taskuary 0.2.0__tar.gz → 0.3.0__tar.gz

{taskuary-0.2.0/taskuary.egg-info → taskuary-0.3.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: taskuary
-Version: 0.2.0
+Version: 0.3.0
 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
@@ -51,6 +51,7 @@ Dynamic: license-file
 [![CI](https://github.com/ldbumble/taskuary/actions/workflows/ci.yml/badge.svg)](https://github.com/ldbumble/taskuary/actions/workflows/ci.yml)
 [![License: MIT](https://img.shields.io/badge/license-MIT-blue.svg)](LICENSE)
 [![Python 3.10+](https://img.shields.io/badge/python-3.10%2B-3776ab.svg)](https://github.com/ldbumble/taskuary)
+[![PyPI](https://img.shields.io/pypi/v/taskuary.svg)](https://pypi.org/project/taskuary/)
 [![PRs welcome](https://img.shields.io/badge/PRs-welcome-brightgreen.svg)](CONTRIBUTING.md)
 [![GitHub stars](https://img.shields.io/github/stars/ldbumble/taskuary?style=social)](https://github.com/ldbumble/taskuary/stargazers)
 
@@ -60,11 +61,11 @@ Dynamic: license-file
 scheduled reports land on one timeline; AI triage says what is real work; the coding CLI
 you already use does it; you approve the result. Runs entirely on your machine.
 
-![Taskuary in use: the timeline of email, chats and scheduled reports; a mail opened to show why triage made it a task and the AI reply drafted for approval; a scheduled report landing with its bar chart and spreadsheet; the Reports pipeline list; and the agent board with live sessions and the note one agent left the next](docs/hero.gif)
+![Taskuary in use: the timeline of email, chats and scheduled reports; a mail opened to show why triage made it a task and the AI reply drafted for approval; a scheduled report landing with its bar chart and spreadsheet; the Reports pipeline list; and the agent board with live sessions and the note one agent left the next](https://raw.githubusercontent.com/ldbumble/taskuary/master/docs/hero.gif)
 
 <sub>Real app, real data — a mail triaged and its reply drafted for your approval, then a scheduled SQL report landing on the same rail with its chart and spreadsheet. Nothing sends without you.</sub>
 
-**Where it is:** early — **v0.2.0**, and moving fast. The funnel, the review queue, the agent
+**Where it is:** early — **v0.3.0**, and moving fast. The funnel, the review queue, the agent
 sessions and the reports pipeline are all real and in daily use on my own inbox; the edges are
 still being knocked off, and breaking changes are possible before 1.0. Issues get answered.
 [Full status and roadmap ↓](#status--roadmap)
@@ -92,7 +93,7 @@ sender (yours to review under Settings → Agent memory).
 
 The general lessons take a stricter road, so one odd Tuesday never becomes a rule:
 
-![Your verdicts become a hypothesis with score s:2; agreeing verdicts add a point and contradictions remove one; at s:4 with proof from two or more people it is promoted into LEARNED.md, which then rides into every triage, draft and agent run](docs/learning-loop.svg)
+![Your verdicts become a hypothesis with score s:2; agreeing verdicts add a point and contradictions remove one; at s:4 with proof from two or more people it is promoted into LEARNED.md, which then rides into every triage, draft and agent run](https://raw.githubusercontent.com/ldbumble/taskuary/master/docs/learning-loop.svg)
 
 **How the memory works, concretely.** Each lesson is one line in `LEARNED.md` (Docs tab) —
 a guess with a score. Say you strip the greeting off three drafts this week; the file soon
@@ -129,7 +130,7 @@ markers always survive. What each one feeds from then on:
 ## Get started
 
 ```bash
-pip install git+https://github.com/ldbumble/taskuary
+pip install taskuary
 taskuary        # opens http://127.0.0.1:7787
 ```
 
@@ -161,10 +162,22 @@ No cloud key at all? Set **Settings → Triage & routing → Triage brain** to y
 and skip step 1 — one brain does everything, slower and pricier per message. See
 [One brain or two](#one-brain-or-two).
 
-Prefer a desktop app? `pip install "taskuary[desktop] @ git+https://github.com/ldbumble/taskuary"`
+Prefer a desktop app? `pip install "taskuary[desktop]"`
 then `taskuary-desktop` — the same UI in a native window. A prebuilt single-file
 `Taskuary.exe` is attached to every CI run.
 
