Skip to content

fix(kanban): CLI dispatch config passthrough + humanizer skill swap (closes #33488, closes #29415) - #34337

Merged
teknium1 merged 1 commit into
mainfrom
hermes/hermes-4f9b6288
May 29, 2026
Merged

fix(kanban): CLI dispatch config passthrough + humanizer skill swap (closes #33488, closes #29415)#34337
teknium1 merged 1 commit into
mainfrom
hermes/hermes-4f9b6288

Conversation

@teknium1

Copy link
Copy Markdown
Contributor

Two small bugs in the kanban dispatcher's CLI surface, both silently degrading distinct workflows. Bundled because the test files and the surrounding code surface overlap.

#33488: hermes kanban dispatch ignored kanban.max_in_progress / max_spawn

The CLI wrapper in hermes_cli/kanban.py:_cmd_dispatch only passed default_assignee and max_in_progress_per_profile (from #34244) through to dispatch_once. The global concurrency cap (kanban.max_in_progress) and the per-tick spawn limit (kanban.max_spawn) were silently dropped, so operators using hermes kanban dispatch as a one-shot or in a custom loop couldn't reach either cap from config — only the gateway-embedded dispatcher honored them.

Fix: read both keys from config in the same _coerce_positive_int helper that already handled max_in_progress_per_profile. CLI --max still wins over config kanban.max_spawn when both are present (explicit operator signal beats default), but absent --max falls back to config.

# After
max_in_progress = _coerce_positive_int(_kanban_cfg.get("max_in_progress"))
cli_max = getattr(args, "max", None)
max_spawn = cli_max if cli_max is not None else _coerce_positive_int(
    _kanban_cfg.get("max_spawn")
)
res = kb.dispatch_once(
    conn,
    max_spawn=max_spawn,
    max_in_progress=max_in_progress,   # NEW: was missing
    failure_limit=...,
    default_assignee=default_assignee,
    max_in_progress_per_profile=max_in_progress_per_profile,
)

CLI now matches gateway dispatcher behavior exactly.

#29415: synthesizer crashed in retry loop on missing skill

hermes_cli/kanban_swarm.py:212 hardcoded skills=["avoid-ai-writing"], a skill that doesn't exist in the bundled skills/ directory or any registered hub source. Every synthesizer worker spawn failed at CLI startup with Unknown skill(s): avoid-ai-writing before the agent loop even started — the dispatcher retried up to failure_limit (default 2), then auto-blocked the task, then dependency rules could re-promote it, looping forever until manual intervention.

Fix: replace with humanizer, which IS bundled at skills/creative/humanizer/SKILL.md (description: "Humanize text: strip AI-isms and add real voice"). That's the obvious intent behind the avoid-ai-writing name, and the skill is platform-portable (linux/macos/windows) so it works on every supported runtime.

Validation

tests/hermes_cli/test_kanban_cli_dispatch_passthrough.py4 new cases:

Test What it checks
test_cli_dispatch_passes_max_in_progress_from_config CLI passes all 4 config keys to dispatch_once: max_in_progress, max_spawn, default_assignee, max_in_progress_per_profile
test_cli_max_flag_overrides_config_max_spawn --max=2 on the CLI wins over kanban.max_spawn: 10 in config
test_cli_invalid_max_in_progress_silently_disables Invalid cap values (0, -1, "abc", "1.5") silently fall through to None (parametrized)
test_kanban_swarm_uses_existing_humanizer_skill Source-level invariant: kanban_swarm.py no longer references avoid-ai-writing AND the replacement humanizer skill exists at the expected on-disk path

Broader kanban suite: 468/468 pass (was 464; +4 new regression tests).

Credit

Infographic

kanban-cli-dispatch-fixes

…g; swap missing 'avoid-ai-writing' skill for bundled humanizer (#33488, #29415)

Two small bugs in the kanban dispatcher's CLI surface that were
silently degrading two distinct workflows. Bundled because the test
files and the surrounding code surface overlap.

