taskuary 0.3.3.2__tar.gz → 0.3.3.4__tar.gz

{taskuary-0.3.3.2/taskuary.egg-info → taskuary-0.3.3.4}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: taskuary
-Version: 0.3.3.2
+Version: 0.3.3.4
 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
@@ -62,7 +62,7 @@ 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=300&release=0.3.3.2)](https://pypi.org/project/taskuary/)
+[![PyPI](https://img.shields.io/pypi/v/taskuary.svg?cacheSeconds=300&release=0.3.3.4)](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)
 
@@ -71,9 +71,9 @@ Dynamic: license-file
 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)
+![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?v=studio)
 
-Taskuary is early—currently **v0.3.3.2**—so breaking changes are still possible before 1.0.
+Taskuary is early—currently **v0.3.3.4**—so breaking changes are still possible before 1.0.
 
 <p align="center">
   <a href="https://taskuary.com/demo/"><img
@@ -85,16 +85,27 @@ Taskuary is early—currently **v0.3.3.2**—so breaking changes are still possi
 
 ## What Taskuary can do
 
-**One Timeline and Assistant**
+**Connect every system. Keep control.**
 
-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.
+Bring your mail, chats, issue trackers, alerts, reports, and AI agents into one control center
+that runs on your machine. Every input lands on a single Timeline with its context, status, and
+available actions intact. Taskuary gives your tools one place to work together while you decide
+what can run, what can send, and what needs your attention.
 
-![Taskuary's unified work pipe and chat, populated only with fictional demo data.](https://raw.githubusercontent.com/ldbumble/taskuary/master/docs/screenshot-chat.png)
+![Taskuary's unified work pipe and chat, populated only with fictional demo data.](https://raw.githubusercontent.com/ldbumble/taskuary/master/docs/screenshot-chat.png?v=0.3.3.2)
 
-**AI triage**
+**Turn incoming work into tasks automatically**
 
-Triage files routine updates, turns real requests into tasks, and puts urgent decisions first.
+AI triage reads every incoming item, understands what it means, and turns actionable requests
+into ready-to-run tasks. Routine updates are filed, questions get draft replies, and urgent
+decisions move to the front. Every choice comes with a reason, and you can change it at any time.
+
+**Watch the work. Approve the outcome.**
+
+Taskuary keeps the system moving and brings you in when your judgment matters. The Assistant
+speaks up when a reply is waiting, a task has gone quiet, a report has failed, or a meeting is
+about to start. Each update arrives once, with the evidence and next actions attached, so you can
+review the work, make the call, and approve what happens next.
 
 **Agents in your own repos**
 
@@ -105,25 +116,25 @@ Send coding work to Claude Code, Codex, Gemini, Cursor, Copilot, or another CLI.
 Watch the terminal, answer questions, review changes, and use the built-in browser without
 losing the session.
 
-**Scheduled checks**
+**Describe a check. Let it run itself.**
 
-Watch databases, files, APIs, cloud logs, reports, or agent skills. Quiet checks stay silent.
+Say what you want watched in plain words and Taskuary builds it: read a folder of CSVs every
+morning, total the AP bills due in the next 30 days, count helpdesk tickets by day. Reports read
+and summarize, on a schedule or on demand, and can come back with an AI summary. Workflows write
+data and keep state. Quiet checks stay quiet, so the only one you hear about is the one that
+failed.
 
 **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)
+![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?v=0.3.3.2)
 
 ### 2. An agent works on it
 
@@ -135,7 +146,7 @@ Tasks move from **Queued** to **Agent working**. The Board shows live progress a
 
 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)
+![The assistant walking the pipe: the item on the table, its drafted reply, and Approve & send waiting on you](https://raw.githubusercontent.com/ldbumble/taskuary/master/docs/screenshot-assistant.png?v=0.3.3.2)
 
 ## Install
 

{taskuary-0.3.3.2 → taskuary-0.3.3.4}/README.md

@@ -1,7 +1,7 @@
 # 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/)
+[![PyPI](https://img.shields.io/pypi/v/taskuary.svg?cacheSeconds=300&release=0.3.3.4)](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)
 
@@ -10,9 +10,9 @@
 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)
+![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?v=studio)
 
-Taskuary is early—currently **v0.3.3.2**—so breaking changes are still possible before 1.0.
+Taskuary is early—currently **v0.3.3.4**—so breaking changes are still possible before 1.0.
 