+Prefer Docker? No Python install on the machine:
+
+```bash
+git clone https://github.com/ldbumble/taskuary && cd taskuary
+docker compose up
+# http://127.0.0.1:7787 — data lives in the taskuary-data volume
+```
+
+The container is the web app (Timeline, Review, Reports, Connectors). Coding CLIs
+(Claude Code, Codex, …) and the WhatsApp bridge stay on the host — they are programs on
+*your* machine. Publish the port past localhost only with `TASKUARY_TOKEN` set.
+
 ## The workspace
 
 One tab per question, two lines each; the details live in the app's own help text.
@@ -178,7 +191,8 @@ One tab per question, two lines each; the details live in the app's own help tex
   right now — a live session counts as working, and a session gone quiet moves its card to
   *waiting on you* with the question showing. Cards working now show a live peephole and
   the files their agent has modified so far; a queued card says whom it waits behind (see
-  [Many agents, one repo](#many-agents-one-repo--no-stepping-on-each-other)).
+  [Many agents, one repo](#many-agents-one-repo--no-stepping-on-each-other)). Flip it to
+  **Studio** for the same board as a floor you can walk around — below.
 - **Tasks** — **the page is a terminal**: your CLI in the task's repo, prompt typed in and
   sent, and you keep talking. Taskuary picks the checkout from the SOUL.md repo map (one
   click to override); the prompt carries the ask, the mail, the files and the rules, so the
@@ -213,9 +227,20 @@ Two principles hold everywhere: **nothing sends or ships without your approval**
 it works the mail (auto-dispatch + auto-draft, both switchable); triage is AI-gated, so
 with no AI connected messages file visibly instead of heuristics spraying tasks.
 
+## The floor — capacity you can see
+
+![The Studio view: an isometric office where each desk is a task and the agent at it is drawn doing what it is actually doing — hunched at a lit screen for a coding session, a pen on a form for a task with no code in it, hand up when it has stopped and is waiting on you. An empty desk is capacity going unused.](https://raw.githubusercontent.com/ldbumble/taskuary/master/docs/screenshot-floor.png)
+
+"How much can run at once" is a number in Settings — **Agents at once**, four by default —
+and on the Board's **Studio** toggle it stops being a number. One desk per slot: a desk with
+somebody at it is a task being worked, an empty desk is capacity you are not using, and the
+posture says which kind of work it is rather than making you read a chip. Drag to turn the
+room, scroll to zoom, click a desk to fly to it; raise the limit in Settings and the floor
+widens rather than crowding.
+
 ## Many agents, one repo — no stepping on each other
 
-![Two agents share one checkout: each working card shows the files ITS agent has modified (claude in the theme files, codex in the report code and its tests), and a third task waits in Queued with the reason written on the card — waiting on TQ-0009, both would modify ReportsView.jsx, starts by itself when it can](docs/screenshot-board.png)
+![Two agents share one checkout: each working card shows the files ITS agent has modified (claude in the theme files, codex in the report code and its tests), and a third task waits in Queued with the reason written on the card — waiting on TQ-0009, both would modify ReportsView.jsx, starts by itself when it can](https://raw.githubusercontent.com/ldbumble/taskuary/master/docs/screenshot-board.png)
 
 Auto-dispatch can put several CLIs to work at once — and the board keeps them out of each
 other's way with three light moves. No locks, no worktrees, no manager agent:
@@ -265,7 +290,7 @@ Plain markdown, all on the Docs tab, all yours to edit. Three you write, two wri
 themselves, and two can **bootstrap themselves from your mail history** (`TRIAGE.md` and
 `STYLE.md` — the Generate from history button). Each feeds exactly the calls it belongs in.
 
-![TRIAGE.md, STYLE.md, SOUL.md and LEARNED.md feed triage and replies on the cheap model; SOUL.md, CODER.md and LEARNED.md feed coding agents on your CLI; DIGEST.md is your own morning read — and TRIAGE.md and STYLE.md can be generated from three months of your own mail](docs/six-docs.svg)
+![TRIAGE.md, STYLE.md, SOUL.md and LEARNED.md feed triage and replies on the cheap model; SOUL.md, CODER.md and LEARNED.md feed coding agents on your CLI; DIGEST.md is your own morning read — and TRIAGE.md and STYLE.md can be generated from three months of your own mail](https://raw.githubusercontent.com/ldbumble/taskuary/master/docs/six-docs.svg)
 
