fix(gateway,plugins): make discord disable actually disable, stop reconnect storm (#30736) by xxxigm · Pull Request #30762 · NousResearch/hermes-agent

xxxigm · 2026-05-23T05:07:10Z

What does this PR do?

Fixes #30736: when a user runs hermes plugins disable platforms/discord, the gateway still loaded the discord adapter, spammed ERROR hermes_plugins.discord_platform.adapter: [Discord] No bot token configured on every start AND every reconnect attempt, and queued discord into the reconnect watcher until the per-platform circuit breaker paused it after 10 failures.

Root cause is a plugin-key mismatch between the CLI and the loader. hermes plugins list / hermes plugins disable use the path-derived key platforms/discord, while PluginManager.discover_and_load scanned plugins/platforms/ with no category prefix and therefore registered the bundled adapter under its manifest name (discord-platform). The disabled-check OR'ed both forms but neither matched, so the plugin loaded as if nothing happened. The bug affects every bundled platform plugin (discord, teams, irc, line, simplex, google_chat).

This PR fixes the root cause and hardens the gateway against the same class of misconfiguration so future drift cannot reproduce the original symptom.

Related Issue

Closes #30736

Type of Change

🐛 Bug fix (non-breaking change that fixes an issue)
✨ New feature (non-breaking change that adds functionality)
🔒 Security fix
📝 Documentation update
✅ Tests (adding or improving test coverage)
♻️ Refactor (no behavior change)
🎯 New skill (bundled or hub)

Changes Made

hermes_cli/plugins.py — fix bundled platform key derivation + add helper (+39 / −10):
- Drop "platforms" from skip_names so the recursive scanner picks up plugins/platforms/ as a category and produces path-derived keys (platforms/discord, platforms/teams, …).
- Remove the now-redundant scan_directory(plugins/platforms) block that would otherwise produce duplicate entries under both keys.
- Add is_platform_plugin_disabled(name) helper that resolves both the new path-derived key and the legacy manifest-name key, so callers don't re-implement the matching logic.
gateway/run.py — skip disabled platform plugins cleanly at startup (+32):
- Before _create_adapter, call is_platform_plugin_disabled(platform.value). If true: log one INFO line that names the platform and the command to re-enable it (hermes plugins enable platforms/<name>) or silence the notice (platforms.<name>.enabled=false), mark runtime status as disabled, and skip the adapter entirely so no _failed_platforms entry / no reconnect watcher activity / no ERROR log spam.
- Wrapped in try/except so a corrupt config can't crash startup — the gateway falls through to the legacy adapter path in that case.
tests/gateway/test_disabled_platform_plugin_30736.py — 17 new test cases across 3 classes (+472 lines):
- TestBundledPlatformKeyAlignment — pins every bundled platform adapter under platforms/<name> so any future refactor that re-introduces the manifest-name key fails immediately.
- TestDisableHonoursBothKeyForms — covers the new path-derived key AND the legacy manifest-name key (back-compat for hand-edited configs), plus is_platform_plugin_disabled() argument hardening ("", None).
- TestGatewaySkipsDisabledPlatformPlugin — boots GatewayRunner with a sentinel _create_adapter that raises if invoked for the disabled platform; asserts the platform is skipped before adapter creation, a single actionable INFO log fires, no ERROR/WARNING leaks, _failed_platforms stays clean, the platform-config enabled=False short-circuit still runs first, and exceptions inside the lookup are caught.

How to Test

Check out this branch and ensure .venv is set up: python3 -m venv .venv && source .venv/bin/activate && pip install -e ".[all,dev]"

Run the new regression tests on their own:

scripts/run_tests.sh tests/gateway/test_disabled_platform_plugin_30736.py -v

Expected: 17 passed.

Run the broader plugin + startup + discord adapter suite to confirm no cross-file regressions:

scripts/run_tests.sh tests/hermes_cli/test_plugins.py tests/hermes_cli/test_plugins_cmd.py tests/gateway/test_runner_startup_failures.py tests/gateway/test_discord_connect.py tests/test_plugin_skills.py

Expected: 196 passed.

End-to-end manual verification of the original symptom:

# Before the fix:
hermes plugins disable platforms/discord
hermes plugins list | grep discord     # prints `disabled`
hermes gateway restart
tail -f ~/.hermes/logs/gateway.log     # ERROR per startup + reconnect storm

# After the fix:
hermes plugins disable platforms/discord
hermes plugins list | grep discord     # prints `disabled`
hermes gateway restart
tail -f ~/.hermes/logs/gateway.log
# → single INFO: "Skipping platform 'discord': its plugin is disabled
#    (run 'hermes plugins enable platforms/discord' to re-enable, or set
#     platforms.discord.enabled=false in config.yaml to silence this notice)."
# → no ERROR, no reconnect attempts, no circuit breaker.

Checklist

Code

I've read the Contributing Guide
My commit messages follow Conventional Commits (fix(plugins): …, fix(gateway): …, test(gateway): …)
I searched for existing PRs to make sure this isn't a duplicate
My PR contains only changes related to this fix (no unrelated commits)
I've run scripts/run_tests.sh tests/gateway/test_disabled_platform_plugin_30736.py and all tests pass
I've added tests for my changes (17 new test cases across 3 classes)
I've tested on my platform: macOS 15.2 (Darwin 24.6.0), Python 3.12

Documentation & Housekeeping

I've updated relevant documentation (README, docs/, docstrings) — N/A (no user-facing CLI/config change)
I've updated cli-config.yaml.example if I added/changed config keys — N/A (no new config keys)
I've updated CONTRIBUTING.md or AGENTS.md if I changed architecture or workflows — N/A
I've considered cross-platform impact (Windows, macOS) per the compatibility guide — tests are hermetic (monkeypatch + mocks, no real filesystem I/O outside tmp_path)
I've updated tool descriptions/schemas if I changed tool behavior — N/A

