Skip to content

fix: handle /reasoning while gateway session is busy#4665

Open
j30231 wants to merge 1 commit into
NousResearch:mainfrom
j30231:fix/gateway-busy-reasoning-command
Open

fix: handle /reasoning while gateway session is busy#4665
j30231 wants to merge 1 commit into
NousResearch:mainfrom
j30231:fix/gateway-busy-reasoning-command

Conversation

@j30231

@j30231 j30231 commented Apr 2, 2026

Copy link
Copy Markdown

Summary

  • dispatch /reasoning immediately from the gateway busy-session fast path
  • prevent busy /reasoning from being queued as plain user text or routed through interrupt handling
  • add regression tests for both the sentinel startup window and an already-running agent

Root cause

Gateway command handling currently has two phases inside _handle_message():

  1. an early busy-session fast path
  2. the normal slash-command dispatcher later in the method

When a session was already busy, /reasoning never reached the normal dispatcher. During the sentinel startup window it was queued into adapter._pending_messages, and those queued entries are later replayed directly into _run_agent(...) as plain user text. During a real in-flight turn it went through the interrupt path instead of the command handler.

That made /reasoning behave unlike /status, even though both are session meta commands that should respond immediately while work is in flight.

Changes

  • treat canonical reasoning as a busy-safe command in the early gateway fast path
  • document why this bypass exists
  • add race-guard regression tests asserting that /reasoning:
    • dispatches immediately while the session is still in the sentinel startup window
    • dispatches immediately while a real agent is running
    • preserves the busy lock instead of queueing or interrupting the active session

Test plan

  • uv run --extra dev python -m pytest -q tests/gateway/test_session_race_guard.py tests/gateway/test_reasoning_command.py -n 0

Notes

  • This intentionally keeps scope narrow to /reasoning.
  • The deeper issue is duplicated busy-session command policy; a safe follow-up would be to centralize the busy-safe command allowlist and evaluate other meta commands separately.

@j30231

j30231 commented Apr 2, 2026

Copy link
Copy Markdown
Author

Heads-up: the GitHub Actions runs for this fork PR landed in action_required before any jobs started (Tests / Docker Build and Publish / Supply Chain Audit). I verified the targeted gateway tests locally with:\n\nuv run --extra dev python -m pytest -q tests/gateway/test_session_race_guard.py tests/gateway/test_reasoning_command.py -n 0\n\nResult: 19 passed.\n\nCould a maintainer approve the workflow runs for this PR so CI can execute on the upstream repo?

teknium1 added a commit that referenced this pull request Apr 19, 2026
@teknium1
fix(gateway): slash commands never interrupt a running agent
9596e1d 
Any recognized slash command now bypasses the Level-1 active-session
guard instead of queueing + interrupting. A mid-run /model (or
/reasoning, /voice, /insights, /title, /resume, /retry, /undo,
/compress, /usage, /provider, /reload-mcp, /sethome, /reset) used to
interrupt the agent AND get silently discarded by the slash-command
safety net — zero-char response, dropped tool calls.

Root cause:
- Discord registers 41 native slash commands via tree.command().
- Only 14 were in ACTIVE_SESSION_BYPASS_COMMANDS.
- The other ~15 user-facing ones fell through base.py:handle_message
  to the busy-session handler, which calls running_agent.interrupt()
  AND queues the text.
- After the aborted run, gateway/run.py:9912 correctly identifies the
  queued text as a slash command and discards it — but the damage
  (interrupt + zero-char response) already happened.

Fix:
- should_bypass_active_session() now returns True for any resolvable
  slash command. ACTIVE_SESSION_BYPASS_COMMANDS stays as the subset
  with dedicated Level-2 handlers (documentation + tests).
- gateway/run.py adds a catch-all after the dedicated handlers that
  returns a user-visible "agent busy — wait or /stop first" response
  for any other resolvable command.
- Unknown text / file-path-like messages are unchanged — they still
  queue.

Also:
- gateway/platforms/discord.py logs the invoker identity on every
  slash command (user id + name + channel + guild) so future
  ghost-command reports can be triaged without guessing.

Tests:
- 15 new parametrized cases in test_command_bypass_active_session.py
  cover every previously-broken Discord slash command.
- Existing tests for /stop, /new, /approve, /deny, /help, /status,
  /agents, /background, /steer, /update, /queue still pass.
- test_steer.py's ACTIVE_SESSION_BYPASS_COMMANDS check still passes.