 <p align="center">
   <a href="https://taskuary.com/demo/"><img
@@ -24,16 +24,27 @@ Taskuary is early—currently **v0.3.3.2**—so breaking changes are still possi
 
 ## What Taskuary can do
 
-**One Timeline and Assistant**
+**Connect every system. Keep control.**
 
-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.
+Bring your mail, chats, issue trackers, alerts, reports, and AI agents into one control center
+that runs on your machine. Every input lands on a single Timeline with its context, status, and
+available actions intact. Taskuary gives your tools one place to work together while you decide
+what can run, what can send, and what needs your attention.
 
-![Taskuary's unified work pipe and chat, populated only with fictional demo data.](https://raw.githubusercontent.com/ldbumble/taskuary/master/docs/screenshot-chat.png)
+![Taskuary's unified work pipe and chat, populated only with fictional demo data.](https://raw.githubusercontent.com/ldbumble/taskuary/master/docs/screenshot-chat.png?v=0.3.3.2)
 
-**AI triage**
+**Turn incoming work into tasks automatically**
 
-Triage files routine updates, turns real requests into tasks, and puts urgent decisions first.
+AI triage reads every incoming item, understands what it means, and turns actionable requests
+into ready-to-run tasks. Routine updates are filed, questions get draft replies, and urgent
+decisions move to the front. Every choice comes with a reason, and you can change it at any time.
+
+**Watch the work. Approve the outcome.**
+
+Taskuary keeps the system moving and brings you in when your judgment matters. The Assistant
+speaks up when a reply is waiting, a task has gone quiet, a report has failed, or a meeting is
+about to start. Each update arrives once, with the evidence and next actions attached, so you can
+review the work, make the call, and approve what happens next.
 
 **Agents in your own repos**
 
@@ -44,25 +55,25 @@ Send coding work to Claude Code, Codex, Gemini, Cursor, Copilot, or another CLI.
 Watch the terminal, answer questions, review changes, and use the built-in browser without
 losing the session.
 
-**Scheduled checks**
+**Describe a check. Let it run itself.**
 
-Watch databases, files, APIs, cloud logs, reports, or agent skills. Quiet checks stay silent.
+Say what you want watched in plain words and Taskuary builds it: read a folder of CSVs every
+morning, total the AP bills due in the next 30 days, count helpdesk tickets by day. Reports read
+and summarize, on a schedule or on demand, and can come back with an AI summary. Workflows write
+data and keep state. Quiet checks stay quiet, so the only one you hear about is the one that
+failed.
 
 **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)
+![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?v=0.3.3.2)
 
 ### 2. An agent works on it
 
@@ -74,7 +85,7 @@ Tasks move from **Queued** to **Agent working**. The Board shows live progress a
 
 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)
+![The assistant walking the pipe: the item on the table, its drafted reply, and Approve & send waiting on you](https://raw.githubusercontent.com/ldbumble/taskuary/master/docs/screenshot-assistant.png?v=0.3.3.2)
 
 ## Install
 

{taskuary-0.3.3.2 → taskuary-0.3.3.4}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "taskuary"
-version = "0.3.3.2"
+version = "0.3.3.4"
 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-0.3.3.2 → taskuary-0.3.3.4}/taskuary/agents.py

@@ -13,6 +13,27 @@ from . import spawn
 from .store import task_ref
 from .clis import preset_args
 
