@akshar5/cohall 0.9.0 → 0.10.0

package/CHANGELOG.md

@@ -1,5 +1,44 @@
 # Changelog
 
+## [0.10.0](https://github.com/AksharP5/cohall/compare/v0.9.0...v0.10.0) (2026-10-04)
+
+
+### Features
+
+* **tasks:** answer worker questions without restarting tasks ([#107](https://github.com/AksharP5/cohall/issues/107)) ([02c11a2](https://github.com/AksharP5/cohall/commit/02c11a27ca5b419eb721ed866331d350322406b8))
+* **tasks:** recover submissions after a lost response ([#113](https://github.com/AksharP5/cohall/issues/113)) ([ced2061](https://github.com/AksharP5/cohall/commit/ced20612e70b44d653b6d0e3470722aa12524cc8))
+* **tasks:** stop delegated coding work at a deadline ([#111](https://github.com/AksharP5/cohall/issues/111)) ([c221892](https://github.com/AksharP5/cohall/commit/c221892981c14478006fecd86c92e74750fe4f53))
+
+
+### Bug Fixes
+
+* **attachments:** bound local file reads after size checks ([#117](https://github.com/AksharP5/cohall/issues/117)) ([3c170e3](https://github.com/AksharP5/cohall/commit/3c170e3040add5216a835dbd9b7285993ee92f8a))
+* **client:** read large task traces without relaxing other limits ([#138](https://github.com/AksharP5/cohall/issues/138)) ([48b7022](https://github.com/AksharP5/cohall/commit/48b70227717a118f8f091f84b8931e3179728ca5))
+* **device:** clean task files before cancellation completes ([#127](https://github.com/AksharP5/cohall/issues/127)) ([9c7910e](https://github.com/AksharP5/cohall/commit/9c7910e862851b2500602a5f0f1b60fa33f0e6a7))
+* **device:** reconnect when relay connections stall ([#129](https://github.com/AksharP5/cohall/issues/129)) ([c734f1d](https://github.com/AksharP5/cohall/commit/c734f1dcee1f1aa55b6a8883ba61a19aa2e48c21))
+* **device:** reject queued work for disabled providers ([#152](https://github.com/AksharP5/cohall/issues/152)) ([baf441b](https://github.com/AksharP5/cohall/commit/baf441b9feae9398a34d72f0bb8d263ca9298ace))
+* **device:** restore task status after reconnect ([#150](https://github.com/AksharP5/cohall/issues/150)) ([902c49a](https://github.com/AksharP5/cohall/commit/902c49ac626a6b5b606cfce4e0b6cc697c495266))
+* **device:** retry relay switches without starting stopped workers ([#134](https://github.com/AksharP5/cohall/issues/134)) ([5a3817b](https://github.com/AksharP5/cohall/commit/5a3817b911f829a386fc11685f36f2abe3495c57))
+* **device:** stop coding providers before worker exit ([#119](https://github.com/AksharP5/cohall/issues/119)) ([5835dc5](https://github.com/AksharP5/cohall/commit/5835dc5223333836f18ad4251f7d8d0035538934))
+* **discovery:** read large device inventories through bounded pages ([#142](https://github.com/AksharP5/cohall/issues/142)) ([e02bf88](https://github.com/AksharP5/cohall/commit/e02bf884f3d7b139b70510feb5f323e5b76029ae))
+* **doctor:** warn about unavailable gateways in automatic mode ([#146](https://github.com/AksharP5/cohall/issues/146)) ([44cbbdb](https://github.com/AksharP5/cohall/commit/44cbbdb1d4d7c4d1b4bff2d6f7015f2888c2c2eb))
+* **mcp:** preserve results when requester waits are cancelled ([#123](https://github.com/AksharP5/cohall/issues/123)) ([ef7ca4a](https://github.com/AksharP5/cohall/commit/ef7ca4a13eddf1418c64e56df9a4eabd991cb998))
