feat(chat): Add per-agent verbosity control across all chat frontends (Discord, Slack, Web)

[edspencer]

Summary

Add a configurable verbosity level per agent across all chat frontends (Discord, Slack, Web) so users can control how much output they see — from "just the final answer" to "everything I'd see in Claude Code."

Reported by community member Oliver on Discord.

Current Behavior

Each agent can configure output via chat.<platform>.output, backed by the shared ChatOutputSchema:

chat:
  discord:
    output:
      tool_results: true
      tool_result_max_length: 900
      system_status: true
      errors: true

This provides basic toggles, but there's no way to:

• Show/hide tool calls themselves (only tool results are toggleable)
• Suppress all intermediate output and only show the final assistant text
• Show streaming/progress events for a fully verbose mode

Users currently have to modify each frontend's implementation to achieve different verbosity levels per agent.

Proposed Solution

Add a verbosity field to the shared ChatOutputSchema so it applies to all three chat frontends:

chat:
  discord:
    output:
      verbosity: minimal
  slack:
    output:
      verbosity: standard
  web:
    output:
      verbosity: verbose

Level	Behavior
minimal	Only final assistant text + errors. No tool calls, no tool results, no system status.
standard	Current default behavior (tool results, system status, errors).
verbose	Everything: tool calls, tool results, system status, streaming/progress events. Closest to the raw Claude Code experience.

The existing granular toggles (tool_results, system_status, etc.) should continue to work and override the verbosity preset when explicitly set.

Implementation Notes

Schema (shared)

• packages/core/src/config/schema.ts — Add verbosity enum to the shared ChatOutputSchema so all platform-specific schemas (DiscordOutputSchema, SlackOutputSchema, etc.) inherit it automatically.

Frontend-specific filtering

Each frontend has its own message handler that would need verbosity-aware filtering:

1. packages/discord/src/manager.ts — onMessage callback (~line 449-605)
2. packages/slack/src/manager.ts — equivalent message handler
3. packages/web/ — web dashboard message display

Shared utilities

Consider extracting a shared shouldShowMessage(messageType, outputConfig) helper into @herdctl/chat so all three frontends apply verbosity rules consistently rather than duplicating the filtering logic.

Scope

• Schema change: ~10 lines
• Shared helper: ~30-50 lines
• Per-frontend wiring: ~20-30 lines each
• Fully backward-compatible — existing configs default to "standard"

[oheckmann74]

yes, this would be great. Does the CLI distinguish between the final assistant text and tool results, thinking etc? I thought that was only in the SDK but is lost when tracking the CLI output

[oheckmann74]

I will play with this a bit today or tomorrow and have a more informed opinion after that.