Fixes #5057. Related: #6252, #10370, #4665.
@teknium1 teknium1 mentioned this pull request Apr 19, 2026
Merged
teknium1 added a commit that referenced this pull request Apr 19, 2026
@teknium1
fix(gateway): slash commands never interrupt a running agent (#12334)
632a807 
Any recognized slash command now bypasses the Level-1 active-session
guard instead of queueing + interrupting. A mid-run /model (or
/reasoning, /voice, /insights, /title, /resume, /retry, /undo,
/compress, /usage, /provider, /reload-mcp, /sethome, /reset) used to
interrupt the agent AND get silently discarded by the slash-command
safety net — zero-char response, dropped tool calls.

Root cause:
- Discord registers 41 native slash commands via tree.command().
- Only 14 were in ACTIVE_SESSION_BYPASS_COMMANDS.
- The other ~15 user-facing ones fell through base.py:handle_message
  to the busy-session handler, which calls running_agent.interrupt()
  AND queues the text.
- After the aborted run, gateway/run.py:9912 correctly identifies the
  queued text as a slash command and discards it — but the damage
  (interrupt + zero-char response) already happened.

Fix:
- should_bypass_active_session() now returns True for any resolvable
  slash command. ACTIVE_SESSION_BYPASS_COMMANDS stays as the subset
  with dedicated Level-2 handlers (documentation + tests).
- gateway/run.py adds a catch-all after the dedicated handlers that
  returns a user-visible "agent busy — wait or /stop first" response
  for any other resolvable command.
- Unknown text / file-path-like messages are unchanged — they still
  queue.

Also:
- gateway/platforms/discord.py logs the invoker identity on every
  slash command (user id + name + channel + guild) so future
  ghost-command reports can be triaged without guessing.

Tests:
- 15 new parametrized cases in test_command_bypass_active_session.py
  cover every previously-broken Discord slash command.
- Existing tests for /stop, /new, /approve, /deny, /help, /status,
  /agents, /background, /steer, /update, /queue still pass.
- test_steer.py's ACTIVE_SESSION_BYPASS_COMMANDS check still passes.

Fixes #5057. Related: #6252, #10370, #4665.
ulasbilgen pushed a commit to ulasbilgen/hermes-adhd-agent that referenced this pull request May 1, 2026
@teknium1
fix(gateway): slash commands never interrupt a running agent (NousRes…
17321a6 
…earch#12334)

Any recognized slash command now bypasses the Level-1 active-session
guard instead of queueing + interrupting. A mid-run /model (or
/reasoning, /voice, /insights, /title, /resume, /retry, /undo,
/compress, /usage, /provider, /reload-mcp, /sethome, /reset) used to
interrupt the agent AND get silently discarded by the slash-command
safety net — zero-char response, dropped tool calls.

Root cause:
- Discord registers 41 native slash commands via tree.command().
- Only 14 were in ACTIVE_SESSION_BYPASS_COMMANDS.
- The other ~15 user-facing ones fell through base.py:handle_message
  to the busy-session handler, which calls running_agent.interrupt()
  AND queues the text.
- After the aborted run, gateway/run.py:9912 correctly identifies the
  queued text as a slash command and discards it — but the damage
  (interrupt + zero-char response) already happened.

Fix:
- should_bypass_active_session() now returns True for any resolvable
  slash command. ACTIVE_SESSION_BYPASS_COMMANDS stays as the subset
  with dedicated Level-2 handlers (documentation + tests).
- gateway/run.py adds a catch-all after the dedicated handlers that
  returns a user-visible "agent busy — wait or /stop first" response
  for any other resolvable command.
- Unknown text / file-path-like messages are unchanged — they still
  queue.

Also:
- gateway/platforms/discord.py logs the invoker identity on every
  slash command (user id + name + channel + guild) so future
  ghost-command reports can be triaged without guessing.

Tests:
- 15 new parametrized cases in test_command_bypass_active_session.py
  cover every previously-broken Discord slash command.
- Existing tests for /stop, /new, /approve, /deny, /help, /status,
  /agents, /background, /steer, /update, /queue still pass.
- test_steer.py's ACTIVE_SESSION_BYPASS_COMMANDS check still passes.