 | document | what it is | who reads it |
 |---|---|---|
@@ -354,14 +379,17 @@ pyinstaller taskuary.spec   # dist/Taskuary.exe - single-file desktop build
 ```
 
 Data lives in `~/.taskuary/` (override with `TASKUARY_HOME`): `taskuary.db` (SQLite),
-`config.toml`, `taskuary.log`. For LAN use set `[server].token` in config and send it as
-the `X-Taskuary-Token` header. CI runs the test matrix on Windows / Linux / macOS ×
-py3.10 / 3.12 on every push and pull request, plus the web build. The single-file
+`config.toml`, `taskuary.log`. Docker uses `/data` inside the container for the same
+files (`TASKUARY_HOST` / `TASKUARY_PORT` / `TASKUARY_TOKEN` overlay `[server]` at
+runtime only — they are never written back). For LAN use set
+`[server].token` in config (or `TASKUARY_TOKEN`) and send it as the `X-Taskuary-Token`
+header. CI runs the test matrix on Windows / Linux / macOS × py3.10 / 3.12 on every
+push and pull request, plus the web build and a Docker image smoke. The single-file
 exe is built on push to master.
 
 ## Status / roadmap
 
-Early (v0.2.0) and moving fast — said up top too, because it should not be something you find
+Early (v0.3.0) and moving fast — said up top too, because it should not be something you find
 out at the bottom.
 
 - [x] AI-gated triage, review queue, resumable agent sessions, hash-chained audit

{taskuary-0.2.0 → taskuary-0.3.0}/README.md

@@ -3,6 +3,7 @@
 [![CI](https://github.com/ldbumble/taskuary/actions/workflows/ci.yml/badge.svg)](https://github.com/ldbumble/taskuary/actions/workflows/ci.yml)
 [![License: MIT](https://img.shields.io/badge/license-MIT-blue.svg)](LICENSE)
 [![Python 3.10+](https://img.shields.io/badge/python-3.10%2B-3776ab.svg)](https://github.com/ldbumble/taskuary)
+[![PyPI](https://img.shields.io/pypi/v/taskuary.svg)](https://pypi.org/project/taskuary/)
 [![PRs welcome](https://img.shields.io/badge/PRs-welcome-brightgreen.svg)](CONTRIBUTING.md)
 [![GitHub stars](https://img.shields.io/github/stars/ldbumble/taskuary?style=social)](https://github.com/ldbumble/taskuary/stargazers)
 
@@ -12,11 +13,11 @@
 scheduled reports land on one timeline; AI triage says what is real work; the coding CLI
 you already use does it; you approve the result. Runs entirely on your machine.
 
-![Taskuary in use: the timeline of email, chats and scheduled reports; a mail opened to show why triage made it a task and the AI reply drafted for approval; a scheduled report landing with its bar chart and spreadsheet; the Reports pipeline list; and the agent board with live sessions and the note one agent left the next](docs/hero.gif)
+![Taskuary in use: the timeline of email, chats and scheduled reports; a mail opened to show why triage made it a task and the AI reply drafted for approval; a scheduled report landing with its bar chart and spreadsheet; the Reports pipeline list; and the agent board with live sessions and the note one agent left the next](https://raw.githubusercontent.com/ldbumble/taskuary/master/docs/hero.gif)
 
 <sub>Real app, real data — a mail triaged and its reply drafted for your approval, then a scheduled SQL report landing on the same rail with its chart and spreadsheet. Nothing sends without you.</sub>
 
-**Where it is:** early — **v0.2.0**, and moving fast. The funnel, the review queue, the agent
+**Where it is:** early — **v0.3.0**, and moving fast. The funnel, the review queue, the agent
 sessions and the reports pipeline are all real and in daily use on my own inbox; the edges are
 still being knocked off, and breaking changes are possible before 1.0. Issues get answered.
 [Full status and roadmap ↓](#status--roadmap)
@@ -44,7 +45,7 @@ sender (yours to review under Settings → Agent memory).
 
 The general lessons take a stricter road, so one odd Tuesday never becomes a rule:
 
-![Your verdicts become a hypothesis with score s:2; agreeing verdicts add a point and contradictions remove one; at s:4 with proof from two or more people it is promoted into LEARNED.md, which then rides into every triage, draft and agent run](docs/learning-loop.svg)
+![Your verdicts become a hypothesis with score s:2; agreeing verdicts add a point and contradictions remove one; at s:4 with proof from two or more people it is promoted into LEARNED.md, which then rides into every triage, draft and agent run](https://raw.githubusercontent.com/ldbumble/taskuary/master/docs/learning-loop.svg)
 
 **How the memory works, concretely.** Each lesson is one line in `LEARNED.md` (Docs tab) —
 a guess with a score. Say you strip the greeting off three drafts this week; the file soon
@@ -81,7 +82,7 @@ markers always survive. What each one feeds from then on:
 ## Get started
 
