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:

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.