Fixes NousResearch#5057. Related: NousResearch#6252, NousResearch#10370, NousResearch#4665.
aj-nt pushed a commit to aj-nt/hermes-agent that referenced this pull request May 1, 2026
@teknium1
fix(gateway): slash commands never interrupt a running agent (NousRes…
9ba3cb9 
…earch#12334)

Any recognized slash command now bypasses the Level-1 active-session
guard instead of queueing + interrupting. A mid-run /model (or
/reasoning, /voice, /insights, /title, /resume, /retry, /undo,
/compress, /usage, /provider, /reload-mcp, /sethome, /reset) used to
interrupt the agent AND get silently discarded by the slash-command
safety net — zero-char response, dropped tool calls.

Root cause:
- Discord registers 41 native slash commands via tree.command().
- Only 14 were in ACTIVE_SESSION_BYPASS_COMMANDS.
- The other ~15 user-facing ones fell through base.py:handle_message
  to the busy-session handler, which calls running_agent.interrupt()
  AND queues the text.
- After the aborted run, gateway/run.py:9912 correctly identifies the
  queued text as a slash command and discards it — but the damage
  (interrupt + zero-char response) already happened.

Fix:
- should_bypass_active_session() now returns True for any resolvable
  slash command. ACTIVE_SESSION_BYPASS_COMMANDS stays as the subset
  with dedicated Level-2 handlers (documentation + tests).
- gateway/run.py adds a catch-all after the dedicated handlers that
  returns a user-visible "agent busy — wait or /stop first" response
  for any other resolvable command.
- Unknown text / file-path-like messages are unchanged — they still
  queue.

Also:
- gateway/platforms/discord.py logs the invoker identity on every
  slash command (user id + name + channel + guild) so future
  ghost-command reports can be triaged without guessing.

Tests:
- 15 new parametrized cases in test_command_bypass_active_session.py
  cover every previously-broken Discord slash command.
- Existing tests for /stop, /new, /approve, /deny, /help, /status,
  /agents, /background, /steer, /update, /queue still pass.
- test_steer.py's ACTIVE_SESSION_BYPASS_COMMANDS check still passes.

Fixes NousResearch#5057. Related: NousResearch#6252, NousResearch#10370, NousResearch#4665.
@alt-glitch alt-glitch added type/bug Something isn't working P2 Medium — degraded but workaround exists comp/gateway Gateway runner, session dispatch, delivery labels May 1, 2026
02356abc pushed a commit to 02356abc/hermes-agent that referenced this pull request May 14, 2026
@teknium1
fix(gateway): slash commands never interrupt a running agent (NousRes…
f87b511 
…earch#12334)

Any recognized slash command now bypasses the Level-1 active-session
guard instead of queueing + interrupting. A mid-run /model (or
/reasoning, /voice, /insights, /title, /resume, /retry, /undo,
/compress, /usage, /provider, /reload-mcp, /sethome, /reset) used to
interrupt the agent AND get silently discarded by the slash-command
safety net — zero-char response, dropped tool calls.

Root cause:
- Discord registers 41 native slash commands via tree.command().
- Only 14 were in ACTIVE_SESSION_BYPASS_COMMANDS.
- The other ~15 user-facing ones fell through base.py:handle_message
  to the busy-session handler, which calls running_agent.interrupt()
  AND queues the text.
- After the aborted run, gateway/run.py:9912 correctly identifies the
  queued text as a slash command and discards it — but the damage
  (interrupt + zero-char response) already happened.

Fix:
- should_bypass_active_session() now returns True for any resolvable
  slash command. ACTIVE_SESSION_BYPASS_COMMANDS stays as the subset
  with dedicated Level-2 handlers (documentation + tests).
- gateway/run.py adds a catch-all after the dedicated handlers that
  returns a user-visible "agent busy — wait or /stop first" response
  for any other resolvable command.
- Unknown text / file-path-like messages are unchanged — they still
  queue.

Also:
- gateway/platforms/discord.py logs the invoker identity on every
  slash command (user id + name + channel + guild) so future
  ghost-command reports can be triaged without guessing.

Tests:
- 15 new parametrized cases in test_command_bypass_active_session.py
  cover every previously-broken Discord slash command.
- Existing tests for /stop, /new, /approve, /deny, /help, /status,
  /agents, /background, /steer, /update, /queue still pass.
- test_steer.py's ACTIVE_SESSION_BYPASS_COMMANDS check still passes.