 ```bash
-pip install git+https://github.com/ldbumble/taskuary
+pip install taskuary
 taskuary        # opens http://127.0.0.1:7787
 ```
 
@@ -113,10 +114,22 @@ No cloud key at all? Set **Settings → Triage & routing → Triage brain** to y
 and skip step 1 — one brain does everything, slower and pricier per message. See
 [One brain or two](#one-brain-or-two).
 
-Prefer a desktop app? `pip install "taskuary[desktop] @ git+https://github.com/ldbumble/taskuary"`
+Prefer a desktop app? `pip install "taskuary[desktop]"`
 then `taskuary-desktop` — the same UI in a native window. A prebuilt single-file
 `Taskuary.exe` is attached to every CI run.
 
+Prefer Docker? No Python install on the machine:
+
+```bash
+git clone https://github.com/ldbumble/taskuary && cd taskuary
+docker compose up
+# http://127.0.0.1:7787 — data lives in the taskuary-data volume
+```
+
+The container is the web app (Timeline, Review, Reports, Connectors). Coding CLIs
+(Claude Code, Codex, …) and the WhatsApp bridge stay on the host — they are programs on
+*your* machine. Publish the port past localhost only with `TASKUARY_TOKEN` set.
+
 ## The workspace
 
 One tab per question, two lines each; the details live in the app's own help text.
@@ -130,7 +143,8 @@ One tab per question, two lines each; the details live in the app's own help tex
   right now — a live session counts as working, and a session gone quiet moves its card to
   *waiting on you* with the question showing. Cards working now show a live peephole and
   the files their agent has modified so far; a queued card says whom it waits behind (see
-  [Many agents, one repo](#many-agents-one-repo--no-stepping-on-each-other)).
+  [Many agents, one repo](#many-agents-one-repo--no-stepping-on-each-other)). Flip it to
+  **Studio** for the same board as a floor you can walk around — below.
 - **Tasks** — **the page is a terminal**: your CLI in the task's repo, prompt typed in and
   sent, and you keep talking. Taskuary picks the checkout from the SOUL.md repo map (one
   click to override); the prompt carries the ask, the mail, the files and the rules, so the
@@ -165,9 +179,20 @@ Two principles hold everywhere: **nothing sends or ships without your approval**
 it works the mail (auto-dispatch + auto-draft, both switchable); triage is AI-gated, so
 with no AI connected messages file visibly instead of heuristics spraying tasks.
 
+## The floor — capacity you can see
+
+![The Studio view: an isometric office where each desk is a task and the agent at it is drawn doing what it is actually doing — hunched at a lit screen for a coding session, a pen on a form for a task with no code in it, hand up when it has stopped and is waiting on you. An empty desk is capacity going unused.](https://raw.githubusercontent.com/ldbumble/taskuary/master/docs/screenshot-floor.png)
+
+"How much can run at once" is a number in Settings — **Agents at once**, four by default —
+and on the Board's **Studio** toggle it stops being a number. One desk per slot: a desk with
+somebody at it is a task being worked, an empty desk is capacity you are not using, and the
+posture says which kind of work it is rather than making you read a chip. Drag to turn the
+room, scroll to zoom, click a desk to fly to it; raise the limit in Settings and the floor
+widens rather than crowding.
+
 ## Many agents, one repo — no stepping on each other
 
-![Two agents share one checkout: each working card shows the files ITS agent has modified (claude in the theme files, codex in the report code and its tests), and a third task waits in Queued with the reason written on the card — waiting on TQ-0009, both would modify ReportsView.jsx, starts by itself when it can](docs/screenshot-board.png)
+![Two agents share one checkout: each working card shows the files ITS agent has modified (claude in the theme files, codex in the report code and its tests), and a third task waits in Queued with the reason written on the card — waiting on TQ-0009, both would modify ReportsView.jsx, starts by itself when it can](https://raw.githubusercontent.com/ldbumble/taskuary/master/docs/screenshot-board.png)
 
 Auto-dispatch can put several CLIs to work at once — and the board keeps them out of each
 other's way with three light moves. No locks, no worktrees, no manager agent:
@@ -217,7 +242,7 @@ Plain markdown, all on the Docs tab, all yours to edit. Three you write, two wri
 themselves, and two can **bootstrap themselves from your mail history** (`TRIAGE.md` and
 `STYLE.md` — the Generate from history button). Each feeds exactly the calls it belongs in.
 
-![TRIAGE.md, STYLE.md, SOUL.md and LEARNED.md feed triage and replies on the cheap model; SOUL.md, CODER.md and LEARNED.md feed coding agents on your CLI; DIGEST.md is your own morning read — and TRIAGE.md and STYLE.md can be generated from three months of your own mail](docs/six-docs.svg)
+![TRIAGE.md, STYLE.md, SOUL.md and LEARNED.md feed triage and replies on the cheap model; SOUL.md, CODER.md and LEARNED.md feed coding agents on your CLI; DIGEST.md is your own morning read — and TRIAGE.md and STYLE.md can be generated from three months of your own mail](https://raw.githubusercontent.com/ldbumble/taskuary/master/docs/six-docs.svg)
 
 | document | what it is | who reads it |
 |---|---|---|
@@ -306,14 +331,17 @@ pyinstaller taskuary.spec   # dist/Taskuary.exe - single-file desktop build
 ```
 
 Data lives in `~/.taskuary/` (override with `TASKUARY_HOME`): `taskuary.db` (SQLite),
-`config.toml`, `taskuary.log`. For LAN use set `[server].token` in config and send it as
-the `X-Taskuary-Token` header. CI runs the test matrix on Windows / Linux / macOS ×
-py3.10 / 3.12 on every push and pull request, plus the web build. The single-file
+`config.toml`, `taskuary.log`. Docker uses `/data` inside the container for the same
+files (`TASKUARY_HOST` / `TASKUARY_PORT` / `TASKUARY_TOKEN` overlay `[server]` at
+runtime only — they are never written back). For LAN use set
+`[server].token` in config (or `TASKUARY_TOKEN`) and send it as the `X-Taskuary-Token`
+header. CI runs the test matrix on Windows / Linux / macOS × py3.10 / 3.12 on every
+push and pull request, plus the web build and a Docker image smoke. The single-file
 exe is built on push to master.
 
