, then the newest line. Reviews and agents
+    used to be EXEMPT from the one-line-per-thread rule rather than winning it, so the older mail on
+    the same thread came up again as "asked you" - TQ-0002 twice after the wrap, Yosef three times
+    in the All list (the 2026-09-03 break test)."""
+    best = {}
+    for r in rows:
+        cid = r.get('ConversationId')
+        if not cid or _feed_skip(r): continue
+        rank = 2 if (r.get('ReviewStatus') == 'pending' and r.get('ReviewId')) else 1 if r.get('AgentWaiting') else 0
+        if cid not in best or rank > best[cid][0]: best[cid] = (rank, r['MessageId'])   # newest first, so ties keep the newest
+    return {cid: mid for cid, (_rank, mid) in best.items()}
+
+
 def from_feed(store, rows: list) -> list:
     out, agents, reviews, threads = [], set(), set(), {}
+    speaks, more = thread_speaker(rows), {}
     for r in rows:
-        if r.get('Direction') == 'out' or r.get('MsgStatus') == 'withdrawn' or r.get('Channel') == 'assistant': continue
-        if r.get('TaskStatus') in ('done', 'dropped') and r.get('ReviewStatus') != 'pending': continue
-        if _OOO.match(str(r.get('Subject') or '')): continue           # an auto-reply is nobody asking anything
-        # one line per conversation: rows come newest first, so the newest speaks for the thread
+        if _feed_skip(r): continue
+        # ONE line per conversation, and it says how much of the thread it stands for: two WhatsApp
+        # lines from the same person are one row, and the Timeline showing two of them read as the
+        # pipe having lost one (the owner, 2026-09-03)
         cid = r.get('ConversationId')
-        if cid and r.get('ReviewStatus') != 'pending' and not r.get('AgentWaiting'):
-            # one line per conversation, and it SAYS how much of the thread it stands for: two
-            # WhatsApp lines from the same person are one row, and the Timeline showing two of them
-            # read as the pipe having lost one (the owner, 2026-09-03)
-            if cid in threads:
-                held = threads[cid]                 # None when the thread's first row made no item at all
-                if held is not None: held['more'] = held.get('more', 0) + 1
-                continue
-            threads[cid] = None                     # filled in below, once the item exists
+        if cid and speaks.get(cid) != r['MessageId']:
+            more[cid] = more.get(cid, 0) + 1
+            continue
         who = r.get('FromName') or r.get('FromEmail') or r.get('SourceName') or r.get('Channel') or ''
         base = dict(who=who, when=r.get('SentAt'), mid=r['MessageId'], tid=r.get('TaskId'), channel=r.get('Channel') or '',
                     category=r.get('Category') or '', preview=r.get('Preview'), cid=cid, email=r.get('FromEmail') or '')
@@ -148,14 +170,23 @@ def from_feed(store, rows: list) -> list:
                                  else ('a reply is drafted for you to send' if r.get('HasDraft') else 'a reply is owed - draft it with AI or write it'),
                              draft=bool(r.get('HasDraft')), summary=agent_found(store, r.get('TaskId')),
                              sig=hashlib.sha1(str(rv.get('DraftText') or '').encode()).hexdigest()[:10],   # the draft's fingerprint: a rewrite is news
-                             **base)); continue
+                             **base))
+            if cid: threads[cid] = out[-1]                 # the draft speaks for its thread
+            continue
         if r.get('AgentWaiting') and r.get('TaskId'):
             if r['TaskId'] in agents: continue
             agents.add(r['TaskId'])
             out.append(_item(f"agent:{r['TaskId']}", 'agent', 'blocked', r.get('Title') or subj, agent=r.get('Working') or 'agent',
-                             why=f"{r.get('Working') or 'the agent'} stopped and is waiting on you", **base)); continue
+                             why=f"{r.get('Working') or 'the agent'} stopped and is waiting on you", **base))
+            if cid: threads[cid] = out[-1]
+            continue
         base['working'] = r.get('Working') or ''             # an agent has it: build() lets these go, by name
         if r.get('Channel') == 'report':
+            # a run whose own first line says it could not summarise (no AI connector) is on the
+            # Timeline and nowhere else - three of them came out of the pipe, one per turn, on a
+            # fresh install's first day (the 2026-09-03 break test)
+            from .reports import NO_BRAIN
+            if NO_BRAIN in str(r.get('Preview') or ''): continue
             sid = report_source_id(store, r.get('SourceName'))
             bad = report_failed(store, sid, subj)
             out.append(_item(f"report:{r['MessageId']}", 'report', 'report', subj, bad=bad, source_id=sid,
@@ -177,7 +208,10 @@ def from_feed(store, rows: list) -> list:
             continue
         if cat == 'info':
             out.append(_item(f"msg:{r['MessageId']}", 'fyi', 'fyi', subj, why=r.get('RouteReason') or 'a person told you something; nothing to do', **base))
-        if cid and cid in threads and threads[cid] is None and out: threads[cid] = out[-1]
+        if cid and out and threads.get(cid) is None: threads[cid] = out[-1]
+    for cid, n in more.items():                   # the rest of each thread, counted on the row that speaks for it
+        held = threads.get(cid)
+        if held is not None: held['more'] = held.get('more', 0) + n
     return out
 
