feat(infinity-slack-bot): dedicated Slack threads per subagent, stream fixes, native stop button by akainth015 · Pull Request #141 · hydro-project/infinity

akainth015 · 2026-10-16T00:25:27Z

Each subagent's output now streams into its own dedicated Slack thread (anchor message + permalink cross-link; replies/buttons inside a subagent thread route back to that subagent). Also fixes stream discontinuity around interposed blocks, frozen 'in progress' task cards, embedded-serialization bugs, and adds support for Slack's native stop button (agent_session_stopped).

Stacked on #140; part 2 of 6 (followed by #142 #143 #144 #145). See commit message for full details.

cloudflare-workers-and-pages · 2026-10-16T00:26:32Z

Deploying infinity with Cloudflare Pages

Latest commit:	`790a43c`
Status:	✅ Deploy successful!
Preview URL:	https://e8d2f8a6.infinity-dc7.pages.dev
Branch Preview URL:	https://pr-slack-subagent-threads.infinity-dc7.pages.dev

View logs

shadaj · 2026-10-16T17:16:38Z

Can we move the stop button logic into #143? Since it only works properly post #142.

akainth015 · 2026-10-16T17:43:13Z

Done — the stop-button logic now ships in #143 alongside the daemon-interrupt wiring (where it actually works end-to-end, per your point about #142). This PR is now just the subagent threads + stream-correctness fixes (~380 lines lighter). #142's own diff is unchanged (interdiff-verified), and the trees from #143 up are byte-identical to before, so #144/#145/#150 diffs didn't churn.

…nstead of prefixing user text When user text interrupted an in-flight completion, the driver rewrote the text to `<interrupt>…` before folding it into history, so the token the model sees was baked into the stored user message. Likewise, an interrupted tool call was settled by storing a synthetic "interrupted" tool result. Interrupts are now first-class history entries: * `TaggedSyntheticKind::Interrupt` is an input marker; `prepare_input` folds it into `InfinityMessage::Interrupt` (settling a trailing unanswered tool call like any other non-result entry). The driver enqueues a marker ahead of the interrupting text instead of mutating it. * The model-facing `<interrupt>` token is derived at render time by `message::render_history` (used by `HistoryManager::get_history`): it is prefixed onto the following user text (byte-for-byte what the model saw before) or rendered standalone if no user text follows. Consecutive interrupts collapse into one token. * An unanswered tool call is settled by an `Interrupt` entry instead of a stored synthetic "Tool call interrupted by user" result. `render_history` renders an interrupt right after a tool call as that result, so the model sees the same request (when text interrupts an in-flight synchronous tool, the result alone now conveys it; no `<interrupt>` prefix follows). History thus ends on an interrupt when the thread was stopped, and the late real result is still dropped as stale. * Stored user text is no longer rewritten. Rows written by older versions (literal `<interrupt>` prefix, synthetic interrupted results) load and render unchanged; replay keeps stripping the legacy prefix and hides interrupt entries (as the live stream does). Note: an older daemon cannot parse histories containing the new `{"kind":"interrupt"}` entry (forward-incompatible on rollback).

…ation Clients can stop a thread's in-flight work without queuing input and without shutting the session down. Builds on interrupts being recorded as history entries (`InfinityMessage::Interrupt`, previous commit). * New `ClientMessage::Interrupt { thread_id }`: interrupts the thread and its descendant subtree. Old daemons skip unparsable frames, so sending it to one is a no-op. * User text and interrupt markers are a batch's *decisive* inputs. One backwards scan (`last_decisive_input`) decides whether the batch interrupts the in-flight completion (any decisive input) and what happens next (the last one): | batch | in-flight work | new completion? | |-------------------------|----------------|-----------------------| | `[interrupt]` | cancelled | no | | `[text, interrupt]` | cancelled | no | | `[interrupt, text]` | cancelled | yes, answers the text | | `[text]` | cancelled | yes (implicit marker) | * A batch ending on an interrupt is recorded by `Thread::record_interrupt()` without a completion: inputs are folded in order, a trailing unanswered tool call is settled as interrupted (RAP cancellation notified, pending choice dismissed; this settlement is factored out of `step_no_defer`), and history is persisted. Exactly one `CompletionFinished` (→ `ResponseDone`) is emitted per interrupted round. * `HistoryManager::promote_harness_settlements()` persists the leading interrupt entries that no completion will validate. History then ends on the interrupt, so the driver idles out and a late tool result is dropped instead of restarting generation. User text stays unvalidated and unpersisted, like any input cancelled before model output. * `SessionManager::interrupt_thread()` delivers interrupt markers to the target and its active descendants; idle descendants are skipped. * `ClientMessage::Interrupt` is handled in the client handler, with thread refs stripped on the remote path like `UserInput`.

…ve rewind) Adds `ClientMessage::ForkSession { thread_id, user_turn, keeps_session_alive }`: create a NEW session whose history is the original's conversation just before its `user_turn`-th user-text turn. The original session is only read, so "rewinding" is an edit-and-fork that keeps it intact. * `SessionManager::fork_session` resolves the turn (0-based over the session's persisted user-text rows, matching the `UserInputEcho` entries in its `Replay`; rows behind a compaction summary still count) and seeds a new root thread with existing `ConversationStore` primitives: the rows before the cut are copied verbatim and the latest compaction summary at or before the cut is carried over. The fork inherits cwd and model and is registered last, so a half-seeded fork never lists. * Like `CreateSession`, the requesting connection is attached to the fork and receives `Connected`; clients then send the (edited) turn as `UserInput`. Non-sessions and out-of-range turns are reported via `DaemonMessage::Error`. `ThreadRef`s are stripped on the remote-forwarding path.