 ## Status / roadmap
 
-Early (v0.2.0) and moving fast — said up top too, because it should not be something you find
+Early (v0.3.0) and moving fast — said up top too, because it should not be something you find
 out at the bottom.
 
 - [x] AI-gated triage, review queue, resumable agent sessions, hash-chained audit

{taskuary-0.2.0 → taskuary-0.3.0}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "taskuary"
-version = "0.2.0"
+version = "0.3.0"
 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-0.3.0/taskuary/__init__.py

@@ -0,0 +1,27 @@
+"""Taskuary - Automate your job: AI triage in, coding agents out, you in charge."""
+# ONE place holds the version and it is pyproject.toml, because that is the one the git tag has
+# to match (RELEASING.md). A second copy here said 0.2.0 through the whole of 0.2.1: the CLI
+# banner, the API and the header pill all reported it, and the header's own tooltip told the
+# owner to restart - which could never have helped, because restarting reloads the same
+# hardcoded string.
+from pathlib import Path
+
+
+def _version() -> str:
+    # pyproject FIRST, and only a source checkout has one sitting next to the package. It is the
+    # file the number lives in, so it is right the moment you pull - whereas installed metadata
+    # is frozen at `pip install` time: an editable install made at 0.2.0 keeps reporting 0.2.0
+    # through every later pull, which is the exact confusion this whole function exists to end.
+    try:
+        for line in (Path(__file__).parent.parent / 'pyproject.toml').read_text(encoding='utf-8').splitlines():
+            if line.startswith('version'): return line.split('=', 1)[1].strip().strip('"\'')
+    except OSError: pass
+    try:                                  # a real wheel: no pyproject beside it, metadata is the truth
+        from importlib.metadata import PackageNotFoundError, version
+        try: return version('taskuary')
+        except PackageNotFoundError: pass
+    except ImportError: pass
+    return '0.0.0+unknown'                # never a stale number pretending to be a real one
+
+
+__version__ = _version()

{taskuary-0.2.0 → taskuary-0.3.0}/taskuary/agents.py

@@ -59,7 +59,10 @@ def _shim_target(path: str) -> list:
     here = os.path.dirname(path)
     found = []
     for tok in re.findall(r'"([^"]+)"', txt):
-        real = os.path.normpath(tok.replace('%dp0%', here).replace('%~dp0', here))
+        # a batch file always writes \ - a separator on Windows, an ordinary character
+        # everywhere else. Translating it means this parser can be exercised by CI on
+        # Linux and macOS too, rather than only on the platform that has the bug.
+        real = os.path.normpath(tok.replace('%dp0%', here).replace('%~dp0', here).replace(chr(92), os.sep))
         if os.path.isfile(real) and real.lower().endswith(('.exe', '.js')): found.append(real)
     exe = next((f for f in found if f.lower().endswith('.exe')), None)
     js = next((f for f in found if f.lower().endswith('.js')), None)
@@ -194,16 +197,24 @@ def task_context(store, task_id: int) -> str:
         lines += [f"- [{m.get('SentAt')}] {m.get('FromName') or m.get('FromEmail')}: {m.get('Subject') or ''}",
                   f"  {str(m.get('BodyText') or '')[:1500]}"]
     lines += ['', 'Thread:'] + [f"- {c.get('Actor')}: {str(c.get('Body'))[:300]}" for c in d['comments']]
+    mem = memory_block(store, d['messages'])
+    if mem: lines += ['', mem]
     return '\n'.join(lines)
 
 
 def memory_block(store, messages: list) -> str:
-    senders = {(m.get('FromEmail') or '').lower() for m in messages if m.get('FromEmail')}
-    domains = {s.rsplit('@', 1)[-1] for s in senders if '@' in s}
-    hits = [f"- {n['Note']}" for n in store.list_memories()
-            if n['Scope'] == 'global' or (n['Scope'] == 'sender' and (n.get('ScopeKey') or '').lower() in senders)
-            or (n['Scope'] == 'sender_domain' and (n.get('ScopeKey') or '').lower() in domains)]
-    return ('Standing notes (learned from the owner - FOLLOW these):\n' + '\n'.join(hits)) if hits else ''
+    """The standing notes an agent working this thread has to follow - the ones that bear on
+    THIS thread, not every note on file. It used to be every match with no cap at all, which
+    grows without limit as the owner keeps giving verdicts; ranking by what the thread actually
+    says puts the ones that matter at the top and says how many were left out."""
+    from .ingest import relevant_notes
+    text = ' '.join(f"{m.get('Subject') or ''} {m.get('BodyText') or ''}" for m in messages)[:4000]
+    notes, left = relevant_notes(store, [(m.get('FromEmail') or '') for m in messages], text)
+    if not notes: return ''
+    return ('Standing notes (learned from the owner - FOLLOW these):\n'
+            + '\n'.join(f'- {n}' for n in notes)
+            + (f'\n({left} more apply to this thread but did not fit - ask before assuming '
+               'nothing else was said.)' if left else ''))
 
