fix(slack): slash-command turns keep the channel prompt, skill binding and source names (#128720, salvage #128722) by teknium1 · Pull Request #139858 · NousResearch/hermes-agent

teknium1 · 2026-10-16T11:55:46Z

A Slack slash-command turn (/hermes <question>, native /<command>) now carries the same channel prompt, bound skills and source names as an ordinary message in that channel, so it no longer drops slack.channel_prompts or flips the session's prompt pins.

SlackAdapter._handle_slash_command builds its event with channel_prompt=_channel_prompt_with_identity(...), auto_skill=resolve_channel_skills(...), and resolves chat_name / user_name with the message path's helpers (_build_message_event).
A sender the gateway rejects costs no Slack lookup (in a channel it is dropped before any lookup; in a DM it still reaches the runner for the pairing/decline answer, without names).
Commands that start no turn (/stop and other interrupt_then_dispatch, /approve, /deny, /pause) are handed over before any users.info / conversations.info, so a cold lookup cannot hold an emergency stop back.
The /goal <text> kickoff turn (gateway/slash_commands_goals.py::_enqueue_goal_turn) keeps the triggering turn's auto_skill along with its channel prompt.
slash_command_text / slash_thread_id move unchanged into plugins/platforms/slack/adapter_slash_payload.py to keep adapter.py under its health cap.
Docs: website/docs/user-guide/messaging/slack.md says slash turns get the channel prompt and binding.

Live repro (real `SlackAdapter._handle_slack_message` / `_handle_slash_command` → real `GatewayRunner._pinned_channel_inputs` + `_pinned_session_context_prompt`, temp `HERMES_HOME`; only the Slack Web API client is stubbed)

message → /hermes what broke? → message in one thread session, channel C0123OPS with channel_prompts: {C0123OPS: "Answer in haiku."}:

	slash `channel_prompt`	slash `chat_name` / `user_name`	distinct pinned (context, channel) prompts
`origin/main` `1274ab7`, thread	`None`	`None` / `None`	2
`origin/main`, DM thread	`None`	`None` / `None`	2
this head, thread	haiku + identity line	`ops` / `Alice`	1
this head, DM thread	haiku + identity line	`Alice` / `Alice`	1

Root cause: _handle_slash_command was the only MessageEvent( site in the Slack adapter that skipped the channel-prompt / name / skill-binding helpers; as a non-internal human turn it re-pinned channel_pin=(None, …) and re-keyed _ephemeral_change_key, and the next message flipped both back.

Tests

tests/plugins/platforms/slack/test_slash_command_channel_prompt.py (9): message → slash → message renders one pinned prompt pair through the real gateway pins (channel, DM, /queue, /goal kickoff via the real _enqueue_goal_turn); rejected/control commands cost no lookup. Red on origin/main (5 failed); deleting only the auto_skill= line in _enqueue_goal_turn turns goal-kickoff red.
tests/gateway/test_slack.py + the new file: 221 passed, 1 skipped.
scripts/check --only health: ok.

Fixes #128720
Salvages #128722 (@jonpol01), cherry-picked with authorship; includes @JoaoMarcos44's /goal kickoff commit from that branch.

Dropped hunks

fix(slack): carry channel prompt and source names on slash-command turns #128722 9b719083 / eb546b6a (per-session slash handoff lane _slash_handoff_tails + cancellation chaining). Resolving names now awaits users.info / conversations.info before the handoff, the same as the message path already does. So two slashes of one session fired together, with the first on a cold lookup, can reach the gateway's /queue FIFO swapped. Once the per-(team, user) and per-(team, channel) caches are warm there is no window. That lane is about queue ordering, not the prompt pins this issue covers, and it adds a future-chaining lane to the adapter, so it is left out for a maintainer call. Only its /pause atom is ported (co-authored). Not re-probed here.
fix(slack): unify slash and message prompt inputs #129046 (@JoaoMarcos44, +930) is a broader rework of the same site. Superseded by this smaller salvage.

Infographic

Review follow-up

Slash FIFO regression (second review F7), fixed in 811e4b92545. Resolving names before the handoff let two /queue slashes of one session reach the gateway swapped when the first had a cold users.info. This head now ports fix(slack): carry channel prompt and source names on slash-command turns #128722 9b719083 / eb546b6a (co-authored with @jonpol01) into plugins/platforms/slack/adapter_slash_lane.py. A slash takes its session's slot before any lookup and hands over only after the slash ahead of it has returned from handle_message. A slash cancelled while waiting still keeps the one behind it waiting. /stop, /approve, /deny and /pause take no slot, so they still go through at once. The first "Dropped hunks" bullet above no longer applies.
Test: test_concurrent_slashes_reach_the_gateway_in_arrival_order goes through the Bolt listener (_handle_hermes_command). It holds the first /hermes queue on a cold lookup and sends a second, plus a /stop in between. /stop is handed over at once, then /queue first and /queue second arrive in order. Red on c271b06ee8d (['/queue second', '/stop']), green now. Slack tests: 249 passed, 1 skipped. scripts/check: ok.

