# Config

> Read and write pawlia configuration settings (interfaces, TTS, transcription, agents, skill-config), override session-local agent selection at runtime, inspect the skill-creator coding configuration (in-process via agents.coder), and toggle private mode. Use this to enable/disable features, change providers, adjust interface settings, configure skill parameters, change active session models, or enable/disable private mode.

- Skill: `cutec-chris/config` (Agent Skill, multi-file: 3 files)
- Install (CLI): `npx skillmds@latest add cutec-chris/config`
- Raw SKILL.md: https://api.skillmd.com/api/skills/cutec-chris/config/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: AI & ML
- License: MIT
- Author: cutec-chris (https://skillmd.com/u/cutec-chris)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/cutec-chris/config

---


# Config Skill

Reads and writes `config.yaml` using dot-notation paths.

## IMPORTANT

Always use the **bash tool** to run the commands below.
`--user-id` and `--session-dir` are NOT needed for this skill.

---

## Settable sections

Only these top-level sections can be modified:

| Section | Examples |
|---------|---------|
| `interfaces` | `interfaces.matrix.always_thread`, `interfaces.web.port` |
| `tts` | `tts.provider`, `tts.hold_audio`, `tts.edge.voice` |
| `transcription` | `transcription.provider`, `transcription.groq.language` |
| `agents` | `agents.default`, `agents.chat`, `agents.skill_runner` |
| `skill-config` | `skill-config.searxng.url`, `skill-config.memory.idle_minutes` |

`providers` and `models` are managed via the web UI and cannot be changed here.

---

## Show current settings

Show all settable config sections:

```
python <scripts_dir>/config.py show
```

Show a single section:

```
python <scripts_dir>/config.py show --section interfaces
```

## Get a specific value

```
python <scripts_dir>/config.py get --path interfaces.matrix.always_thread
```

## Set a value

Values are parsed as YAML scalars: `true`/`false` become booleans, numbers become integers/floats, everything else is a string.

```
python <scripts_dir>/config.py set --path interfaces.matrix.always_thread --value true
python <scripts_dir>/config.py set --path tts.provider --value edge
python <scripts_dir>/config.py set --path tts.edge.voice --value de-DE-KatjaNeural
python <scripts_dir>/config.py set --path transcription.groq.language --value de
python <scripts_dir>/config.py set --path agents.default --value fast
python <scripts_dir>/config.py set --path skill-config.memory.idle_minutes --value 10
```

**TTS validation:** writes to `tts.provider`, `tts.piper.model`, and `tts.edge.voice` are validated — an unknown value is rejected with a list of valid options instead of being written. A wrong voice silently breaks TTS in the VoIP call, so always pick from the returned `available_voices`. To force a value not in the list (e.g. an English Edge voice), append `--force`.

## Switch the active chat model (session shorthand)

Show the current session override and the effective active chat model:

```
python <scripts_dir>/config.py model
```

Switch to a different model for this session (takes effect immediately, no restart needed):

```
python <scripts_dir>/config.py model --name qwen3.5:latest
```

This is shorthand for overriding `agents.chat`.
The model name must match a key in the `models` section of config.yaml.

This is shorthand for overriding `session.agents.chat`.

## Override runtime `agents:` selection

Show all session overrides:

```
python <scripts_dir>/config.py agent
```

Show one override path:

```
python <scripts_dir>/config.py agent --path chat
python <scripts_dir>/config.py agent --path skills.browser
```

Set an override path:

```
python <scripts_dir>/config.py agent --path chat --value smart,fast
python <scripts_dir>/config.py agent --path default --value smart,fast
python <scripts_dir>/config.py agent --path skills.browser --value fast
```

Clear one override path:

```
python <scripts_dir>/config.py agent --path skills.browser --off
```

Supported runtime paths:
- `default`
- `chat`
- `skill_runner`
- `vision`
- `compiler`
- `skills.<name>`

## Switch the TTS voice (per-user)

The voice command is **provider-aware** — it reads `tts.provider` from `config.yaml` and lists/validates voices for that provider. Always run `voice` (without `--name`) first to see the current `available_voices` for the active provider, then pick one from that list. A name that doesn't match the provider is rejected to avoid silently breaking TTS.