## #33488: hermes kanban dispatch ignored kanban.max_in_progress / max_spawn

The CLI wrapper in hermes_cli/kanban.py:_cmd_dispatch only passed
default_assignee and max_in_progress_per_profile through to
dispatch_once. The global concurrency cap (kanban.max_in_progress)
and the per-tick spawn limit (kanban.max_spawn) were silently dropped,
so operators using 'hermes kanban dispatch' as a one-shot or in a
custom loop couldn't reach either cap from config — only the gateway
embedded dispatcher honored them.

Fix: read both keys from config in the same coerce-positive-int
helper that already handled max_in_progress_per_profile. CLI --max
still wins over config kanban.max_spawn when both are present
(explicit operator signal beats default), but absent --max falls
back to config.

## #29415: synthesizer crashed in retry loop on missing skill

hermes_cli/kanban_swarm.py:212 hardcoded skills=['avoid-ai-writing'],
a skill that doesn't exist in the bundled skills/ directory or any
registered hub source. Every synthesizer worker spawn failed at CLI
startup with 'Unknown skill(s): avoid-ai-writing' before the agent
loop even started — the dispatcher retried up to failure_limit
(default 2), then auto-blocked the task, then dependency rules could
re-promote it, looping forever until manual intervention.

Fix: replace with 'humanizer' which is bundled at
skills/creative/humanizer/SKILL.md (description: 'Humanize text:
strip AI-isms and add real voice'). That's the obvious intent behind
the 'avoid-ai-writing' name, and the skill is platform-portable
(linux/macos/windows) so it works on every supported runtime.

## Tests

tests/hermes_cli/test_kanban_cli_dispatch_passthrough.py — 4 cases:
- CLI passes max_in_progress / max_spawn / default_assignee /
  max_in_progress_per_profile from config to dispatch_once
- CLI --max flag overrides config kanban.max_spawn
- Invalid cap values (0, -1, 'abc', '1.5') silently fall through to None
- kanban_swarm.py no longer references 'avoid-ai-writing' AND the
  replacement 'humanizer' skill exists at the expected on-disk path

Kanban suite: 468/468 pass (was 464; +4 new regression tests).
@github-actions

Copy link
Copy Markdown
Contributor

🔎 Lint report: hermes/hermes-4f9b6288 vs origin/main

ruff

Total: 0 on HEAD, 0 on base (➖ 0)

🆕 New issues: none

✅ Fixed issues: none

Unchanged: 0 pre-existing issues carried over.

ty (type checker)

Total: 9582 on HEAD, 9581 on base (🆕 +1)

🆕 New issues (1):

Rule Count
unresolved-import 1
First entries
tests/hermes_cli/test_kanban_cli_dispatch_passthrough.py:17: [unresolved-import] unresolved-import: Cannot resolve imported module `pytest`

✅ Fixed issues: none

Unchanged: 5050 pre-existing issues carried over.

Diagnostics are surfaced as warnings — this check never fails the build.

@teknium1
teknium1 merged commit 69b74c1 into main May 29, 2026
24 of 25 checks passed
@teknium1
teknium1 deleted the hermes/hermes-4f9b6288 branch May 29, 2026 04:00
@alt-glitch alt-glitch added type/bug Something isn't working P3 Low — cosmetic, nice to have comp/cli CLI entry point, hermes_cli/, setup wizard comp/plugins Plugin system and bundled plugins labels May 29, 2026
KKT-OPT pushed a commit to KKT-OPT/hermes-agent that referenced this pull request May 31, 2026
…g; swap missing 'avoid-ai-writing' skill for bundled humanizer (NousResearch#33488, NousResearch#29415) (NousResearch#34337)

Two small bugs in the kanban dispatcher's CLI surface that were
silently degrading two distinct workflows. Bundled because the test
files and the surrounding code surface overlap.