 
 # dispatch() lived here: one open->close HEADLESS run on a task, the CLI working and closing

{taskuary-0.2.0 → taskuary-0.3.0}/taskuary/aws.py

@@ -129,6 +129,44 @@ def discover(store, cfg: dict, connector_id: int, actor: str = 'owner') -> dict:
     return out
 
 
+def catalog(store, service: str = None) -> dict:
+    """What is callable here, WITHOUT calling AWS: botocore ships the service and operation
+    models on disk, so both lists are local metadata - no credentials, no region, no request.
+
+    Typing `service` and `operation` by hand meant knowing boto3's naming from memory and
+    finding out you had it wrong only when a scheduled report failed. `seen` is what discovery
+    actually found in this account, first in the list, because 400-odd alphabetical service
+    names is a haystack rather than a choice. `read` marks the operations that only look:
+    a report is a thing you read, and describe_/list_/get_ is nearly always what you want.
+    """
+    import botocore.session
+    from botocore import xform_name
+    sess = botocore.session.get_session()
+    # the objects discovery actually found, each WITH THE REGION it was found in. Handing back a
+    # bare name was a real bug: a log group is (name, region) - /aws/lambda/x in us-east-2 and the
+    # one in us-east-1 are unrelated groups - so a name picked from a list and run against the
+    # card's FIRST region answers "The specified log group does not exist", which is true and
+    # completely misleading. The region travels with the pick now.
+    objs = [(s['Address'] or '', json.loads(s.get('ConfigJson') or '{}').get('region') or '')
+            for s in store.list_sources(active_only=False) if s['Channel'] == 'aws']
+    groups = sorted(({'name': a[len('logs://'):], 'region': r} for a, r in objs if a.startswith('logs://')),
+                    key=lambda g: (g['region'], g['name']))
+    buckets = sorted(({'name': a[len('s3://'):], 'region': r} for a, r in objs if a.startswith('s3://')),
+                     key=lambda b: b['name'])
+    seen = sorted({'logs' for _ in groups[:1]} | {'s3' for _ in buckets[:1]})
+    out = {'seen': seen, 'services': sorted(sess.get_available_services()),
+           'log_groups': groups, 'buckets': buckets}
+    if service:
+        try:
+            names = [xform_name(o) for o in sess.get_service_model(service).operation_names]
+        except Exception as e:
+            return {**out, 'service': service, 'operations': [], 'error': str(e)[:200]}
+        read = [n for n in names if n.startswith(('list_', 'describe_', 'get_', 'head_', 'query', 'scan', 'select_'))]
+        out.update({'service': service, 'operations': sorted(read) + sorted(set(names) - set(read)),
+                    'read': sorted(read)})
+    return out
+
+
 def poll_source(store, cfg: dict, src: dict, since, llm=None, file_only=False) -> int:
     """One discovered object in tasks/feed mode. s3://bucket -> a Timeline item per NEW
     object; logs://group -> ONE batched item of the new matching events (default pattern:
@@ -212,15 +250,50 @@ def run_s3_object(cfg: dict):
     return rows_out(rows, lim, unit='objects', mine=mine)
 
