Turn Repeated Work Into Cron Jobs, Skills, and Agentic Workflows

Recurring automation in Hermes has three layers. A cron job determines when work runs. A skill defines the reusable procedure. An agentic workflow is the full loop in which Hermes gathers information, reasons over it, creates an output, reviews the result, and delivers or acts on it.

These layers should be built in that order of maturity: first prove the task manually, then refine the procedure, then package it as a skill, and finally schedule it.

A cron job runs on a schedule. It can execute every morning, every Friday, during trading hours, or at several points throughout the day. The schedule starts the work without requiring a fresh user message.

A skill is a packaged playbook. It contains the method for performing a repeated task while allowing the inputs to change. The research in a morning briefing changes every day, but the expected sections and quality standard remain stable. The skill preserves that standard.

An agentic workflow is broader than one prompt. It can include several steps and decisions. A typical loop is:

1. Retrieve information from connected applications.
2. Filter and prioritize what matters.
3. Create the requested output.
4. Review the output against the intended format.
5. Correct weak or missing elements.
6. Deliver the result through the selected channel.
7. Update the project board and memory.
8. Repeat on the next schedule.

The morning briefing demonstrates all three layers. Hermes uses calendar access, Gmail, Slack, documents, and research tools. It identifies meetings, urgent messages, team activity, follow-up tasks, preparation needs, and relevant topics. It then sends a short practical briefing to Telegram at a fixed time.

The highest-leverage use of an agent is not asking it to perform the same useful task every day. It is teaching the task once, validating the result, and making the agent perform it consistently without another request.

Cron jobs convert availability into action. Skills convert preference into repeatability. Agentic workflows connect multiple applications and reasoning steps into an operating process.

The combination can support client updates, competitor monitoring, trend research, invoice follow-up, daily planning, overnight work, and many other recurring responsibilities.

Build a morning briefing as the model workflow.

First, connect the required applications. Hermes should be able to read the calendar, email, team chat, and relevant documents.

Second, request one example briefing. Specify the required contents:

• Today's calendar and upcoming meetings.
• Preparation or research needed before meetings.
• Urgent or important email from the previous 24 hours.
• Drafts and links where action is required.
• A short summary of important team conversations.
• Relevant documents and follow-up tasks.
• Selected trends or topics worth attention.
• A concise format that becomes even shorter when little is important.

Third, review the example. Remove unnecessary sections, adjust the language, change the ordering, and require links or drafts where they improve usefulness. The first result is a prototype, not the final workflow.

Fourth, schedule it for the desired time through a cron job. Confirm the next run appears in Hermes and, when Multica is connected, confirm that the same recurring job appears as an autopilot.

Fifth, ask Hermes to convert the refined briefing process into a skill. Save the skill in Hermes and Multica so the same playbook can be reused by other profiles or team members.

Use the same pattern for other recurring work. For a weekly client report, retrieve project updates, summarize progress, identify risks, create the report, and send it. For competitor monitoring, scrape selected sites, compare price or offer changes, alert the user, and update the project. For unpaid invoices, inspect the connected billing system, draft follow-ups, and route them for review or sending.

A recurring workflow multiplies both good and bad design. If the procedure is weak, the same weak output will reappear on schedule. If a connector fails, the briefing may be incomplete. If the prompt is too broad, the output can become long and noisy.

Scheduled jobs also consume model usage without a fresh decision each time. Inspect how often they run, what they retrieve, and whether the value justifies the work. Keep approval in the loop for external messages or financial actions until the workflow has been validated.

Prove the task, refine the result, package the procedure as a skill, and schedule it with a cron job. That sequence turns a useful interaction into a durable agentic workflow.