Show current voice + available voices for the active provider:

```
python <scripts_dir>/config.py voice
```

Set a voice (persists in `workspace/memory/voice_override.txt`):

```
# When tts.provider = piper:
python <scripts_dir>/config.py voice --name de_DE-thorsten-low

# When tts.provider = edge:
python <scripts_dir>/config.py voice --name de-DE-KatjaNeural
```

Clear the override (falls back to global `tts.piper.model` / `tts.edge.voice`):

```
python <scripts_dir>/config.py voice --off
```

**How `available_voices` is discovered:**
- Piper: glob configured model directories for `*.onnx`. Directory precedence is `PAWLIA_PIPER_DIR`, `PIPER_VOICE_DIR`, `tts.piper.voice_dir`, `tts.piper.model_dir`, the directory of `tts.piper.model` if it is a path, then `/app/piper` (the voice files baked into the VoIP image). **This list is authoritative** — a voice not listed does not exist on disk. Setting a non-listed Piper voice is always rejected, even with `--force`, because Piper would immediately crash with "Model file doesn't exist" and break TTS. If a user asks for a Piper voice that isn't available, tell them so and offer one from the list; never try to force it.
- Edge: `edge_tts.list_voices()` (full Microsoft catalog). If the dynamic listing is unreachable (`edge_tts` not installed / no internet) the list may be empty — only then may `--force` be used to write a voice name the user explicitly asked for:

  ```
  python <scripts_dir>/config.py voice --name en-US-AriaNeural --force
  ```

## Private mode

Enable private mode for the current session (messages won't be saved):

```
python <scripts_dir>/config.py private
```

Enable private mode for a specific thread:

```
python <scripts_dir>/config.py private --thread <thread_id>
```

Disable private mode:

```
python <scripts_dir>/config.py private --off
python <scripts_dir>/config.py private --thread <thread_id> --off
```

## Session config (per-user overrides)

`session/<user>/config.yaml` stores per-user overrides that mirror the global `config.yaml` structure.
Supported sections: `agents`, `tts`, `disabled_skills`, `user`.

The `user` section holds personal settings that govern how the model perceives
the user's environment. Currently:

- `user.timezone` — IANA name like `Europe/Berlin`. When set, the model sees
  the user's local time in its system prompt and in daily log timestamps.
  When unset, the model sees server-local time (UTC in container deploys),
  which leads to confused replies like "kurz vor fünf" at 06:56 local. Always
  set this if the user has mentioned their timezone, location, or working hours.
  Validated via Python's `zoneinfo` — typos are rejected with a hint.

Show the full session config:

```
python <scripts_dir>/config.py session
```

Show one section:

```
python <scripts_dir>/config.py session --section agents
python <scripts_dir>/config.py session --section tts
python <scripts_dir>/config.py session --section disabled_skills
python <scripts_dir>/config.py session --section user
```

Read a specific value:

```
python <scripts_dir>/config.py session --get-path agents.chat
python <scripts_dir>/config.py session --get-path user.timezone
python <scripts_dir>/config.py session --get-path tts.voice
```

Write a specific value:

```
python <scripts_dir>/config.py session --set-path agents.chat --set-value smart
python <scripts_dir>/config.py session --set-path tts.voice --set-value de_DE-thorsten-low
python <scripts_dir>/config.py session --set-path user.timezone --set-value Europe/Berlin
```

The `model`, `agent`, and `voice` shortcut commands continue to work and write into this file.

**Timezone validation:** writes to `user.timezone` are validated via `zoneinfo` —
an unknown name is rejected with a hint pointing at the IANA tz database list.
Take effect immediately: the next message will see the new local time in the
system prompt. Clear with `--set-value` omitted (no value → field removed).

## Disabled skills (per-session)

Prevent specific skills from being available in this session, even if globally configured.

List currently disabled skills:

```
python <scripts_dir>/config.py disabled-skills
```

Disable a skill:

```
python <scripts_dir>/config.py disabled-skills --add researcher
python <scripts_dir>/config.py disabled-skills --add browser
```

Re-enable a skill:

```
python <scripts_dir>/config.py disabled-skills --remove researcher
```

Changes take effect immediately for the current session.

## Coding backend (skill-creator)

The skill-creator writes and fixes skill scripts in-process via the
`coder` agent from `agents.coder` in `config.yaml` (falls back to
`agents.default` and then to the first defined model). To change the
model, set `agents.coder` with `config.py set agents.coder <model-key>` —
there is no CLI backend to install or switch.

Show the current configuration:

```bash
python <scripts_dir>/config.py coding
```

The response reports the resolved `coder` agent and model. Use the
`set` command to change the model key.

## Output

All commands return JSON. On success: `{"success": true, ...}`. On error: `{"success": false, "error": "..."}`.

After `set`, the response includes `"value_read_back"` — the value actually written to disk. Always compare it against what you intended to set and report any discrepancy to the user.

**Note:** Changes to config.yaml (via `set`) take effect after the next restart. The `model`, `agent`, `voice`, and `disabled-skills` commands take effect immediately without restart. Model and agent overrides are session-wide; only `private --thread` remains thread-local.