 
@@ -227,6 +261,7 @@ def from_agents(store) -> list:
     """Live sessions parked on a question - whatever the feed window, an agent waiting is waiting."""
     from . import terminal as term, waitroom
     out = []
+    fresh_setup = (datetime.now() - timedelta(minutes=SETUP_GRACE_MIN)).strftime('%Y-%m-%d %H:%M:%S')
     try: live = term.live_sessions(tail=6)
     except Exception: return out
     for t in live:
@@ -234,9 +269,19 @@ def from_agents(store) -> list:
         if not tid: continue
         task = store.get_task(tid) or {}
         if task.get('SourceRef') == 'assistant:dock' or task.get('Status') in ('done', 'dropped'): continue
+        # A WALK-THROUGH the owner just started is a conversation they are IN. It parked because it
+        # said its piece and is waiting for their next line - and the by-the-way bar announced
+        # "assistant stopped on TQ-0009 and is waiting on you" sixty seconds after they opened it
+        # (the 2026-09-03 break test). It comes back as a blocked row only once it has been left.
+        if task.get('SourceRef') == 'assistant:setup' and _ts(t.get('started')) >= fresh_setup: continue
         waiting = t.get('waiting') if t.get('waiting') is not None else (t.get('idle') or 0) >= term.IDLE_WAITING
         if not waiting: continue
         tail = [str(x).strip() for x in (t.get('tail') or []) if str(x).strip()]
+        # ...read off the RENDERED screen when there is one, with the TUI's chrome dropped: the
+        # card showed a theme toolbar where the agent's question belonged (2026-09-03)
+        if t.get('sid'):
+            try: tail = [x for x in term.asking_lines(t['sid'], 4)] or tail
+            except Exception as e: logger.debug(f'funnel: no rendered screen for {t.get("sid")} - {e}')
         asking = waitroom.looks_like_question(tail)
         agent = t.get('agent') or t.get('label') or 'agent'
         out.append(_item(f"agent:{tid}", 'agent', 'blocked', task.get('Title') or f'task {tid}', who=agent, when=t.get('started'),
@@ -391,11 +436,20 @@ def _aged_out(i: dict, now: datetime, hours: int) -> bool:
     return bool(when) and when < now - timedelta(hours=hours)
 