+* **providers:** reject answers lost to event limits ([#122](https://github.com/AksharP5/cohall/issues/122)) ([5d2877b](https://github.com/AksharP5/cohall/commit/5d2877b07d3d4cc9db351a102c0a4e1145126efd))
+* **relay:** preserve Bot questions completed while offline ([#151](https://github.com/AksharP5/cohall/issues/151)) ([024a319](https://github.com/AksharP5/cohall/commit/024a3199a171506963ac888d06bfccbb2ff5a0f7))
+* **relay:** release Bot slots after rejected questions ([#154](https://github.com/AksharP5/cohall/issues/154)) ([6714f25](https://github.com/AksharP5/cohall/commit/6714f25536df87aeaccf8cfa887fccbffc72ac0a))
+* **relay:** settle worker replies after history pruning ([#149](https://github.com/AksharP5/cohall/issues/149)) ([4bec2fd](https://github.com/AksharP5/cohall/commit/4bec2fd1e74873d0964bdf3bf15b8a9e9d03c1ea))
+* **service:** honor the Linux user config directory ([#155](https://github.com/AksharP5/cohall/issues/155)) ([4d74a5e](https://github.com/AksharP5/cohall/commit/4d74a5edb591b7d197c2e26d70b83b06e82ff255))
+* **service:** keep pnpm workers on the upgraded version ([#145](https://github.com/AksharP5/cohall/issues/145)) ([d72fc2c](https://github.com/AksharP5/cohall/commit/d72fc2cac0eadbb1991bdd848780f1a52b37b8a9))
+* **setup:** preserve worker settings when repairing setup ([#153](https://github.com/AksharP5/cohall/issues/153)) ([9845f49](https://github.com/AksharP5/cohall/commit/9845f4982996c920e8f1aa3519264164d859f3bd))
+* **tasks:** finish offline deadlines and confirm cancellation ([#118](https://github.com/AksharP5/cohall/issues/118)) ([639d185](https://github.com/AksharP5/cohall/commit/639d185b41572eb8c67f5f59c6d6fdbceb77a357))
+* **tasks:** honor wait timeouts through relay responses ([#126](https://github.com/AksharP5/cohall/issues/126)) ([7918a3c](https://github.com/AksharP5/cohall/commit/7918a3c96a3c5079f5389b09d7adcc5202e7108b))
+* **tasks:** preserve cancellation acknowledgements after reconnect ([#110](https://github.com/AksharP5/cohall/issues/110)) ([a5d503e](https://github.com/AksharP5/cohall/commit/a5d503e88a4a1585a595e2fea5ef1cf49e354c12))
+* **upgrade:** keep unfinished restarts when changing versions ([#141](https://github.com/AksharP5/cohall/issues/141)) ([32bebb0](https://github.com/AksharP5/cohall/commit/32bebb079cebb66c6731000f9e2c0764cc04f798))
+* **upgrade:** preserve newer installations when latest is older ([#103](https://github.com/AksharP5/cohall/issues/103)) ([1932a6a](https://github.com/AksharP5/cohall/commit/1932a6a359f5481a0ba11f9b6368b46c85c48553))
+* **upgrade:** preserve unfinished restarts and validate recovery ([#137](https://github.com/AksharP5/cohall/issues/137)) ([c5c7930](https://github.com/AksharP5/cohall/commit/c5c7930ee1943bc1dc6bc847ed648b8368a3d2e8))
+* **upgrade:** reject Windows tasks using a different installation ([#131](https://github.com/AksharP5/cohall/issues/131)) ([5c3987f](https://github.com/AksharP5/cohall/commit/5c3987f75ad3532788235c7195e120f65e75276f))
+* **upgrade:** verify global installations before changing them ([#156](https://github.com/AksharP5/cohall/issues/156)) ([068e619](https://github.com/AksharP5/cohall/commit/068e619490173d1bbfcfbebcb218c8559ce61a2a))
+
 ## [0.9.0](https://github.com/AksharP5/cohall/compare/v0.8.0...v0.9.0) (2026-09-30)
 