Review follow-up

dc5dee3: _early_reject_unauthorized now takes the decoded thread_id and passes it to _is_sender_authorized and to the build_source fallback. Both call sites pass it: the slash path, and the message path (where the omission predates this PR; it now runs after _session_thread_ts). Before this, a thread-scoped profile_routes entry was checked against the default profile's pairing store, so a caller approved only in the routed profile had /stop and messages in that thread dropped (Greptile + maintainer review). New test test_thread_routed_sender_is_checked_against_the_routed_thread, through _handle_slash_command and _handle_slack_message: fails before (nothing reaches the runner), passes after.

_handle_slash_command built its MessageEvent without channel_prompt and its source without chat_name/user_name, unlike the message path. A /hermes turn ran without the configured channel_prompts and the bot identity line, and as a human turn it re-pinned channel_pin to None and re-keyed the pinned session-context prompt; the next ordinary message flipped both back, so every slash turn cost two agent rebuilds and two prompt-cache misses. Synthetic events copied from it (/retry, /queue, /steer fallback, /goal kickoff) inherited the gap. Resolve the names and the channel prompt with the same helpers the message path uses.

`/hermes <question>` can open a new session, and bound skills load only when a session starts. The slash event carried no auto_skill, so a session opened that way never loaded the channel's slack.channel_skill_bindings, unlike one opened by an ordinary message. Resolve it the same way the message path does.

The message path rejects an unauthorized sender before any lookup (5bb933e); the slash path now resolves the channel and user name, so a sender the gateway was going to reject cost a conversations.info and a users.info first. Ask the same early gate before resolving. In a channel the runner only drops such a sender, so stop there. In a DM it answers per unauthorized_dm_behavior (pairing code or decline), so the event still reaches it, without names, exactly as before this branch. Docs: N/A, slack.md already says an unauthorized DM gets the pairing prompt.

…aits for no Slack lookup The names the slash path resolves (users.info, conversations.info) were awaited before the event reached the active-session dispatcher. With a shared channel session, a second operator's /stop sat behind their uncached users.info while the first operator's worker kept running. /stop, /new, /reset (the interrupt-then-dispatch commands) and /approve, /deny start no turn of their own, so they are handed over before any lookup. Commands that do start a turn (/queue, /retry, /goal, ...) and free-form questions keep the names and the channel prompt.

_enqueue_goal_turn rebuilt the kickoff turn with the message id and channel prompt but not auto_skill, so the turn the agent actually runs for `/goal <text>` carried no channel skill binding. A resume continuation still carries none. Ported from #129046 (5ee2a10) onto #128722 as a credited delta, per the maintainer review of #129046. The code line is unchanged; the docstring line was adjusted to name bound skills when porting. Refs #130316

/pause engages the emergency stop and starts no turn, like /stop and /approve, so a cold users.info must not hold it back. Ported from 9b71908 in #128722 (its arrival-order handoff lane is left out). Co-authored-by: John Paul Soliva <soliva.johnpaul@icloud.com>

… prompt Drive message -> slash -> message in one thread session through the real gateway pins (_pinned_channel_inputs / _pinned_session_context_prompt) and the real /goal kickoff enqueue, so the class is judged by the cache outcome, not only by event fields. Co-authored-by: John Paul Soliva <soliva.johnpaul@icloud.com>

Move the two pure slash-payload helpers (gateway text, thread anchor) out of the adapter facade so the slash-turn parity fix keeps adapter.py under its health cap. No behaviour change.

…ding

github-actions · 2026-10-16T11:56:14Z

૮ >ﻌ< ა ci review

_{ran on dc5dee3 — fix(slack): early auth reject checks the session thread's ro}

debug info

CI timings

CI timings · View report · View job

Wall time 9m58s vs 15m21s (-35.1%). 10 job(s) slower, 8 faster, 1 unchanged.

Python tests / Run tests (1/2): +98.0s
OS-specific tests / Windows-only tests: +98.0s
OS-specific tests / Windows-only tests (arm64, 1/2): -91.0s
OS-specific tests / Windows-only tests (arm64, 2/2): +67.0s
Detect affected areas: +12.0s

greptile-apps · 2026-10-16T12:01:09Z

Confidence Score: 4/5

[Medium impact] Slash commands now preserve channel prompt and skill binding.

Fix the thread-aware access check before merging so valid routed slash commands are not dropped.