## NousResearch#33488: hermes kanban dispatch ignored kanban.max_in_progress / max_spawn

The CLI wrapper in hermes_cli/kanban.py:_cmd_dispatch only passed
default_assignee and max_in_progress_per_profile through to
dispatch_once. The global concurrency cap (kanban.max_in_progress)
and the per-tick spawn limit (kanban.max_spawn) were silently dropped,
so operators using 'hermes kanban dispatch' as a one-shot or in a
custom loop couldn't reach either cap from config — only the gateway
embedded dispatcher honored them.

Fix: read both keys from config in the same coerce-positive-int
helper that already handled max_in_progress_per_profile. CLI --max
still wins over config kanban.max_spawn when both are present
(explicit operator signal beats default), but absent --max falls
back to config.

## NousResearch#29415: synthesizer crashed in retry loop on missing skill

hermes_cli/kanban_swarm.py:212 hardcoded skills=['avoid-ai-writing'],
a skill that doesn't exist in the bundled skills/ directory or any
registered hub source. Every synthesizer worker spawn failed at CLI
startup with 'Unknown skill(s): avoid-ai-writing' before the agent
loop even started — the dispatcher retried up to failure_limit
(default 2), then auto-blocked the task, then dependency rules could
re-promote it, looping forever until manual intervention.

Fix: replace with 'humanizer' which is bundled at
skills/creative/humanizer/SKILL.md (description: 'Humanize text:
strip AI-isms and add real voice'). That's the obvious intent behind
the 'avoid-ai-writing' name, and the skill is platform-portable
(linux/macos/windows) so it works on every supported runtime.

## Tests

tests/hermes_cli/test_kanban_cli_dispatch_passthrough.py — 4 cases:
- CLI passes max_in_progress / max_spawn / default_assignee /
  max_in_progress_per_profile from config to dispatch_once
- CLI --max flag overrides config kanban.max_spawn
- Invalid cap values (0, -1, 'abc', '1.5') silently fall through to None
- kanban_swarm.py no longer references 'avoid-ai-writing' AND the
  replacement 'humanizer' skill exists at the expected on-disk path

Kanban suite: 468/468 pass (was 464; +4 new regression tests).
alt-glitch pushed a commit that referenced this pull request Jun 14, 2026
…g; swap missing 'avoid-ai-writing' skill for bundled humanizer (#33488, #29415) (#34337)

Two small bugs in the kanban dispatcher's CLI surface that were
silently degrading two distinct workflows. Bundled because the test
files and the surrounding code surface overlap.

## #33488: hermes kanban dispatch ignored kanban.max_in_progress / max_spawn

The CLI wrapper in hermes_cli/kanban.py:_cmd_dispatch only passed
default_assignee and max_in_progress_per_profile through to
dispatch_once. The global concurrency cap (kanban.max_in_progress)
and the per-tick spawn limit (kanban.max_spawn) were silently dropped,
so operators using 'hermes kanban dispatch' as a one-shot or in a
custom loop couldn't reach either cap from config — only the gateway
embedded dispatcher honored them.

Fix: read both keys from config in the same coerce-positive-int
helper that already handled max_in_progress_per_profile. CLI --max
still wins over config kanban.max_spawn when both are present
(explicit operator signal beats default), but absent --max falls
back to config.

## #29415: synthesizer crashed in retry loop on missing skill

hermes_cli/kanban_swarm.py:212 hardcoded skills=['avoid-ai-writing'],
a skill that doesn't exist in the bundled skills/ directory or any
registered hub source. Every synthesizer worker spawn failed at CLI
startup with 'Unknown skill(s): avoid-ai-writing' before the agent
loop even started — the dispatcher retried up to failure_limit
(default 2), then auto-blocked the task, then dependency rules could
re-promote it, looping forever until manual intervention.

Fix: replace with 'humanizer' which is bundled at
skills/creative/humanizer/SKILL.md (description: 'Humanize text:
strip AI-isms and add real voice'). That's the obvious intent behind
the 'avoid-ai-writing' name, and the skill is platform-portable
(linux/macos/windows) so it works on every supported runtime.

