Cron Retry

Add this to your `HEARTBEAT.md`:

```markdown
## Cron Recovery Check
Check for cron jobs with lastStatus: "error". If the error matches network patterns (connection error, sendMessage failed, fetch failed, ETIMEDOUT, ECONNREFUSED), retry the job using cron tool with action: "run" and the job ID. Report what was recovered.
```

That's it. On each heartbeat, failed network jobs get retried automatically.

## How It Works

1. On heartbeat, check all cron jobs via `cron list`
2. Filter for jobs where `lastStatus = "error"` and `enabled = true`
3. Check if `lastError` matches network-related patterns
4. Re-run eligible jobs via `cron run`
5. Report results

## Network Error Patterns (Retryable)

These errors indicate transient network issues worth retrying:

- `Network request.*failed`
- `Connection error`
- `ECONNREFUSED`
- `ETIMEDOUT`
- `ENOTFOUND`
- `sendMessage.*failed`
- `fetch failed`
- `socket hang up`

## What Gets Retried vs Skipped

**Retried:**
- Network timeouts
- Connection refused
- Message send failures
- DNS lookup failures

**Skipped (not retried):**
- Logic errors (bad config, missing data)
- Auth failures
- Disabled jobs
- Jobs that just ran successfully

## Manual Recovery Check

To check and retry failed jobs manually:

```bash
# List all jobs and their status
clawdbot cron list

# Find failed jobs
clawdbot cron list | jq '.jobs[] | select(.state.lastStatus == "error") | {name, error: .state.lastError}'

# Retry a specific job
clawdbot cron run --id <JOB_ID>
```

## Agent Implementation

When implementing the heartbeat check:

```
1. Call cron tool with action: "list"
2. For each job in response.jobs:
   - Skip if job.enabled !== true
   - Skip if job.state.lastStatus !== "error"
   - Check if job.state.lastError matches network patterns
   - If retryable: call cron tool with action: "run", jobId: job.id
3. Report: "Recovered X jobs" or "No failed jobs to recover"
```

## Example Scenario

1. **7:00 PM** — Evening briefing cron fires
2. **Network hiccup** — Telegram send fails
3. **Job marked** `lastStatus: "error"`, `lastError: "Network request for 'sendMessage' failed!"`
4. **7:15 PM** — Connection restored, heartbeat runs
5. **Skill detects** the failed job, sees it's a network error
6. **Retries** the job → briefing delivered
7. **Reports**: "Recovered 1 job: evening-wrap-briefing"

## Safety

- Only retries transient network errors
- Respects job enabled state
- Won't create retry loops (checks lastRunAtMs)
- Reports all recovery attempts