Screenshots / Logs

$ scripts/run_tests.sh tests/gateway/test_disabled_platform_plugin_30736.py -v
17 passed in 4.71s

$ scripts/run_tests.sh tests/hermes_cli/test_plugins.py tests/hermes_cli/test_plugins_cmd.py \
    tests/gateway/test_runner_startup_failures.py tests/gateway/test_discord_connect.py \
    tests/test_plugin_skills.py
196 passed in 5.18s

$ scripts/run_tests.sh tests/gateway/test_disabled_platform_plugin_30736.py \
    tests/gateway/test_runner_startup_failures.py tests/gateway/test_discord_connect.py \
    tests/gateway/test_discord_free_response.py tests/gateway/test_discord_channel_controls.py
90 passed in 6.63s

…usResearch#30736) `hermes plugins list` and `hermes plugins disable` always referred to bundled platform plugins by their path-derived key (`platforms/discord`, `platforms/teams`, …). But the loader scanned `plugins/platforms/` with no category prefix, which keyed those same plugins by their manifest name (`discord-platform`, `teams-platform`, …). The mismatch meant `hermes plugins disable platforms/discord` saved `platforms/discord` to `plugins.disabled`, the loader looked up `discord-platform`, no match fired, and the plugin loaded as if nothing happened. Drop `platforms` from `skip_names` and remove the redundant `scan_directory(plugins/platforms)` so the recursive scanner picks the directory up as a category. Bundled platform plugins now register under `platforms/<name>`, matching what the CLI reports and what the disable command writes. Also expose `is_platform_plugin_disabled(name)` so the gateway runtime can check either key form (path-derived or legacy manifest name) without re-implementing the matching logic.

…Research#30736) When a platform's `enabled` flag in config.yaml is still true but the user has explicitly disabled the matching plugin (e.g. `hermes plugins disable platforms/discord`), the startup loop now short-circuits before `_create_adapter`. Without this guard the adapter would be instantiated, `connect()` would fail with "No bot token configured", the platform would be queued into the reconnect watcher, and the user would see an ERROR-per-attempt log storm forever after — exactly the behaviour reported in NousResearch#30736. The skip path: * logs one INFO line that names the platform and the command needed to re-enable it (`hermes plugins enable platforms/<name>`) or to silence the notice (`platforms.<name>.enabled=false`), * records the runtime status as `disabled` so `gateway status` / dashboards reflect the actual situation, * never enters `_failed_platforms`, so the reconnect watcher never fires for it. Wrapped in `try/except` so a corrupt config can't crash startup — the gateway falls back to the existing adapter path in that case.

…usResearch#30736) 17 tests across three classes: * `TestBundledPlatformKeyAlignment` — pins every bundled platform adapter under `platforms/<name>` so any future refactor that reintroduces the manifest-name key fails immediately. * `TestDisableHonoursBothKeyForms` — covers the new path-derived key AND the legacy manifest-name key (back-compat for hand-edited configs), plus `is_platform_plugin_disabled()` argument hardening. * `TestGatewaySkipsDisabledPlatformPlugin` — boots `GatewayRunner` with a sentinel `_create_adapter` that raises if invoked for the disabled platform. Asserts the platform is skipped before adapter creation, a single actionable INFO log fires, no ERROR/WARNING leaks, the `_failed_platforms` reconnect queue stays clean, and that lookup failures fall through to the legacy adapter path.

grepsuzette · 2026-05-24T13:42:59Z

Sick of discord. But this PR is a bit too big...
Ideologically and practically, works as well rm -fr plugins/platforms/discord

melroy89 · 2026-05-24T14:04:47Z

Sick of discord. But this PR is a bit too big... Ideologically and practically, works as well rm -fr plugins/platforms/discord

I'm afraid a simple hermes update will restore the discord platform bot stuff.

aguysomewhere · 2026-05-24T20:53:48Z

Sick of discord. But this PR is a bit too big... Ideologically and practically, works as well rm -fr plugins/platforms/discord

So I tried this in my container and went out of my way to remove perms for the user:

They restored when I brought the container up. I guess the init processes run as root. I would appreciate the merge but could also go for something leaner. I have zero knowledge of this code base but would love for Discord to go tf away.

Thanks all...

Edit: for now I volume mounted an empty directory as read only and that works - in case this helps anyone

xxxigm added 3 commits May 23, 2026 12:05

alt-glitch mentioned this pull request May 25, 2026

Discord platform: auto-enables when discord.py is present, and hermes plugins disable silently no-ops #32058

Open

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

fix(gateway,plugins): make discord disable actually disable, stop reconnect storm (#30736)#30762

fix(gateway,plugins): make discord disable actually disable, stop reconnect storm (#30736)#30762
xxxigm wants to merge 3 commits into
NousResearch:mainfrom
xxxigm:fix/30736-discord-disabled-still-connects

xxxigm commented May 23, 2026 •

edited

Loading

Uh oh!

grepsuzette commented May 24, 2026 •

edited

Loading

Uh oh!

melroy89 commented May 24, 2026

Uh oh!

aguysomewhere commented May 24, 2026 •

edited

Loading

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants

Conversation

xxxigm commented May 23, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

What does this PR do?

Related Issue

Type of Change

Changes Made

How to Test

Checklist

Code

Documentation & Housekeeping

Screenshots / Logs

Uh oh!

grepsuzette commented May 24, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

melroy89 commented May 24, 2026

Uh oh!

aguysomewhere commented May 24, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants

xxxigm commented May 23, 2026 •

edited

Loading

grepsuzette commented May 24, 2026 •

edited

Loading

aguysomewhere commented May 24, 2026 •

edited

Loading