Findings

Routed commands get dropped ▶

Fix with agent prompt

### Issue 1
plugins/platforms/slack/adapter.py:6081-6083
The new early access check omits the decoded `thread_id`. If a shared Slack bot has a thread-specific `profile_routes` entry and the caller is approved only in the destination profile's pairing store, this checks the default profile instead. In a channel, the command is then dropped before its correctly routed source is built. Even `/stop` never reaches the runner.

Pass `thread_id` through `_early_reject_unauthorized` to both authorization paths. Add a regression test with a thread route and separate pairing stores.

---

For each issue above, determine whether it is valid and should be fixed. If so, fix it directly.

Summary

Slack slash turns now carry the channel prompt, bound skills, and source names used by ordinary messages.

Slack slash turns keep the channel prompt and skill binding.
Slack looks up names only for slash commands that need them.

Review was based on code inspection; tests were not run.

Diagram

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A[Slack slash payload] --> B[Decode text and thread]
    B --> C[Channel and DM gates]
    C --> D[Early access check without thread]
    D -->|Channel rejected| E[Drop command]
    D -->|Continue| F[Build source with thread]
    F --> G[Attach prompt and skills]
    G --> H{Control command or rejected DM?}
    H -->|No| I[Resolve source names]
    H -->|Yes| J[Hand event to gateway]
    I --> J

_{Reviews (1) · Last reviewed commit: "docs(slack): slash-command turns get the..." · Reviewed by Greptile}

greptile-apps · 2026-10-16T12:01:13Z

+        rejected = self._early_reject_unauthorized(user_id, channel_id, is_dm)
+        if rejected and not is_dm:
+            return


Routed commands get dropped

The new early access check omits the decoded thread_id. If a shared Slack bot has a thread-specific profile_routes entry and the caller is approved only in the destination profile's pairing store, this checks the default profile instead. In a channel, the command is then dropped before its correctly routed source is built. Even /stop never reaches the runner.

Pass thread_id through _early_reject_unauthorized to both authorization paths. Add a regression test with a thread route and separate pairing stores.

Prompt To Fix With AI

This is a comment left during a code review. Path: plugins/platforms/slack/adapter.py Line: 6081-6083 Comment: **Routed commands get dropped** The new early access check omits the decoded `thread_id`. If a shared Slack bot has a thread-specific `profile_routes` entry and the caller is approved only in the destination profile's pairing store, this checks the default profile instead. In a channel, the command is then dropped before its correctly routed source is built. Even `/stop` never reaches the runner. Pass `thread_id` through `_early_reject_unauthorized` to both authorization paths. Add a regression test with a thread route and separate pairing stores. --- For each issue above, determine whether it is valid and should be fixed. If so, fix it directly.

Fixed in dc5dee3: _early_reject_unauthorized takes thread_id and passes it to _is_sender_authorized and the build_source fallback, from both the slash path and the (pre-existing) message path. test_thread_routed_sender_is_checked_against_the_routed_thread drives both entry points with a thread-scoped auth check (red before).

teknium1

P0 rapid-response review — verdict: sound, fold in Greptile's thread fix before the merge-ask.

Slash turns now carry channel_prompt (via _channel_prompt_with_identity, same as the message path), auto_skill, chat_name, user_name, so the pins match the next message. Skipping the Slack lookups for control commands (/stop, /approve, /deny, /pause) is the right call: a cold users.info must not hold an emergency stop.

Real finding (Greptile, confirmed against main): the new early reject at _handle_slash_command calls _early_reject_unauthorized(user_id, channel_id, is_dm) without the decoded thread_id. BasePlatformAdapter._is_sender_authorized and _make_adapter_auth_check both take thread_id precisely so a thread-scoped profile_routes entry checks the routed profile's pairing store. Without it a caller approved only in the routed profile is rejected in a channel and the command is dropped before the runner sees it — /stop included. The message path has the same omission (pre-existing, _handle_slack_message), so fix both: add a thread_id kwarg to _early_reject_unauthorized, pass it through to _is_sender_authorized and the build_source fallback, and add one invariant test with a thread route.

Nothing pushed; review only.

teknium1 · 2026-10-16T12:28:35Z

+        # A sender the gateway rejects must not cost Slack lookups first (the message path's rule).
+        # In a channel the runner only drops them; in a DM it answers per unauthorized_dm_behavior
+        # (pairing code or decline), so it still gets the event, without names.
+        rejected = self._early_reject_unauthorized(user_id, channel_id, is_dm)


Pass thread_id here (and thread it into _is_sender_authorized(..., thread_id=...) / the fallback build_source). Without it a thread-routed profile's paired user is rejected and even /stop is dropped in a channel. The message path's call has the same gap; fix both.

Done in dc5dee3 (both call sites, regression through _handle_slash_command and _handle_slack_message).