 
+LOG_PAGE_CAP = 12        # one report is worth this many FilterLogEvents calls, not unbounded
+
+
 def run_cloudwatch_logs(cfg: dict):
     """{"log_group", "pattern", "hours": 24} - recent events from a CloudWatch log group.
     Pattern uses CloudWatch filter syntax: '?ERROR ?Exception' means any of these words."""
     from .reports import row_limit, rows_out
     lim, mine = row_limit(cfg)
     start = int((datetime.now() - timedelta(hours=float(cfg.get('hours') or 24))).timestamp() * 1000)
-    kw = {'logGroupName': cfg['log_group'], 'startTime': start, 'limit': min(lim + 1, 10000)}
+    hours = float(cfg.get('hours') or 24)
+    kw = {'logGroupName': cfg['log_group'], 'startTime': start, 'limit': min(max(lim + 1, 1000), 10000)}
     if cfg.get('pattern'): kw['filterPattern'] = cfg['pattern']
-    ev = client(cfg, 'logs').filter_log_events(**kw).get('events') or []
+    where = regions(cfg)[0]
+    try:
+        cl = client(cfg, 'logs', where)
+        # FilterLogEvents PAGES, and its pages are not full ones. It scans log streams and hands
+        # back whatever that pass found, with a nextToken - so a lambda (one stream per
+        # invocation) answered a single call with nine events while the rest of the day sat behind
+        # the token. "9 events" then read as "a quiet day", which for a report asked to find
+        # ERRORS is the worst way to be wrong. Page until the cap is covered or the token runs
+        # out; PAGE_CAP stops a busy group from turning one report into a thousand calls.
+        ev, tok, pages = [], None, 0
+        while pages < LOG_PAGE_CAP:
+            r = cl.filter_log_events(**kw, **({'nextToken': tok} if tok else {}))
+            ev += r.get('events') or []
+            tok, pages = r.get('nextToken'), pages + 1
+            if not tok or len(ev) > lim: break
+    except Exception as e:
+        # "The specified log group does not exist" is true and useless: it does exist, in another
+        # region. AWS cannot say that because it only knows the region it was asked about - but we
+        # know which one we asked, so we say it.
+        if 'ResourceNotFoundException' not in type(e).__name__ and 'does not exist' not in str(e): raise
+        raise RuntimeError(f"no log group {cfg['log_group']} in {where or 'the default region'} - a log "
+                           'group belongs to ONE region, so the same name in another region is a '
+                           'different group. Pick it from the list again (each entry says its region), '
+                           "or set this source's region to the one it lives in.") from None
+    # newest first: when the cap bites, the recent end of the window is the half worth keeping,
+    # and AWS hands these back oldest-first
+    ev.sort(key=lambda e: e.get('timestamp') or 0, reverse=True)
     rows = [{'at': datetime.fromtimestamp(e['timestamp'] / 1000).strftime('%Y-%m-%d %H:%M:%S'),
              'stream': e.get('logStreamName'), 'message': (e.get('message') or '').strip()[:500]} for e in ev]
-    return rows_out(rows, lim, unit='events', mine=mine)
+    hrs = f'{hours:g}h'
+    head, body = rows_out(rows, lim, unit=f'events in the last {hrs}', mine=mine)
+    # a scan that stopped at the page bound is not "all of it", and only this code knows
+    if tok and len(ev) <= lim:
+        head += f' — stopped after {LOG_PAGE_CAP} pages, so there may be more; narrow it with a filter pattern'
+    return head, body

{taskuary-0.2.0 → taskuary-0.3.0}/taskuary/blackboard.py

@@ -116,11 +116,11 @@ def drain(store):
     """A session ended (or a slot freed up): start what was queued, in arrival order. Anything
     whose blocker is still working stays put; anything whose task moved on is just cleared."""
     from . import terminal as term
-    from .ingest import AUTO_SESSIONS
+    from .ingest import auto_sessions
     if not _DRAINING.acquire(blocking=False): return
     try:
         for q in store.queued_dispatches():
-            if len([t for t in term.SESSIONS.values() if t.alive]) >= AUTO_SESSIONS: return
+            if len([t for t in term.SESSIONS.values() if t.alive]) >= auto_sessions(store): return
             b = q.get('BehindTaskId')
             if b and (term.for_task(b) or any(r['TaskId'] == b for r in store.running_runs())): continue
             t = store.get_task(q['TaskId']) or {}

{taskuary-0.2.0 → taskuary-0.3.0}/taskuary/channels.py

@@ -14,12 +14,18 @@ from .github import _h as gh_headers, list_accessible_repos
 from .ingest import ingest_message
 
 GRAPH = 'https://graph.microsoft.com/v1.0'
-MAIL_SELECT = 'id,subject,from,receivedDateTime,sentDateTime,bodyPreview,body,conversationId,webLink,hasAttachments,isRead'
+MAIL_SELECT = 'id,subject,from,toRecipients,ccRecipients,receivedDateTime,sentDateTime,bodyPreview,body,conversationId,webLink,hasAttachments,isRead'
 
 
 def _cfg(c): return json.loads(c.get('ConfigJson') or '{}')
 
 
+def _addrs(rows) -> list:
+    """The addresses off a Graph recipient list. Triage needs to know whether the mailbox is
+    on the To line or merely in Cc - being copied on other people's work is not an assignment."""
+    return [a for a in ((r.get('emailAddress') or {}).get('address') or '' for r in rows or []) if a]
+
+
 def graph_creds(store, c):
     """Effective Graph credentials for a connector: its own, else the Outlook connector's
     saved app (Teams shares it by design), else the AZURE_* env vars (in graph_token).
@@ -132,6 +138,14 @@ def test_connector(store, cid: int) -> dict:
         elif c['Type'] == 'whatsapp':
             from .messengers import wa_test
             detail = wa_test(store, store.get_connector(c['ConnectorId'], with_secret=True))
+        elif c['Type'] in ('exa', 'tavily', 'firecrawl', 'reader'):
+            # a real call, not a key-shape check: these all fail the same way (401) and the
+            # owner should find that out here rather than from an empty report on Monday
+            from .reports import REGISTRY, resolve_cfg
+            probe = ({'url': 'https://example.com'} if c['Type'] in ('firecrawl', 'reader')
+                     else {'query': 'taskuary local-first ai task hub', 'num': 1})
+            head, _body = REGISTRY[c['Type']](resolve_cfg(store, {**probe, 'type': c['Type'], 'max_rows': 1}))
+            detail = f'{c["Type"]} answered: {head}'
         elif c['Type'] in ('jira', 'asana', 'monday', 'clickup', 'todoist'):
             from . import pm
             detail = pm.test(store, store.get_connector(c['ConnectorId'], with_secret=True))
@@ -411,7 +425,7 @@ def _graph_user(tok, oid, cache):
 _HOSTED = re.compile(r'<img[^>]+src="([^"]*?/hostedContents/[^"]*?)"', re.I)
 