+_CLI_CHILDREN = set()
+_CLI_CHILDREN_LOCK = threading.Lock()
+
+
+def shutdown_cli_children():
+    """Stop headless CLI calls that are still running when Taskuary shuts down.
+
+    Normal calls remove themselves when their one-shot process exits. Keeping the live Popen
+    objects here gives the application lifespan a deterministic cleanup road instead of leaving
+    an invisible Claude/Codex process behind after the server is gone.
+    """
+    with _CLI_CHILDREN_LOCK:
+        children = list(_CLI_CHILDREN)
+        _CLI_CHILDREN.clear()
+    for child in children:
+        try:
+            if child.poll() is None: child.kill()
+        except Exception:
+            pass
+    return len(children)
+
 
 def _git(cwd, *args):
     try:
@@ -371,6 +392,13 @@ def run_cli(profile: dict, prompt: str, trace, resume: str = None, cancel=None,
     # to the saved profile, which is also used to open the interactive terminal TUI).
     is_codex = _cli_name(name) == 'codex' and bool(args) and args[0] in ('exec', 'e')
     if is_codex and '--json' not in args: args.append('--json')
+    # Same reasoning for the trust check: codex refuses to start outside a git repo with "Not
+    # inside a trusted directory and --skip-git-repo-check was not specified", and half of what
+    # this app asks a CLI runs in ~/.taskuary/scratch on purpose - the classifier, the drafter,
+    # STYLE.md generation - none of which is a checkout. Headlessly there is nobody to answer the
+    # trust prompt, so every one of those died before the model was reached (issue #34). The flag
+    # only relaxes WHERE codex may run; the sandbox flags decide what it may do.
+    if is_codex and '--skip-git-repo-check' not in args: args.append('--skip-git-repo-check')
     cmd = _resolve_cmd(name) + args
     # which model works it: profile default, or a per-run override from the UI. The flag
     # name is configurable because every CLI spells it differently (claude/codex: --model).
@@ -393,6 +421,7 @@ def run_cli(profile: dict, prompt: str, trace, resume: str = None, cancel=None,
         p = spawn.popen(cmd, stdin=subprocess.PIPE, stdout=subprocess.PIPE, stderr=subprocess.PIPE,
                              text=True, encoding='utf-8', errors='replace', cwd=cwd, shell=False,
                              env=env)
+        with _CLI_CHILDREN_LOCK: _CLI_CHILDREN.add(p)
     except PermissionError as e:
         # which() found something that cannot be executed from here. "Not installed" sent people
         # off to reinstall a CLI that was already there; the reason is in denied_msg.
@@ -466,6 +495,7 @@ def run_cli(profile: dict, prompt: str, trace, resume: str = None, cancel=None,
         p.wait()
     finally:
         killer.cancel()
+        with _CLI_CHILDREN_LOCK: _CLI_CHILDREN.discard(p)
     err_t.join(5)      # the exit code can land before the stderr reader has appended - 'boom' read as 'no output' on a fast CI box
     if p.returncode != 0:
         if cancel is not None and cancel.is_set(): raise RuntimeError('cancelled')

{taskuary-0.3.3.2 → taskuary-0.3.3.4}/taskuary/assistant.py

@@ -327,7 +327,8 @@ CONTRACT = ('\n\nYou are writing your POST on the owner\'s Timeline - the short
             'already and yours is ignored for those; this is for idea:* lines>", '
             '"why": "<one line: what this rests on - the mail, the date, the silence, the pattern - named as it appears in what you '
             'were given (sender, subject, mid, TQ-ref), so the owner can check it>", "mid": <the message id it is '
-            'about, or null>, "task": "<idea:* only - a task title the owner could accept as-is, or null>"}], '
+            'about, or null>, "task": "<idea:* only - a task title the owner could accept as-is, or null>", '
+            '"kind": "<when task is not null: coding if it needs a repository/system changed, otherwise general>"}], '
             '"notes": "<your note to the next check, under 120 words: FACTS AND TIMINGS ONLY - what you looked at and found nothing in, '
             'the date or silence length at which something becomes worth raising, a fact you settled so it need not be worked out again. '
             'Never a standing rule about what to ignore or what is noise: the instruction decides that, and a note that says '
@@ -537,7 +538,7 @@ def raised(store, days: float = 2) -> str:
     return '\n'.join(out) or '(the assistant raised nothing in this window)'
 