 

package/README.md

@@ -178,8 +178,10 @@ cohall relay use https://new-relay.example.com
 ```
 
 The switch verifies the existing credentials against the restored relay before
-changing local configuration, then restarts an active device service. Remote
-addresses require HTTPS by default. Use `--allow-http` only when another layer,
+changing local configuration, then restarts an active device service. Stopped
+workers stay stopped. Repeat the command to retry a failed restart, even when
+the address is already saved. Remote addresses require HTTPS by default. Use
+`--allow-http` only when another layer,
 such as Tailscale, already encrypts the connection.
 
 The backup contains the SQLite relay state, the plaintext owner token, and a
@@ -201,20 +203,27 @@ npx -y @akshar5/cohall delegate \
   --context 'Why: the deployment failed after local checks passed. Need: root cause, evidence, and recommended next step.'
 ```
 
-The command waits for a result by default. Queue longer work with `--no-wait`,
-then inspect it later:
+The command waits for a result or a clarification question by default. Queue
+longer work with `--no-wait`, then inspect it later:
 
 ```bash
 npx -y @akshar5/cohall wait <task-id> --timeout 1800
 npx -y @akshar5/cohall trace <task-id> --follow
 ```
 
+If a submission response is lost, retry with a UUID v4 generated before the
+first attempt. Supply it through `--request-id` or MCP `delegate.request_id` and
+keep the input and client credential unchanged. Cohall returns the original
+task's current status. Use a device ID or `@device-id/bot-id` target so discovery
+changes cannot affect the retry. See [retrying submissions](docs/integrations.md#retrying-submissions)
+for limits and retention.
+
 Task status and traces include the worker's latest progress note when it reports
 one. Workers can publish a brief milestone with `cohall progress --message
 "Running tests"`; the task ID is inherited during delegated work.
 
-Completed tasks also appear in the sending client's inbox. This covers queued
-work and tasks that finish after a wait times out:
+Questions awaiting answers and completed tasks appear in the sending client's
+inbox. This covers queued work and tasks that finish after a wait times out:
 
 ```bash
 npx -y @akshar5/cohall inbox
@@ -223,14 +232,31 @@ npx -y @akshar5/cohall inbox ack <task-id>
 ```
 
 The inbox shows short previews; `status` returns the full result. Acknowledge a
-task after handling it. Synchronous `delegate` calls acknowledge their own
-results automatically. The inbox shows up to 20 oldest entries and sets
-`hasMore` when more are waiting; acknowledge handled entries to reveal the rest.
+completed task after handling it. Synchronous `delegate` calls acknowledge their
+completed results automatically. Questions require an answer or cancellation.
+The inbox shows up to 20 oldest entries and sets `hasMore` when more are waiting;
+handle entries to reveal the rest.
 Each client credential has its own inbox, including
 client-only pairings. The relay keeps at most 1,000 completed tasks by default,
 so an old unacknowledged result can leave the inbox when history is pruned.
 Tasks created before the inbox was added have no inbox entry.
 
+If essential information is missing, a worker can run
+`cohall request-input --question 'Which branch should I use?'` and end its turn.
+The task becomes `needs_input`, frees the worker slot, and returns the question
+as `input_request` in `status` or a waiting command. The original requester or
+relay owner answers using its question ID:
+
+```bash
+cohall answer <task-id> --request-id <question-id> --message 'Use main.'
+cohall wait <task-id> --timeout 1800
+```
+
+The same task resumes on the same target with its saved provider session when
+available. Answers remain in its resumed prompt after a restart. Upgrade the
+relay, requester, and worker first. See [clarification and resume](docs/integrations.md#clarification-and-resume)
+for MCP, limits, and worker instructions.
+
 Attach up to two explicit files to a coding task. Each file may be up to 256 KiB.
 The target receives them as temporary files and can return up to two files by
 writing them to the output directory named in its task prompt. List task files
@@ -253,11 +279,23 @@ relay and target worker before sending attachments; older workers do not
 advertise attachment support. Named Grok Bots do not support file attachments.
 If a long result and output files exceed the 1 MiB transfer limit, the task
 completes with its text result and a notice that the files were omitted.
+Provider text results must fit 128 KiB. Codex and OpenCode JSON events are capped
+at 1 MiB each. If an event is discarded at that limit, the provider must return
+a later text answer or the task fails explicitly.
 
 `--timeout` accepts 5 to 86400 seconds and defaults to 900. It limits how long
 the command waits; the task continues after that. Invalid timeout values are
 rejected before any work is sent, including with `--no-wait`.
 