…ream-correctness fixes ## Subagent threads * Subagent output (daemon events whose `thread_id` differs from the session root) routes into a dedicated Slack thread, created lazily via an anchor message ("🤖 Subagent …") with a permalink cross-link in the root thread. * Replies/button clicks inside a subagent thread route back to that subagent (`SendInput.target_thread`); per-(thread, subagent) stream state. ## Stream correctness * Streams are finalized before any discrete block (buttons, errors, cross-links) is posted into their thread, fixing out-of-order edits above interposed blocks. * Task cards are tracked per stream message (`open_tasks`) and completed on every stop/split/recovery path so spinners can never freeze; late completions are dropped instead of resurrecting streams. * `SlackAction` serialization fixes for the embedded binary format (no `skip_serializing_if`, string-adapted `serde_json::Value` fields). Docs updated; new sim/unit tests cover routing, finalization, and task cards. (Native stop-button support lands with the interrupt wiring in a later commit, where it works end to end.)

…ream-correctness fixes ## Subagent threads * Subagent output (daemon events whose `thread_id` differs from the session root) routes into a dedicated Slack thread, created lazily via an anchor message ("🤖 Subagent …") with a permalink cross-link in the root thread. * Replies/button clicks inside a subagent thread route back to that subagent (`SendInput.target_thread`); per-(thread, subagent) stream state. ## Stream correctness * Streams are finalized before any discrete block (buttons, errors, cross-links) is posted into their thread, fixing out-of-order edits above interposed blocks. * Task cards are tracked per stream message (`open_tasks`) and completed on every stop/split/recovery path so spinners can never freeze; late completions are dropped instead of resurrecting streams. * `SlackAction` serialization fixes for the embedded binary format (no `skip_serializing_if`, string-adapted `serde_json::Value` fields). Docs updated; new sim/unit tests cover routing, finalization, and task cards. Ported from hydro-project/infinity#141.

akainth015 · 2026-11-04T23:40:34Z

Moved to hydro-project/infinity-slack@2b5d03bc0ed2. The native stop button / interrupt parts were dropped.

Base automatically changed from pr/clippy-199 to main October 16, 2026 00:36

akainth015 requested a review from a team October 16, 2026 00:36

akainth015 mentioned this pull request Oct 16, 2026

chore: fix clippy 1.99 fallout #140

Merged

akainth015 added the stacked-pr label Oct 16, 2026

akainth015 added this pull request to stack #149 October 16, 2026 17:01

akainth015 force-pushed the pr/slack-subagent-threads branch from f05c004 to b95c6e5 Compare October 16, 2026 17:02

akainth015 removed the stacked-pr label Oct 16, 2026

akainth015 force-pushed the pr/slack-subagent-threads branch 2 times, most recently from 72afe98 to 4973257 Compare October 19, 2026 20:31

akainth015 removed this pull request from stack #149 October 19, 2026 22:08

akainth015 changed the base branch from main to pr/daemon-rewind October 19, 2026 22:08

akainth015 force-pushed the pr/slack-subagent-threads branch from 4973257 to 19a0968 Compare October 19, 2026 23:51

akainth015 force-pushed the pr/daemon-rewind branch 2 times, most recently from 1508d93 to 39a3427 Compare October 20, 2026 00:16

akainth015 force-pushed the pr/slack-subagent-threads branch from 19a0968 to 2a2b13c Compare October 20, 2026 00:17

akainth015 marked this pull request as draft October 20, 2026 02:52

akainth015 force-pushed the pr/daemon-rewind branch from 39a3427 to a8da36b Compare October 20, 2026 17:42

akainth015 force-pushed the pr/slack-subagent-threads branch from 2a2b13c to 2001008 Compare October 20, 2026 17:42

akainth015 added 4 commits October 22, 2026 00:22

akainth015 force-pushed the pr/slack-subagent-threads branch from 2001008 to 790a43c Compare October 22, 2026 01:18

akainth015 force-pushed the pr/daemon-rewind branch from a8da36b to 3c24168 Compare October 22, 2026 01:18

akainth015 force-pushed the pr/daemon-rewind branch from 3c24168 to c426040 Compare October 29, 2026 22:57

akainth015 closed this Nov 4, 2026

akainth015 deleted the pr/slack-subagent-threads branch November 4, 2026 23:43

Navigation Menu

Repository navigation

feat(infinity-slack-bot): dedicated Slack threads per subagent, stream fixes, native stop button - #141

feat(infinity-slack-bot): dedicated Slack threads per subagent, stream fixes, native stop button#141
akainth015 wants to merge 4 commits into
pr/daemon-rewindfrom
pr/slack-subagent-threads

akainth015 commented Oct 16, 2026 •

edited

Loading

Uh oh!

cloudflare-workers-and-pages Bot commented Oct 16, 2026 •

edited

Loading

Uh oh!

shadaj commented Oct 16, 2026

Uh oh!

akainth015 commented Oct 16, 2026

Uh oh!

akainth015 commented Nov 4, 2026

Uh oh!

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Conversation

akainth015 commented Oct 16, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

cloudflare-workers-and-pages Bot commented Oct 16, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Deploying infinity with Cloudflare Pages

Uh oh!

shadaj commented Oct 16, 2026

Uh oh!

akainth015 commented Oct 16, 2026

Uh oh!

akainth015 commented Nov 4, 2026

Uh oh!

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

akainth015 commented Oct 16, 2026 •

edited

Loading

cloudflare-workers-and-pages Bot commented Oct 16, 2026 •

edited

Loading