…y in arrival order Names resolved for a slash turn are awaited before the handoff and Bolt runs callbacks concurrently, so a /queue whose cold users.info answered late reached the gateway FIFO after a later one. A slash now reserves its session's slot before any lookup and hands over only after the slash ahead returned from handle_message; a slash cancelled while waiting still chains the one behind it. /stop, /approve, /deny, /pause take no slot. Ported from #128722 (9b71908, eb546b6) into adapter_slash_lane.py. Co-authored-by: John Paul Soliva <soliva.johnpaul@icloud.com>

_early_reject_unauthorized dropped the decoded thread_id, so a thread-scoped profile_routes entry checked the default profile's pairing store: a caller approved only in the routed profile had slash commands (even /stop) and messages in that thread dropped. Pass thread_id from both the slash and the message path to _is_sender_authorized and the build_source fallback.

jonpol01 and others added 9 commits October 16, 2026 04:39

refactor(slack): slash payload decoding lives in adapter_slash_payload

575e77d

Move the two pure slash-payload helpers (gateway text, thread anchor) out of the adapter facade so the slash-turn parity fix keeps adapter.py under its health cap. No behaviour change.

docs(slack): slash-command turns get the channel prompt and skill bin…

c271b06

…ding

greptile-apps Bot reviewed Oct 16, 2026

View reviewed changes

teknium1 commented Oct 16, 2026

View reviewed changes

teknium1 and others added 2 commits October 16, 2026 07:10

alt-glitch added P3 Low — cosmetic, nice to have P0 Critical — data loss, security, crash loop and removed P0 Critical — data loss, security, crash loop P3 Low — cosmetic, nice to have labels Oct 16, 2026

teknium1 added the ci-reviewed applied to manually approve dangerous changes label Oct 16, 2026

teknium1 merged commit f41ac07 into main Oct 16, 2026
56 checks passed

teknium1 deleted the fix/bp0-pins-adapters branch October 16, 2026 21:22

This was referenced Oct 16, 2026

fix(slack): carry channel prompt and source names on slash-command turns #128722

Closed

[Bug]: bound-skill auto-load runs skill_view / dependency setup synchronously on the gateway event loop #140256

Open

andrexibiza mentioned this pull request Oct 17, 2026

fix(gateway): keep queued context and channel prompt pins coherent #140286

Draft

7 of 8 tasks

Navigation Menu

Repository navigation

fix(slack): slash-command turns keep the channel prompt, skill binding and source names (#128720, salvage #128722) - #139858

fix(slack): slash-command turns keep the channel prompt, skill binding and source names (#128720, salvage #128722)#139858
teknium1 merged 11 commits into
mainfrom
fix/bp0-pins-adapters

teknium1 commented Oct 16, 2026 •

edited

Loading

Uh oh!

github-actions Bot commented Oct 16, 2026 •

edited

Loading

CI timings · View report · View job

Uh oh!

greptile-apps Bot commented Oct 16, 2026 •

edited

Loading

Summary

Diagram

Uh oh!

greptile-apps Bot Oct 16, 2026

Uh oh!

teknium1 Oct 16, 2026

Uh oh!

teknium1 left a comment

Uh oh!

teknium1 Oct 16, 2026

Uh oh!

teknium1 Oct 16, 2026

Uh oh!

Uh oh!

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

Conversation

teknium1 commented Oct 16, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Live repro (real SlackAdapter._handle_slack_message / _handle_slash_command → real GatewayRunner._pinned_channel_inputs + _pinned_session_context_prompt, temp HERMES_HOME; only the Slack Web API client is stubbed)

Tests

Dropped hunks

Infographic

Review follow-up

Review follow-up

Uh oh!

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

Uh oh!

૮ >ﻌ< ა ci review

debug info

CI timings · View report · View job

Uh oh!

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

Uh oh!

Confidence Score: 4/5

Findings

Summary

Diagram

Uh oh!

greptile-apps Bot Oct 16, 2026

Choose a reason for hiding this comment

Uh oh!

teknium1 Oct 16, 2026

Choose a reason for hiding this comment

Uh oh!

teknium1 left a comment

Choose a reason for hiding this comment

Uh oh!

teknium1 Oct 16, 2026

Choose a reason for hiding this comment

Uh oh!

teknium1 Oct 16, 2026

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

teknium1 commented Oct 16, 2026 •

edited

Loading

Live repro (real `SlackAdapter._handle_slack_message` / `_handle_slash_command` → real `GatewayRunner._pinned_channel_inputs` + `_pinned_session_context_prompt`, temp `HERMES_HOME`; only the Slack Web API client is stubbed)

github-actions Bot commented Oct 16, 2026 •

edited

Loading

greptile-apps Bot commented Oct 16, 2026 •

edited

Loading