 
-def hosted_images(tok: str, html: str, cap: int = 4) -> list:
+def hosted_images(tok: str, html: str, cap: int = 4, where: str = '') -> list:
     out = []
     for i, url in enumerate(dict.fromkeys(_HOSTED.findall(html or '')).keys()):
         if i >= cap: break
@@ -419,7 +433,14 @@ def hosted_images(tok: str, html: str, cap: int = 4) -> list:
             r = requests.get(unescape(url), headers={'Authorization': f'Bearer {tok}'}, timeout=30)
             r.raise_for_status()
         except Exception as e:
-            logger.warning(f'teams hosted image fetch failed: {e}')
+            # one readable line, not the signed hostedContents URL twice over: it is 900
+            # characters of query string, it says nothing the status code does not, and two
+            # of them per message buried the rest of the sync log. 403 here is the membership
+            # check described below - expected, not a fault to hunt.
+            code = getattr(getattr(e, 'response', None), 'status_code', None)
+            why = ('Microsoft refused it for this chat (403 - the app is not a member)' if code == 403
+                   else f'HTTP {code}' if code else str(e)[:120])
+            logger.warning(f'teams image {i + 1} not readable{f" in {where}" if where else ""}: {why}')
             continue
         ct = (r.headers.get('Content-Type') or 'image/png').split(';')[0].strip()
         out.append({'id': hashlib.sha1(url.encode()).hexdigest()[:16],
@@ -453,7 +474,7 @@ def ingest_teams_chats(store, upn: str, tok: str, since, llm=None, file_only=Fal
         # fetched BEFORE triage, like the mail path: the classifier has to see the screenshot
         # to judge it, and afterwards is too late
         raw_html = (m.get('body') or {}).get('content') or ''
-        atts = hosted_images(tok, raw_html)
+        atts = hosted_images(tok, raw_html, where=topic or f'chat with {name}')
         # a picture we could not FETCH must not vanish silently the way it used to. Graph
         # refuses hostedContents unless the app registration carries ChatMessage.Read.All, and
         # "the screenshot was the whole ask" is exactly the message you cannot afford to read
@@ -629,7 +650,7 @@ def _since(s, backfill_days: int = 0):
     return min(last, datetime.now() - timedelta(days=backfill_days)) if backfill_days else last
 
 
-def poll_channels(store, backfill_days: int = 0) -> int:
+def poll_channels(store, backfill_days: int = 0, progress=None) -> int:
     """Ingest new items for every connection the owner marked as a TRIGGER, through the
     same triage funnel (incl. the configured AI, if any). A connection without the trigger
     role is still usable by agents and reports - it just never creates work on its own.
@@ -651,6 +672,11 @@ def poll_channels(store, backfill_days: int = 0) -> int:
         if (not roles & {'trigger', 'feed'} and not (c['Type'] == 'github' and _gh_explicit(store))
                 and not (c['Type'] in CLOUD and _cloud_explicit(store, CH2SRC[c['Type']]))): continue
         file_only = 'trigger' not in roles
+        # said as it happens, not at the end: the timeline is refreshing while this runs, so
+        # "reading Outlook" beside rows that are already arriving beats a spinner and a wait
+        if progress:
+            try: progress(c['Type'], n)
+            except Exception: pass
         full = store.get_connector(c['ConnectorId'], with_secret=True)
         try:
             if c['Type'] in ('outlook', 'teams'):
@@ -708,6 +734,7 @@ def poll_channels(store, backfill_days: int = 0) -> int:
                             'external_id': f"graph:{m['id']}", 'channel': 'email',
                             'subject': m.get('subject'), 'body': _body(m),
                             'from_name': frm.get('name'), 'from_email': frm.get('address'),
+                            'to': _addrs(m.get('toRecipients')), 'cc': _addrs(m.get('ccRecipients')),
                             'conversation_id': m.get('conversationId'), 'sent_at': _local(m.get('receivedDateTime') or ''),
                             'source_link': m.get('webLink'), 'source_name': s['Address'],
                             'images': images_for_triage(store, atts)}, llm=llm)