 
-def parse(text: str, cands: list, max_lines: int = MAX_LINES) -> list:
+def parse(store, text: str, cands: list, max_lines: int = MAX_LINES) -> list:
     """The model's list, kept honest: a key it invents must be idea:*, a candidate key keeps its
     kind and its buttons, and the text is the model's when it gave one. Every line keeps its WHY -
     the hub's facts for a candidate (plus the model's read on them), the model's own for an idea -
@@ -558,6 +559,8 @@ def parse(text: str, cands: list, max_lines: int = MAX_LINES) -> list:
             mid = s.get('mid') if isinstance(s.get('mid'), int) else None
             title = _short(s.get('task'), 120) or None
             act = {'type': 'task', 'mid': mid, 'title': title} if title and mid else {'type': 'message', 'mid': mid} if mid else {'type': 'note'}
+            if act['type'] == 'task' and str(s.get('kind') or '').lower() in ('coding', 'general'):
+                act['kind'] = str(s['kind']).lower()
             # a line that NAMES a task belongs to it, whatever else it carries: without the tid the
             # pipe could not tell that an agent had the work, so "TQ-0329 July financials hasn't moved"
             # sat in 'slipped' while the task said in_progress (the owner, 2026-09-03)
@@ -767,7 +770,7 @@ def think(store, cands: list, llm, instruction: str = None, max_lines: int = MAX
         from .llm import readable_images
         images = readable_images(store, _people_context(store)[1])
     text = llm(system, user, max_tokens=POST_TOKENS, **({'images': images} if images else {}))
-    return parse(text, cands, max_lines), _notes(text), user
+    return parse(store, text, cands, max_lines), _notes(text), user
 
 
 def facts(store, watch_source_ids=None, watch_sources=None, systems_only: bool = False) -> str:
@@ -1101,9 +1104,40 @@ def act(store, idea_id: int, verb: str, actor: str = 'owner', llm=None, days: in
     elif verb == 'task':
         if not a.get('mid'): raise ValueError('this idea is not about a message, so there is nothing to make a task from')
         from . import ingest
-        tid = ingest.task_from_message(store, a['mid'], actor, 'coding')
+        message = store.get_message(a['mid']) or {}
+        prior = store.get_task(message.get('TaskId')) if message.get('TaskId') else None
+        # The task kind is the router. The old path always launched `_auto_code`, even when the
+        # message already belonged to a GENERAL task; that is how TQ-0367 was silently reopened
+        # in a repository after it had been completed. An explicit kind from this assistant's
+        # one triage wins, then the existing task kind, with coding retained only as the legacy
+        # default for older idea rows that carry neither.
+        kind = str(a.get('kind') or (prior or {}).get('Kind') or 'coding').lower()
+        if kind not in ('coding', 'general'): kind = 'coding'
+        if prior and prior.get('Status') in ('done', 'dropped'):
+            # This is NEW work noticed after an earlier task was completed. A message can point
+            # at only one task, so do not steal it from history or rename/reopen the completed
+            # task. Carry the evidence into a fresh assistant-owned task instead.
+            title = str(a.get('title') or i.get('Text') or 'Assistant follow-up').strip()[:200]
+            source_text = str(message.get('BodyText') or '').strip()
+            summary = str(i.get('Text') or '').strip()
+            if source_text and source_text not in summary: summary += f'\n\nSource message: {source_text}'
+            tid = store.create_task({'Title': title, 'Summary': summary[:2000], 'Kind': kind,
+                                     'Source': 'assistant', 'SourceRef': f'assistant:idea:{idea_id}'}, actor)
+            store.add_comment(tid, 'assistant', 'assistant_agent',
+                              f'Created from assistant idea {idea_id}; related completed task {task_ref(prior["TaskId"])}.')
+            store.audit('task', tid, 'create_from_assistant_idea', actor,
+                        detail={'idea_id': idea_id, 'message_id': a['mid'], 'related_task_id': prior['TaskId']})
+        else:
+            tid = ingest.task_from_message(store, a['mid'], actor, kind)
         if a.get('title'): store.update_task(tid, {'Title': str(a['title'])[:200]}, actor)
-        if store.get_settings().get('coder_auto_enabled') == '1': ingest._spawn(ingest._auto_code, store, tid)
+        task = store.get_task(tid) or {}
+        brief = str(a.get('title') or i.get('Text') or task.get('Title') or '').strip()
+        why = str(a.get('why') or '').strip()
+        if why: brief += f'\n\nWhy the assistant raised it: {why}'
+        if kind == 'general':
+            ingest._spawn(_auto_general, store, tid, brief)
+        elif store.get_settings().get('coder_auto_enabled') == '1':
+            ingest._spawn(ingest._auto_code, store, tid)
         out |= {'taskId': tid, 'ref': task_ref(tid)}
     elif verb == 'snooze':
         until = (datetime.now() + timedelta(days=max(1, int(days or 1)))).strftime('%Y-%m-%d %H:%M:%S')
@@ -1119,3 +1153,15 @@ def act(store, idea_id: int, verb: str, actor: str = 'owner', llm=None, days: in
         else: learn.learn_from(store, ev)
     store.audit('idea', idea_id, verb, actor, detail={'kind': i.get('Kind')})
     return out
+
+
+def _auto_general(store, tid: int, brief: str) -> None:
+    """The regular-agent counterpart to ingest._auto_code for an accepted assistant idea."""
+    from . import general
+    try:
+        session = general.start_session(store, tid, actor='router')
+        store.add_comment(tid, 'router', 'agent', 'sent to the regular agent from the assistant')
+        session.send_prompt(brief)
+    except Exception as e:
+        logger.warning(f'regular-agent dispatch failed for task {tid}: {e}')
+        store.add_comment(tid, 'router', 'agent', f'Regular-agent start failed: {str(e)[:200]}')