## Tests

tests/hermes_cli/test_kanban_cli_dispatch_passthrough.py — 4 cases:
- CLI passes max_in_progress / max_spawn / default_assignee /
  max_in_progress_per_profile from config to dispatch_once
- CLI --max flag overrides config kanban.max_spawn
- Invalid cap values (0, -1, 'abc', '1.5') silently fall through to None
- kanban_swarm.py no longer references 'avoid-ai-writing' AND the
  replacement 'humanizer' skill exists at the expected on-disk path

Kanban suite: 468/468 pass (was 464; +4 new regression tests).
T02200059 pushed a commit to T02200059/hermes-agent that referenced this pull request Jun 18, 2026
…g; swap missing 'avoid-ai-writing' skill for bundled humanizer (NousResearch#33488, NousResearch#29415) (NousResearch#34337)

Two small bugs in the kanban dispatcher's CLI surface that were
silently degrading two distinct workflows. Bundled because the test
files and the surrounding code surface overlap.

## NousResearch#33488: hermes kanban dispatch ignored kanban.max_in_progress / max_spawn

The CLI wrapper in hermes_cli/kanban.py:_cmd_dispatch only passed
default_assignee and max_in_progress_per_profile through to
dispatch_once. The global concurrency cap (kanban.max_in_progress)
and the per-tick spawn limit (kanban.max_spawn) were silently dropped,
so operators using 'hermes kanban dispatch' as a one-shot or in a
custom loop couldn't reach either cap from config — only the gateway
embedded dispatcher honored them.

Fix: read both keys from config in the same coerce-positive-int
helper that already handled max_in_progress_per_profile. CLI --max
still wins over config kanban.max_spawn when both are present
(explicit operator signal beats default), but absent --max falls
back to config.

## NousResearch#29415: synthesizer crashed in retry loop on missing skill

hermes_cli/kanban_swarm.py:212 hardcoded skills=['avoid-ai-writing'],
a skill that doesn't exist in the bundled skills/ directory or any
registered hub source. Every synthesizer worker spawn failed at CLI
startup with 'Unknown skill(s): avoid-ai-writing' before the agent
loop even started — the dispatcher retried up to failure_limit
(default 2), then auto-blocked the task, then dependency rules could
re-promote it, looping forever until manual intervention.

Fix: replace with 'humanizer' which is bundled at
skills/creative/humanizer/SKILL.md (description: 'Humanize text:
strip AI-isms and add real voice'). That's the obvious intent behind
the 'avoid-ai-writing' name, and the skill is platform-portable
(linux/macos/windows) so it works on every supported runtime.

## Tests

tests/hermes_cli/test_kanban_cli_dispatch_passthrough.py — 4 cases:
- CLI passes max_in_progress / max_spawn / default_assignee /
  max_in_progress_per_profile from config to dispatch_once
- CLI --max flag overrides config kanban.max_spawn
- Invalid cap values (0, -1, 'abc', '1.5') silently fall through to None
- kanban_swarm.py no longer references 'avoid-ai-writing' AND the
  replacement 'humanizer' skill exists at the expected on-disk path

Kanban suite: 468/468 pass (was 464; +4 new regression tests).
waefrebeorn pushed a commit to waefrebeorn/slermes that referenced this pull request Jul 2, 2026
…g; swap missing 'avoid-ai-writing' skill for bundled humanizer (NousResearch#33488, NousResearch#29415) (NousResearch#34337)

Two small bugs in the kanban dispatcher's CLI surface that were
silently degrading two distinct workflows. Bundled because the test
files and the surrounding code surface overlap.

## NousResearch#33488: hermes kanban dispatch ignored kanban.max_in_progress / max_spawn

The CLI wrapper in hermes_cli/kanban.py:_cmd_dispatch only passed
default_assignee and max_in_progress_per_profile through to
dispatch_once. The global concurrency cap (kanban.max_in_progress)
and the per-tick spawn limit (kanban.max_spawn) were silently dropped,
so operators using 'hermes kanban dispatch' as a one-shot or in a
custom loop couldn't reach either cap from config — only the gateway
embedded dispatcher honored them.

