Run Claude on a cronjob for morning briefings and email tasks
User built a Claude-in-a-loop with cron that produces morning briefings and podcasts and handles tasks emailed or texted to it. Task: SyneRyder implemented a custom 'Claude-in-a-loop' using a cron job to execute `claude -p` during early morning hours. The system accesses personal context and long-term memory to generate morning briefings and personalized podcasts. Additionally, incoming emails, voice texts, or Telegram messages trigger sessions where Claude processes forwarded tasks and replies. Human involvement is limited to initial setup and reviewing outputs, as the source does not detail specific manual interventions during execution.; tools: Claude. Human role: Source evidence is unavailable regarding specific human roles or manual interventions in this workflow.; limitations: Effectiveness relies heavily on providing extensive personal context and accessible long-term memory; a bare scheduled prompt yields poor results. A commenter warns that using `claude -p` with subscription accounts may violate terms of service, risking account enforcement due to usage patterns. Integration requires secure handling of personal data and reliable access to communication channels like email and Telegram..
Tools
Published on Agent Work Atlas:
Overview
User built a Claude-in-a-loop with cron that produces morning briefings and podcasts and handles tasks emailed or texted to it.
Workflow
SyneRyder implemented a custom 'Claude-in-a-loop' using a cron job to execute `claude -p` during early morning hours. The system accesses personal context and long-term memory to generate morning briefings and personalized podcasts. Additionally, incoming emails, voice texts, or Telegram messages trigger sessions where Claude processes forwarded tasks and replies. Human involvement is limited to initial setup and reviewing outputs, as the source does not detail specific manual interventions during execution.
Reported result
The author reports that with sufficient context, the agent independently surfaces useful and sometimes surprising suggestions overnight without explicit prompting. No quantitative metrics or specific success rates are provided in the source. A measurable acceptance check would be to track the number of actionable insights or completed tasks derived from the morning briefings over a two-week period.
Human role
Source evidence is unavailable regarding specific human roles or manual interventions in this workflow.
Limitations
Effectiveness relies heavily on providing extensive personal context and accessible long-term memory; a bare scheduled prompt yields poor results. A commenter warns that using `claude -p` with subscription accounts may violate terms of service, risking account enforcement due to usage patterns. Integration requires secure handling of personal data and reliable access to communication channels like email and Telegram.
Try it
Act as an automation engineer. I want to set up a personal AI assistant using Claude. 1. **Context Setup**: Create a file `[replaceable_personal_context.txt]` containing my daily schedule, key projects, and preferences. 2. **Cron Job**: Write a script that runs `claude -p` with the context file and a prompt to generate a morning briefing. Schedule this for `[replaceable_time, e.g., 06:00]`. 3. **Email Integration**: Configure a rule to forward emails from `[replaceable_sender_domain]` to a script that triggers a Claude session to summarize and suggest replies. 4. **Dry Run**: Execute the cron job manually once. Review the output for accuracy and relevance. 5. **Human Confirmation**: Do not enable the automated email forwarding until I approve the dry run briefing. **Deliverables**: A bash script for the cron job, a Python script for email handling, and a sample morning briefing. **Success Metric**: The briefing must include at least 3 actionable items relevant to my current projects. **Error Handling**: If the context file is missing, log an error and skip execution. Ask clarifying questions if my specific email provider or OS is not `[replaceable_os]` or `[replaceable_email_provider]`.
Source reports are not independently verified. Publication date is the catalog publication date, not the original article date.