{taskuary-0.3.3.2 → taskuary-0.3.3.4}/taskuary/blackboard.py

@@ -99,6 +99,22 @@ def wall(store, cwd: str = '', limit: int = 12) -> list:
     return store.notes(norm(cwd) if cwd else None, limit)
 
 
+def live_wall(store, cwd: str = '', limit: int = 60) -> list:
+    """Task notes for sessions alive now, plus the shared house lane.
+
+    Board notes are durable, but the UI calls this surface Live handoff. Filtering only at the
+    daily roll-up left notes from closed sessions looking current for hours (or indefinitely).
+    """
+    from . import terminal as term
+    try:
+        live_tids = {int(s['taskId']) for s in term.live_sessions(tail=0, details=False)
+                     if s.get('taskId')}
+    except Exception:
+        live_tids = set()
+    rows = store.notes(norm(cwd) if cwd else None, max(int(limit) * 5, 300))
+    return [n for n in rows if not n.get('TaskId') or int(n['TaskId']) in live_tids][:int(limit)]
+
+
 def house_wall(store, limit: int = 8) -> list:
     """The lane with no checkout in it: what the assistant chat and the owner leave for
     everyone. A chat has no repository, so this is the whole of its wall."""

{taskuary-0.3.3.2 → taskuary-0.3.3.4}/taskuary/channels.py

@@ -466,6 +466,57 @@ def _same_words(a: str, b: str, n: int = 160) -> bool:
     return bool(x) and x == y
 