Fix: read both keys from config in the same coerce-positive-int
helper that already handled max_in_progress_per_profile. CLI --max
still wins over config kanban.max_spawn when both are present
(explicit operator signal beats default), but absent --max falls
back to config.

## NousResearch#29415: synthesizer crashed in retry loop on missing skill

hermes_cli/kanban_swarm.py:212 hardcoded skills=['avoid-ai-writing'],
a skill that doesn't exist in the bundled skills/ directory or any
registered hub source. Every synthesizer worker spawn failed at CLI
startup with 'Unknown skill(s): avoid-ai-writing' before the agent
loop even started — the dispatcher retried up to failure_limit
(default 2), then auto-blocked the task, then dependency rules could
re-promote it, looping forever until manual intervention.

Fix: replace with 'humanizer' which is bundled at
skills/creative/humanizer/SKILL.md (description: 'Humanize text:
strip AI-isms and add real voice'). That's the obvious intent behind
the 'avoid-ai-writing' name, and the skill is platform-portable
(linux/macos/windows) so it works on every supported runtime.

## Tests

tests/hermes_cli/test_kanban_cli_dispatch_passthrough.py — 4 cases:
- CLI passes max_in_progress / max_spawn / default_assignee /
  max_in_progress_per_profile from config to dispatch_once
- CLI --max flag overrides config kanban.max_spawn
- Invalid cap values (0, -1, 'abc', '1.5') silently fall through to None
- kanban_swarm.py no longer references 'avoid-ai-writing' AND the
  replacement 'humanizer' skill exists at the expected on-disk path

Kanban suite: 468/468 pass (was 464; +4 new regression tests).
santhreal pushed a commit to santhreal/hermes-agent that referenced this pull request Jul 13, 2026
…g; swap missing 'avoid-ai-writing' skill for bundled humanizer (NousResearch#33488, NousResearch#29415) (NousResearch#34337)

Two small bugs in the kanban dispatcher's CLI surface that were
silently degrading two distinct workflows. Bundled because the test
files and the surrounding code surface overlap.

## NousResearch#33488: hermes kanban dispatch ignored kanban.max_in_progress / max_spawn

The CLI wrapper in hermes_cli/kanban.py:_cmd_dispatch only passed
default_assignee and max_in_progress_per_profile through to
dispatch_once. The global concurrency cap (kanban.max_in_progress)
and the per-tick spawn limit (kanban.max_spawn) were silently dropped,
so operators using 'hermes kanban dispatch' as a one-shot or in a
custom loop couldn't reach either cap from config — only the gateway
embedded dispatcher honored them.

Fix: read both keys from config in the same coerce-positive-int
helper that already handled max_in_progress_per_profile. CLI --max
still wins over config kanban.max_spawn when both are present
(explicit operator signal beats default), but absent --max falls
back to config.

## NousResearch#29415: synthesizer crashed in retry loop on missing skill

hermes_cli/kanban_swarm.py:212 hardcoded skills=['avoid-ai-writing'],
a skill that doesn't exist in the bundled skills/ directory or any
registered hub source. Every synthesizer worker spawn failed at CLI
startup with 'Unknown skill(s): avoid-ai-writing' before the agent
loop even started — the dispatcher retried up to failure_limit
(default 2), then auto-blocked the task, then dependency rules could
re-promote it, looping forever until manual intervention.

Fix: replace with 'humanizer' which is bundled at
skills/creative/humanizer/SKILL.md (description: 'Humanize text:
strip AI-isms and add real voice'). That's the obvious intent behind
the 'avoid-ai-writing' name, and the skill is platform-portable
(linux/macos/windows) so it works on every supported runtime.