 
+RUN_STALE_MIN = 20        # a 'running' run row nobody has touched for this long is not working anything
+
 def working_tids(store) -> set:
-    """Tasks an agent has right now - a live session or a running headless run. Nothing about them is
-    the owner's to do until the agent stops."""
+    """Tasks an agent has right now - a live session, or a headless run that is actually running.
+    Nothing about them is the owner's to do until the agent stops.
+
+    A run row left at 'running' by a session that died used to be proof enough: TQ-0006 sat in the
+    working lane with no session and "nothing for you", and the outage task vanished from the pipe
+    altogether - not read, not offered, not findable (the 2026-09-03 break test). A row nobody has
+    touched for RUN_STALE_MIN is a corpse, not a worker."""
     from . import terminal as term
-    out = {r['TaskId'] for r in store.running_runs() if r.get('TaskId')}
+    fresh = (datetime.now() - timedelta(minutes=RUN_STALE_MIN)).strftime('%Y-%m-%d %H:%M:%S')
+    out = {r['TaskId'] for r in store.running_runs()
+           if r.get('TaskId') and str(r.get('UpdatedAt') or r.get('StartedAt') or '') >= fresh}
     try:
         for t in term.live_sessions(tail=0):
             if not t.get('taskId'): continue
@@ -428,11 +482,23 @@ def build(store, now: datetime = None, keep_surfaced: bool = False) -> dict:
     # an agent mid-job: nothing to do here yet, whatever the mail or the idea says about the task - so it
     # rides at the TOP of the pipe as 'in hand', and drops to the front when the agent stops or asks
     busy = working_tids(store)
+    live_tids = busy | {i['tid'] for i in items if i['kind'] == 'agent' and i.get('tid')}   # working, parked or asking: an agent is on it
+    stale_before = (now - timedelta(minutes=RUN_STALE_MIN)).strftime('%Y-%m-%d %H:%M:%S')
     # ...and a task whose STATUS says in_progress is in the middle of being worked, whether or not a
     # session is alive right now: the owner reads it that way ("it's in middle of working... it should
     # say working so it's not in funnel"), and it comes back to the front the moment the watcher moves
     # it to waiting or the agent asks (2026-09-03).
-    def held(tid): return bool(tid) and (tid in busy or (store.get_task(tid) or {}).get('Status') == 'in_progress')
+    # 'in_progress' means an agent is mid-job, and the owner reads it that way even between sessions
+    # ("it's in middle of working... it should say working so it's not in funnel"). But nobody moves
+    # the status back when a session DIES, so the status alone held a task in the working lane for
+    # ever: TQ-0006 sat there with no session and "nothing for you", and the outage task fell out of
+    # the pipe entirely (the 2026-09-03 break test). Mid-job is a live agent, or a task somebody has
+    # touched in the last RUN_STALE_MIN; anything older is abandoned, and comes back to the owner.
+    def mid_job(tid):
+        t = store.get_task(tid) or {}
+        if t.get('Status') != 'in_progress': return False
+        return tid in live_tids or str(t.get('UpdatedAt') or t.get('CreatedAt') or '') >= stale_before
+    def held(tid): return bool(tid) and (tid in busy or mid_job(tid))
     def in_hand(i): return i['kind'] not in ('agent', 'review', 'action') and (i.get('working') or held(i.get('tid')))
     # ...under the SAME key the parked agent will have (agent:<tid>), so shown-once and the page's live
     # row follow the task through stopping and starting instead of losing it at each change

{taskuary-0.3.3.0 → taskuary-0.3.3.2}/taskuary/ingest.py

@@ -263,6 +263,13 @@ def ingest_message(store, msg: dict, actor: str = 'router', llm=None, file_only:
             with _PENDING_LOCK: _PENDING[mid] = msg
             return {'status': 'queued', 'task_id': None, 'message_id': mid}
 
+    # a chat opener with nothing behind it yet: on the timeline, and no task, no draft, no agent.
+    # The line that follows it carries the ask, and the reader is shown this one as its opening.
+    if is_opener(msg):
+        mid = _land(store, msg, None, 'filed')
+        store.add_route(mid, None, 'file', None, 'an opening line on a chat - waiting for the ask it opens', [], 'triage')
+        logger.info(f"ingest: chat opener filed, waiting for the point - {(msg.get('body') or '')[:40]}")
+        return {'status': 'filed', 'task_id': None, 'message_id': mid}
     r = route(msg, store.snapshots(), float(cfg.get('attach_threshold', 0.42)))
     r = own_thread_only(store, msg, r)
     # ...and on a chat, the room it shares with the task is not a reason to join it
@@ -734,6 +741,22 @@ def is_chat(msg: dict) -> bool:
     return str(msg.get('channel') or '').lower() in CHAT_CHANNELS
 
 
+# "hey" is not an ask. On WhatsApp it became a reply task with a drafted "Hey - what's up?", the
+# real question arrived two lines later and attached to it, and the pipe offered the GREETING for
+# approval (the 2026-09-03 break test). A chat opener waits for the sentence it opens: it is filed
+# on the timeline, and the next line - which exchange_lines hands the reader as context - is the ask.
+_OPENER = re.compile(r"^\s*(hi|hey+|hello+|yo|sup|hiya|morning|good (morning|afternoon|evening)|shalom|hey there|you there|u there"
+                     r"|quick (q|question)|got a (sec|second|minute|min)|are you (there|around|free)|can i ask you something"
+                     r"|knock knock|\W*)\W*$", re.I)
+
+def is_opener(msg: dict) -> bool:
+    """A chat line that opens a conversation and asks nothing - the greeting before the point."""
+    if not is_chat(msg): return False
+    body = ' '.join(str(msg.get('body') or '').split())
+    if not body or len(body) > 60: return False
+    return bool(_OPENER.match(body))
+
+
 def _secs(a: str, b: str) -> float:
     """Seconds between two 'YYYY-MM-DD HH:MM:SS' stamps; inf when either is unreadable, so a
     missing timestamp never passes for "typed a moment ago"."""