+To stop coding work at a fixed time, pass `--deadline <UTC-ISO-timestamp>` to
+`delegate`, or `deadline` to the MCP tool. The future deadline is saved on the
+task and stays unchanged through retries and clarification. Expired tasks fail
+with `Task deadline exceeded`. Tasks awaiting their first dispatch fail on the
+relay. Tasks that may have reached a worker stay `cancelling` until it confirms
+termination, including after a disconnect. Upgrade the requester, relay, and
+worker before using deadlines. Named Grok Bots do not support them.
+See [task deadlines](docs/integrations.md#task-deadlines) for details.
+
 Reuse the returned `thread_id` for follow-ups so the target resumes its provider
 session. Queued follow-ups pick up the preceding turn's session when they start,
 including after a worker restart.
@@ -318,7 +356,8 @@ A Cohall thread records the exchange and lets follow-ups find the same Bot.
 It does not create an isolated Grok Bot conversation: messages sent in the
 Grok Bot app share that Bot's history. Bot tasks use the Bot's own permissions
 and computer context, so omit `--workspace`. Queued Bot tasks that have never
-been dispatched can be cancelled; active Bot turns must be stopped in Grok Bot because
+been dispatched and tasks paused for clarification can be cancelled. Active Bot
+turns must be stopped in Grok Bot because
 the gateway cannot safely cancel a specific Cohall turn.
 A dispatched Bot request remains non-cancellable through Cohall if a disconnect
 or relay restart puts it back in the queue, even if its acceptance message was lost.
@@ -326,13 +365,16 @@ When upgrading an older relay without dispatch records, outstanding Bot requests
 are conservatively treated as potentially dispatched.
 
 Cohall includes a local callback command in each Bot request. After finishing,
-the Bot hands its result back by running `cohall reply <task-id> --message-file
-<path>` on its computer. The callback also accepts `--message -` for stdin,
-`--message <text>`, or `--error <text>` when the Bot cannot complete the task.
+the Bot hands its result back by running the supplied command,
+`cohall reply <task-id> --run-id <run-id> --message-file <path>`, on its computer.
+Preserve the supplied task and run IDs. The callback also accepts `--message -`
+for stdin, `--message <text>`, or `--error <text>` when the Bot cannot complete
+the task. Use `--question <text>` to ask for essential missing information, then
+end the Bot turn; the sender's answer resumes the same task with a new callback.
 It records the result locally without relay credentials or transcript scraping.
 The task becomes completed when the worker receives this callback; a reply in
 the Grok Bot chat alone does not complete it. If no callback arrives within six
-hours of the first dispatch, Cohall reports failure. That deadline does not
+hours of that turn's first dispatch, Cohall reports failure. That deadline does not
 stop the Bot's ongoing work.
 
 Pending or uncertain prompt acceptance is checked until the gateway confirms
@@ -389,7 +431,9 @@ Every device must already run Cohall 0.5.0 or newer; upgrade older installations
 individually once before using all-device upgrades.
 Each target must use a global npm, Bun, or pnpm installation that can upgrade
 itself. Preview the targets with `cohall upgrade --all --dry-run`, or pin an
-exact release with `--to 1.2.3`. `cohall upgrades` returns the 50 newest results.
+exact release with `--to 1.2.3`. Updated daemons leave newer installations untouched
+when `latest` is older; an exact `--to` version can intentionally roll back.
+`cohall upgrades` returns the 50 newest results.
 Abandonment is an owner recovery action for a permanently unreachable target;
 it does not interrupt an upgrade that is already executing. Forgetting an
 offline device also closes its outstanding maintenance operation.
@@ -422,18 +466,22 @@ configuration.
 - The relay must be reachable to submit new work, but persisted tasks survive a
   relay restart.
 - The relay retains the newest 1,000 terminal tasks by default so history cannot
-  grow without bound.
+  grow without bound. It acknowledges replayed task and upgrade replies after
+  pruning so workers can discard local replies; this does not restore deleted
+  history.
 - Paired clients can ask a device's local provider to act with that user's normal
   authority. Pair only devices and users you trust.
 - Workspace roots are enforced after resolving symlinks, credentials are
   role-separated, and task traces omit prompts, results, tokens, and provider
-  session IDs.
+  session IDs. Traces include worker progress and clarification text; inspect
+  those fields before sharing them.
 
 ## Documentation
 
 - [Installation, pairing, providers, and upgrades](docs/install.md)
 - [Grok Bot computer setup, Tailscale, and recovery](docs/grok-bot.md)
 - [Agent skill and MCP integrations](docs/integrations.md)
+- [Muse requester setup and proxy access](docs/muse.md)
 - [Linux, macOS, and Windows services](docs/services.md)
 - [Contributing](CONTRIBUTING.md)