## Tests

tests/hermes_cli/test_kanban_cli_dispatch_passthrough.py — 4 cases:
- CLI passes max_in_progress / max_spawn / default_assignee /
  max_in_progress_per_profile from config to dispatch_once
- CLI --max flag overrides config kanban.max_spawn
- Invalid cap values (0, -1, 'abc', '1.5') silently fall through to None
- kanban_swarm.py no longer references 'avoid-ai-writing' AND the
  replacement 'humanizer' skill exists at the expected on-disk path

Kanban suite: 468/468 pass (was 464; +4 new regression tests).
donbowman pushed a commit to donbowman/hermes-agent that referenced this pull request Jul 13, 2026
…g; swap missing 'avoid-ai-writing' skill for bundled humanizer (NousResearch#33488, NousResearch#29415) (NousResearch#34337)

Two small bugs in the kanban dispatcher's CLI surface that were
silently degrading two distinct workflows. Bundled because the test
files and the surrounding code surface overlap.

## NousResearch#33488: hermes kanban dispatch ignored kanban.max_in_progress / max_spawn

The CLI wrapper in hermes_cli/kanban.py:_cmd_dispatch only passed
default_assignee and max_in_progress_per_profile through to
dispatch_once. The global concurrency cap (kanban.max_in_progress)
and the per-tick spawn limit (kanban.max_spawn) were silently dropped,
so operators using 'hermes kanban dispatch' as a one-shot or in a
custom loop couldn't reach either cap from config — only the gateway
embedded dispatcher honored them.

Fix: read both keys from config in the same coerce-positive-int
helper that already handled max_in_progress_per_profile. CLI --max
still wins over config kanban.max_spawn when both are present
(explicit operator signal beats default), but absent --max falls
back to config.

## NousResearch#29415: synthesizer crashed in retry loop on missing skill

hermes_cli/kanban_swarm.py:212 hardcoded skills=['avoid-ai-writing'],
a skill that doesn't exist in the bundled skills/ directory or any
registered hub source. Every synthesizer worker spawn failed at CLI
startup with 'Unknown skill(s): avoid-ai-writing' before the agent
loop even started — the dispatcher retried up to failure_limit
(default 2), then auto-blocked the task, then dependency rules could
re-promote it, looping forever until manual intervention.

Fix: replace with 'humanizer' which is bundled at
skills/creative/humanizer/SKILL.md (description: 'Humanize text:
strip AI-isms and add real voice'). That's the obvious intent behind
the 'avoid-ai-writing' name, and the skill is platform-portable
(linux/macos/windows) so it works on every supported runtime.

## Tests

tests/hermes_cli/test_kanban_cli_dispatch_passthrough.py — 4 cases:
- CLI passes max_in_progress / max_spawn / default_assignee /
  max_in_progress_per_profile from config to dispatch_once
- CLI --max flag overrides config kanban.max_spawn
- Invalid cap values (0, -1, 'abc', '1.5') silently fall through to None
- kanban_swarm.py no longer references 'avoid-ai-writing' AND the
  replacement 'humanizer' skill exists at the expected on-disk path

Kanban suite: 468/468 pass (was 464; +4 new regression tests).
Gravezzz pushed a commit to Gravezzz/hermes-agent that referenced this pull request Jul 21, 2026
…g; swap missing 'avoid-ai-writing' skill for bundled humanizer (NousResearch#33488, NousResearch#29415) (NousResearch#34337)

Two small bugs in the kanban dispatcher's CLI surface that were
silently degrading two distinct workflows. Bundled because the test
files and the surrounding code surface overlap.

## NousResearch#33488: hermes kanban dispatch ignored kanban.max_in_progress / max_spawn

The CLI wrapper in hermes_cli/kanban.py:_cmd_dispatch only passed
default_assignee and max_in_progress_per_profile through to
dispatch_once. The global concurrency cap (kanban.max_in_progress)
and the per-tick spawn limit (kanban.max_spawn) were silently dropped,
so operators using 'hermes kanban dispatch' as a one-shot or in a
custom loop couldn't reach either cap from config — only the gateway
embedded dispatcher honored them.

Fix: read both keys from config in the same coerce-positive-int
helper that already handled max_in_progress_per_profile. CLI --max
still wins over config kanban.max_spawn when both are present
(explicit operator signal beats default), but absent --max falls
back to config.