Fixes NousResearch#5057. Related: NousResearch#6252, NousResearch#10370, NousResearch#4665.
gweeteve pushed a commit to gweeteve/hermes-agent that referenced this pull request Jun 2, 2026
@teknium1
fix(gateway): slash commands never interrupt a running agent (NousRes…
86088a6 
…earch#12334)

Any recognized slash command now bypasses the Level-1 active-session
guard instead of queueing + interrupting. A mid-run /model (or
/reasoning, /voice, /insights, /title, /resume, /retry, /undo,
/compress, /usage, /provider, /reload-mcp, /sethome, /reset) used to
interrupt the agent AND get silently discarded by the slash-command
safety net — zero-char response, dropped tool calls.

Root cause:
- Discord registers 41 native slash commands via tree.command().
- Only 14 were in ACTIVE_SESSION_BYPASS_COMMANDS.
- The other ~15 user-facing ones fell through base.py:handle_message
  to the busy-session handler, which calls running_agent.interrupt()
  AND queues the text.
- After the aborted run, gateway/run.py:9912 correctly identifies the
  queued text as a slash command and discards it — but the damage
  (interrupt + zero-char response) already happened.

Fix:
- should_bypass_active_session() now returns True for any resolvable
  slash command. ACTIVE_SESSION_BYPASS_COMMANDS stays as the subset
  with dedicated Level-2 handlers (documentation + tests).
- gateway/run.py adds a catch-all after the dedicated handlers that
  returns a user-visible "agent busy — wait or /stop first" response
  for any other resolvable command.
- Unknown text / file-path-like messages are unchanged — they still
  queue.

Also:
- gateway/platforms/discord.py logs the invoker identity on every
  slash command (user id + name + channel + guild) so future
  ghost-command reports can be triaged without guessing.

Tests:
- 15 new parametrized cases in test_command_bypass_active_session.py
  cover every previously-broken Discord slash command.
- Existing tests for /stop, /new, /approve, /deny, /help, /status,
  /agents, /background, /steer, /update, /queue still pass.
- test_steer.py's ACTIVE_SESSION_BYPASS_COMMANDS check still passes.

Fixes NousResearch#5057. Related: NousResearch#6252, NousResearch#10370, NousResearch#4665.
Egavasyug pushed a commit to Egavasyug/hermes-agent that referenced this pull request Jun 10, 2026
@teknium1
fix(gateway): slash commands never interrupt a running agent (NousRes…
fcb9611 
…earch#12334)

Any recognized slash command now bypasses the Level-1 active-session
guard instead of queueing + interrupting. A mid-run /model (or
/reasoning, /voice, /insights, /title, /resume, /retry, /undo,
/compress, /usage, /provider, /reload-mcp, /sethome, /reset) used to
interrupt the agent AND get silently discarded by the slash-command
safety net — zero-char response, dropped tool calls.

Root cause:
- Discord registers 41 native slash commands via tree.command().
- Only 14 were in ACTIVE_SESSION_BYPASS_COMMANDS.
- The other ~15 user-facing ones fell through base.py:handle_message
  to the busy-session handler, which calls running_agent.interrupt()
  AND queues the text.
- After the aborted run, gateway/run.py:9912 correctly identifies the
  queued text as a slash command and discards it — but the damage
  (interrupt + zero-char response) already happened.

Fix:
- should_bypass_active_session() now returns True for any resolvable
  slash command. ACTIVE_SESSION_BYPASS_COMMANDS stays as the subset
  with dedicated Level-2 handlers (documentation + tests).
- gateway/run.py adds a catch-all after the dedicated handlers that
  returns a user-visible "agent busy — wait or /stop first" response
  for any other resolvable command.
- Unknown text / file-path-like messages are unchanged — they still
  queue.

Also:
- gateway/platforms/discord.py logs the invoker identity on every
  slash command (user id + name + channel + guild) so future
  ghost-command reports can be triaged without guessing.

Tests:
- 15 new parametrized cases in test_command_bypass_active_session.py
  cover every previously-broken Discord slash command.
- Existing tests for /stop, /new, /approve, /deny, /help, /status,
  /agents, /background, /steer, /update, /queue still pass.
- test_steer.py's ACTIVE_SESSION_BYPASS_COMMANDS check still passes.

Fixes NousResearch#5057. Related: NousResearch#6252, NousResearch#10370, NousResearch#4665.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

No reviews

Assignees

No one assigned

Labels

comp/gateway Gateway runner, session dispatch, delivery P2 Medium — degraded but workaround exists type/bug Something isn't working

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

2 participants

@j30231 @alt-glitch