 
+def retire_draft_answered_elsewhere(store, tid: int | None, sent: dict) -> list:
+    """Resolve reply drafts made obsolete by a reply the owner sent in the native app.
+
+    A provider sync can land well after triage wrote its draft.  The draft belongs to the
+    inbound message it was written against, so an owner line later on that same conversation
+    is the missing verdict: the reply already went out.  Action approvals are deliberately
+    excluded -- sending a chat line cannot approve an unrelated proposed action.
+    """
+    conv, sent_at = sent.get('ConversationId'), sent.get('SentAt')
+    if not conv or not sent_at:
+        return []
+    stale = [rv for rv in store.list_reviews('pending')
+             if rv.get('TaskId') == tid and rv.get('Kind') != 'action'
+             and rv.get('ConversationId') == conv
+             and (not rv.get('SentAt') or rv['SentAt'] <= sent_at)]
+    if not stale:
+        return []
+
+    place = SENT_FROM.get(sent.get('Channel'), 'outside Taskuary')
+    for rv in stale:
+        store.decide_review(rv['ReviewId'], 'superseded', None, 'you',
+                            f'The owner replied {place}; this draft was not sent.')
+
+    # Treat this exactly like a successful send through Review.  Reply-only work closes, while
+    # owner-controlled work and a task whose coding agent is still running keep their safeguards.
+    if tid:
+        from . import verdicts
+        verdicts._settle_task_after_sent_reply(store, stale[0], 'you', True)
+
+    task = store.get_task(tid) if tid else {}
+    body = re.sub(r'\s+', ' ', sent.get('BodyText') or '').strip()
+    excerpt = f': "{body[:180]}"' if body else ''
+    if not tid:
+        subject = (store.get_message(stale[0].get('MessageId')) or {}).get('Subject') or 'this thread'
+        where = subject
+        state = ''
+    else:
+        where = f'TQ-{tid:04d}'
+        state = ' The task is still open because it has other work in progress.' \
+            if task.get('Status') not in ('done', 'dropped') else ' I marked the task done.'
+    from . import concierge, funnel, general
+    concierge.record(store, general.dock_task(store)[0]['TaskId'], 'assistant',
+                     f'You replied {place} on {where}{excerpt}. '
+                     f'I removed the unused draft; the reply was taken care of.{state}')
+    funnel.invalidate()
+    # The review transition already wakes the UI, but this final wake happens after the durable
+    # Assistant line was written, avoiding a race where an open chat fetched one write too early.
+    store._poke('feed-changed', 'task-changed', task_id=tid)
+    return stale
+
+
 def ingest_own_message(store, msg: dict, why: str, keep_unmatched: bool = True) -> int:
     """Anything YOU sent - a mail reply, a line in a chat - never gets its own timeline row and
     never becomes work: it rides INSIDE the thread as a 'context' message, which is what the
@@ -486,6 +537,8 @@ def ingest_own_message(store, msg: dict, why: str, keep_unmatched: bool = True)
                              'SentAt': msg.get('sent_at'), 'BodyText': msg.get('body'),
                              'SourceLink': msg.get('source_link'), 'Status': 'context',
                              'MailMetaJson': json.dumps(msg.get('mail_meta')) if msg.get('mail_meta') else None})
+    sent = store.get_message(mid) or {}
+    retire_draft_answered_elsewhere(store, tid, sent)
     if not tid: return 1
     store.add_route(mid, tid, 'attach', None, why, [], 'router')
     # ...unless Taskuary is reading back its OWN send. That reply already has its line on the
@@ -745,7 +798,7 @@ def ingest_github_issues(store, src: dict, tok: str, since, llm=None, file_only=
     issues_mode, prs_mode = gh_modes(src, file_only)
     if issues_mode == 'off' and prs_mode == 'off': return 0
     n = 0
-    from .github import list_items
+    from .github import body_images, list_items
     for i in reversed(list_items(tok, repo, since=since.astimezone().isoformat())):
         if TQ_ISSUE.match(i.get('title') or ''): continue
         is_pr = 'pull_request' in i
@@ -761,6 +814,10 @@ def ingest_github_issues(store, src: dict, tok: str, since, llm=None, file_only=
             'from_name': who, 'from_email': f'{who}@users.noreply.github.com',
             'conversation_id': f"gh:{repo}#{i['number']}", 'sent_at': _local(i.get('updated_at') or ''),
             'source_link': i.get('html_url'), 'source_name': repo,
+            # the screenshot IS the report: read it before the row exists, or the classifier
+            # judges an issue template whose headings are empty (see images_for_triage above)
+            'images': (body_images(tok, i.get('body') or '')
+                       if str(store.get_settings().get('vision_enabled') or '1') == '1' else []),
             'no_auto': not gh_auto_ok(src, i.get('author_association'))},
             llm=llm, file_only=mode == 'feed')
         n += out['status'] != 'duplicate'
@@ -897,7 +954,7 @@ def _poll_jobs(store, only=None):
         # tool-only card, a Sync that pulled nothing, and no error anywhere).
         phone_guide = (c['Type'] == 'whatsapp'
                        and store.get_settings().get('phone_assistant') == '1'
-                       and 'notify' in roles)
+                       and bool(_cfg(c).get('assistant_chat') or _cfg(c).get('notify_chat')))
         if (not roles & {'trigger', 'feed'} and not phone_guide
                 and not (c['Type'] == 'github' and _gh_explicit(store))
                 and not (c['Type'] in CLOUD and _cloud_explicit(store, CH2SRC[c['Type']]))): continue