## NousResearch#29415: synthesizer crashed in retry loop on missing skill

hermes_cli/kanban_swarm.py:212 hardcoded skills=['avoid-ai-writing'],
a skill that doesn't exist in the bundled skills/ directory or any
registered hub source. Every synthesizer worker spawn failed at CLI
startup with 'Unknown skill(s): avoid-ai-writing' before the agent
loop even started — the dispatcher retried up to failure_limit
(default 2), then auto-blocked the task, then dependency rules could
re-promote it, looping forever until manual intervention.

Fix: replace with 'humanizer' which is bundled at
skills/creative/humanizer/SKILL.md (description: 'Humanize text:
strip AI-isms and add real voice'). That's the obvious intent behind
the 'avoid-ai-writing' name, and the skill is platform-portable
(linux/macos/windows) so it works on every supported runtime.

## Tests

tests/hermes_cli/test_kanban_cli_dispatch_passthrough.py — 4 cases:
- CLI passes max_in_progress / max_spawn / default_assignee /
  max_in_progress_per_profile from config to dispatch_once
- CLI --max flag overrides config kanban.max_spawn
- Invalid cap values (0, -1, 'abc', '1.5') silently fall through to None
- kanban_swarm.py no longer references 'avoid-ai-writing' AND the
  replacement 'humanizer' skill exists at the expected on-disk path

Kanban suite: 468/468 pass (was 464; +4 new regression tests).
leewenjie pushed a commit to leewenjie/hermes-agent that referenced this pull request Aug 7, 2026
…g; swap missing 'avoid-ai-writing' skill for bundled humanizer (NousResearch#33488, NousResearch#29415) (NousResearch#34337)

Two small bugs in the kanban dispatcher's CLI surface that were
silently degrading two distinct workflows. Bundled because the test
files and the surrounding code surface overlap.

## NousResearch#33488: hermes kanban dispatch ignored kanban.max_in_progress / max_spawn

The CLI wrapper in hermes_cli/kanban.py:_cmd_dispatch only passed
default_assignee and max_in_progress_per_profile through to
dispatch_once. The global concurrency cap (kanban.max_in_progress)
and the per-tick spawn limit (kanban.max_spawn) were silently dropped,
so operators using 'hermes kanban dispatch' as a one-shot or in a
custom loop couldn't reach either cap from config — only the gateway
embedded dispatcher honored them.

Fix: read both keys from config in the same coerce-positive-int
helper that already handled max_in_progress_per_profile. CLI --max
still wins over config kanban.max_spawn when both are present
(explicit operator signal beats default), but absent --max falls
back to config.

## NousResearch#29415: synthesizer crashed in retry loop on missing skill

hermes_cli/kanban_swarm.py:212 hardcoded skills=['avoid-ai-writing'],
a skill that doesn't exist in the bundled skills/ directory or any
registered hub source. Every synthesizer worker spawn failed at CLI
startup with 'Unknown skill(s): avoid-ai-writing' before the agent
loop even started — the dispatcher retried up to failure_limit
(default 2), then auto-blocked the task, then dependency rules could
re-promote it, looping forever until manual intervention.

Fix: replace with 'humanizer' which is bundled at
skills/creative/humanizer/SKILL.md (description: 'Humanize text:
strip AI-isms and add real voice'). That's the obvious intent behind
the 'avoid-ai-writing' name, and the skill is platform-portable
(linux/macos/windows) so it works on every supported runtime.

## Tests

tests/hermes_cli/test_kanban_cli_dispatch_passthrough.py — 4 cases:
- CLI passes max_in_progress / max_spawn / default_assignee /
  max_in_progress_per_profile from config to dispatch_once
- CLI --max flag overrides config kanban.max_spawn
- Invalid cap values (0, -1, 'abc', '1.5') silently fall through to None
- kanban_swarm.py no longer references 'avoid-ai-writing' AND the
  replacement 'humanizer' skill exists at the expected on-disk path

Kanban suite: 468/468 pass (was 464; +4 new regression tests).
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

comp/cli CLI entry point, hermes_cli/, setup wizard comp/plugins Plugin system and bundled plugins P3 Low — cosmetic, nice to have type/bug Something isn't working

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants