---
name: user-context
description: Inspect, export, update, reset, or manage CareerSignal durable profile, evidence, preferences, history, and privacy state. Use on `/CSsettings`, `/CShistory`, `/CSexport`, `/CSreset`, or explicit requests to change saved CareerSignal information.
---

# CareerSignal User Context

Use durable local state when execution is available.

Default state path:
1. `$CAREER_SIGNAL_HOME` if set;
2. `$PLUGIN_DATA/career-signal-os/` when the OpenAI host provides writable plugin data;
3. `$CLAUDE_PLUGIN_DATA/career-signal-os/` when Claude provides plugin data;
4. `$CODEX_HOME/state/plugins/career-signal-os/` if `$CODEX_HOME` exists;
5. `~/.career-signal-os/`.

State files:
- `profile.json`
- `evidence.json`
- `voice.json`
- `settings.json`
- `post_history.json`
- `topic_memory.json`
- `source_cache.json`
- `feedback.json`
- `onboarding.json`

Use:
`python3 skills/user-context/scripts/context_cli.py <command>`

Commands:
- `init`
- `status --json`
- `get-profile`
- `get-evidence`
- `get-settings`
- `get-history`
- `update-settings --file FILE`
- `export --out FILE`
- `reset`
- `import-onboarding --file FILE`
- `record-post --file FILE`
- `mark-published --id ID [--url URL]`
- `record-feedback --file FILE`

Reset must back up existing state before clearing it.

Do not store passwords, API keys, tokens, or raw confidential documents.

Manual mode must remain fully usable without connectors.

## Command-specific behavior

- `/CSsettings`: read current settings first, apply only requested changes with `update-settings`, and do not overwrite unrelated preferences.
- `/CSexport`: export the complete portable state package.
- `/CSreset`: explain that reset backs up current local state, then reset only after explicit confirmation.
- `/CShistory`: use `get-history` for saved post records when local state is available.

## Destructive-state protection

CLI-level reset and restore are confirmation-gated:

- `reset --confirm RESET`
- `restore-backup --path <backup> --confirm RESTORE`

Do not call either until the user explicitly confirms the destructive action.