{taskuary-0.3.3.2 → taskuary-0.3.3.4}/taskuary/clis.py

@@ -54,9 +54,19 @@ READONLY = {'claude': (('--dangerously-skip-permissions',), ('--tools', '')),
 # classifier above, which gets no tools at all because the text it classifies is untrusted input.
 # Claude's built-ins are named explicitly, and every MCP tool is denied because MCP permissions
 # are separate from --tools and a connector may expose writes beside reads.
+REPORT_TOOLS = 'Read,Glob,Grep,WebFetch,WebSearch'
+
+# --tools and --allowedTools answer different questions and a report needs BOTH: --tools says which
+# built-ins EXIST in this run, --allowedTools says which may be used without asking a human. With
+# the bypass flag dropped and only --tools given, WebFetch and WebSearch existed but still prompted,
+# and a headless report has nobody to click: every fetch came back "Claude requested permissions to
+# use WebFetch, but you haven't granted it yet" and the run narrated the refusal instead of the
+# trending page. Read/Glob/Grep never prompt, which is why the gap stayed invisible for two months
+# (the GitHub Trending report, 2026-09-04). Granting the same five is not a widening - the set the
+# run may use is still exactly the set that exists.
 REPORT_READ = {
     'claude': (('--dangerously-skip-permissions',),
-               ('--tools', 'Read,Glob,Grep,WebFetch,WebSearch', '--disallowedTools', 'mcp__*')),
+               ('--tools', REPORT_TOOLS, '--allowedTools', REPORT_TOOLS, '--disallowedTools', 'mcp__*')),
     'codex': READONLY['codex'],
     'gemini': READONLY['gemini'],
 }

{taskuary-0.3.3.2 → taskuary-0.3.3.4}/taskuary/coder.py

@@ -224,8 +224,8 @@ def wrap(store, tid: int, close: bool = True, actor: str = 'owner', sid: str = N
     report = resolution_text(rep)
     store.add_comment(tid, actor, 'human', 'Closed the session - wrapped up from what was on screen.')
     store.add_comment(tid, agent, 'agent', f'CODER REPORT\n{report}')
-    # The compact result stays deliberately scannable. The transcript does not disappear behind
-    # that compression: preserve it as a Markdown artifact and link it from the same Agent work card.
+    # The compact result and final answer become the readable Markdown artifact. terminal.py has
+    # already filed the raw PTY stream for recovery; duplicating it here made the reader unusable.
     artifact = None
     try:
         from . import session_artifacts
@@ -283,7 +283,15 @@ def raise_reply(store, task_id: int, mid: int, run_id: int, rep: dict,
                                 'Status': 'pending', 'Reason': 'coder finished the work - reply awaiting approval'})
     # The report is intentionally compact UI copy. Draft from the complete final response (or
     # transcript fallback), so every question the session answered remains available here.
-    try: responder.write_draft(store, task_id, rid, complete_result or resolution_text(rep), 'coder')
+    try:
+        try: deliver = json.loads((held or {}).get('Deliver') or '{}') or {}
+        except (TypeError, ValueError): deliver = {}
+        if deliver.get('channel') and deliver.get('kind') != 'zoho_invoice':
+            from . import outbox
+            outbox.redraft_review(store, store.get_review(rid),
+                                  complete_result or resolution_text(rep))
+        else:
+            responder.write_draft(store, task_id, rid, complete_result or resolution_text(rep), 'coder')
     except Exception as e: logger.warning(f'reply draft failed for task {task_id}: {e}')
     # the ping that matters most: work FINISHED and its reply is sitting in Review on you
     if (store.get_settings().get('notify_level') or 'needs_me') != 'off':