Scheduled-agents review by Anthropic’s guide
‘The last 24 hours’ in a scheduled task skips or repeats work: fixed windows, silent read failures and risky posting flagged.
384 saves on X, as of Posted
Prompt
Read this guide: https://claude.dev/blog/building-effective-agent-automations/
Then review all my scheduled agent tasks against it: cron, launchd, scheduled GitHub Actions workflows, and scripts that call claude -p or codex exec. If any run on a server or in the cloud where this machine can't see them, tell me which config to paste in. Find three kinds of problems:
1. Tasks that read a fixed time window, like 'the last 24 hours' or 'since yesterday'.
2. Tasks that report 'nothing new' even when a read fails: if a token has expired or an MCP server won't connect, the report doesn't show it.
3. Tasks that can drop a post, post twice, overspend or have too much access: a post marked as sent before it's confirmed, no duplicate check, no spending cap per run, or a token that can write to a source the task only reads.
For each issue, quote the line, explain what could go wrong, and suggest the smallest change.
Show me the recommendations first. Don't change anything yet.Next Send it to Opus 5.5 in Claude Code on the machine that runs your scheduled tasks.
Anthropic just published an official full guide to scheduled agents. It calls out a common mistake: telling the agent to read “the last 24 hours.” You can ask Opus 5.5 to use it to review the scheduled tasks you have running.
If a run starts late, it misses a stretch. If it starts early, it repeats items. The guide has the agent keep a bookmark for each source, and the next run picks up from it. If your prompts say “since yesterday” or “the last 7 days,” change those too.
Send this prompt to Opus 5.5.
Share this tip
