feat(scheduler): add env_vars parameter for per-job environment overrides

[cagataycali]

Summary

Scheduled jobs can now inject arbitrary environment variables into the spawned DevDuck instance (e.g. TELEGRAM_BOT_TOKEN, API keys, feature flags).

Previously only model, tools, system_prompt, and max_tokens could be overridden per-job — anything else had to live in the global process env, which breaks when one scheduler runs jobs for different services.

Changes

• New env_vars: Optional[Dict[str, str]] parameter on scheduler(action='add', ...)
• Persisted to jobs.json → survives restarts
• Applied via the existing _EnvOverride context manager → global env stays clean after the job completes
• Docstring + status line updated

Example

scheduler(
    action="add",
    name="telegram-morning",
    schedule="0 9 * * *",
    prompt="send daily summary to chat",
    tools="telegram",
    env_vars={
        "TELEGRAM_BOT_TOKEN": "...",
        "TELEGRAM_CHAT_ID": "...",
    },
)

Each fired job will have those vars set for the duration of the spawned DevDuck session, then cleaned up.

Backwards compatibility

✅ Fully backwards compatible — env_vars defaults to None, existing jobs keep working. Old jobs.json entries without the field are handled via job.get("env_vars") or {}.

[github-actions]

✅ Autonomous Review - LGTM

Analyzed the changes and they look excellent:

Strengths

1. Clean API Design - env_vars: Optional[Dict[str, str]] = None parameter follows existing patterns
2. Proper Integration - Uses existing _EnvOverride context manager → no env pollution
3. Persistence - Stored in jobs.json → survives scheduler restarts
4. Backwards Compatible - Defaults to None, handles via job.get("env_vars") or {}
5. Well Documented - Clear examples in PR description + docstring updated

Implementation Quality

• ✅ Minimal diff (13 additions, 1 deletion) - surgical change
• ✅ Reuses existing infrastructure (_EnvOverride)
• ✅ Follows type hints consistently
• ✅ No breaking changes to existing jobs

Use Case Impact

Solves real problem: Multi-service schedulers can now inject credentials per-job without global env conflicts. Perfect for Telegram bots, API integrations, feature flags.

Recommendation

Merge-ready ✅

The PR is mergeable, implementation is clean, and backwards compatibility is maintained. No blocking issues found.

---

Reviewed autonomously by DevDuck GitHub Agent

[github-actions]

✅ Merged successfully!

The scheduler now supports per-job environment variables via the new env_vars parameter. This enables isolated configuration for different scheduled tasks without polluting the global environment.

Key capabilities unlocked:

• Telegram/Slack/WhatsApp bot tokens per scheduled job
• API keys and secrets scoped to specific tasks
• Feature flags and configuration overrides per job
• Full backwards compatibility with existing jobs

Example usage:

scheduler(
    action="add",
    name="telegram-morning",
    schedule="0 9 * * *",
    prompt="send daily summary",
    tools="telegram",
    env_vars={
        "TELEGRAM_BOT_TOKEN": "...",
        "TELEGRAM_CHAT_ID": "..."
    }
)

Changes: 1 file modified (+13/-1)
Commit: f5a27bd