Harvest Automation

Overview of Harvest Automation skill

What Harvest Automation does

Harvest Automation is a Claude Code skill for operating Harvest through natural language, using the Rube MCP server and Composio Harvest toolkit. It helps you log time, inspect and update time entries, manage projects, clients, and tasks, and pull billing or reporting data without switching from your terminal to the Harvest UI.

This Harvest Automation skill is best for freelancers, agencies, engineering teams, consultants, and operations users who already track work in Harvest and want faster, lower-friction workflow automation.

Best-fit jobs and users

Use Harvest Automation when the job is specific and account-backed: “log 3.5 hours to project X and task Y,” “list this week’s unsubmitted time,” “create a task for an existing project,” or “pull time entries for invoicing review.” It is less useful for vague planning tasks that do not require Harvest data or actions.

The strongest fit is repetitive administrative work where accuracy matters: time entry cleanup, project/task lookup, invoice preparation, and client/project maintenance.

Important adoption requirements

The repository contains a single SKILL.md; there are no helper scripts, rules folders, or bundled reference files. The key requirement is MCP access: the skill declares rube as required and expects you to add the Rube MCP server at https://rube.app/mcp, then authenticate your Harvest account.

Before installing, confirm your environment supports Claude Code skills and MCP servers, and that you are comfortable granting tool access to a live Harvest workspace.

How to Use Harvest Automation skill

Harvest Automation install and setup path

Install the skill from the source repository with your skill manager, for example:

npx skills add ComposioHQ/awesome-claude-skills --skill "Harvest Automation"

Then configure Claude Code to use the Rube MCP server:

1. Add the MCP server URL: https://rube.app/mcp
2. Follow the authentication link when Claude prompts for Harvest access
3. Test with a read-only request before creating or updating records

Start by reading composio-skills/harvest-automation/SKILL.md. Because this skill has no separate README.md, metadata.json, scripts, or references, SKILL.md is the authoritative source for supported workflows and tool names.

Inputs the skill needs to work reliably

Harvest Automation usage improves dramatically when you provide real Harvest identifiers and date details. For time entries, the skill may need:

• project_id
• task_id
• spent_date in YYYY-MM-DD
• hours for duration-based accounts
• started_time and ended_time for timestamp-based accounts
• notes describing the work

A weak prompt is: “Log my dev time for today.”

A stronger prompt is: “Use Harvest Automation to log 3.5 hours for 2026-01-15 on project 12345, task 67890, notes: Implemented OAuth callback handling and tests. If the project-task pairing is invalid, stop and tell me before creating anything.”

Practical workflow for safer automation

For create, update, and invoice-adjacent actions, use a two-step workflow:

1. Ask Harvest Automation to list or retrieve the relevant project, task, client, or time entry.
2. Confirm the exact record IDs before asking it to create or update anything.

This reduces mistakes caused by similar project names, stale task assignments, or ambiguous dates. For reporting and invoicing review, ask for a filtered list first, such as “list time entries for client X from Monday through Friday,” then request summaries only after the raw entries look correct.

Prompt patterns that improve output quality

Good prompts combine intent, constraints, and confirmation behavior:

• “List time entries for project 12345 from 2026-01-01 to 2026-01-31; group by task and flag entries missing notes.”
• “Create a Harvest project for client Acme, name Website Retainer Q1, but show me the proposed fields before calling the create tool.”
• “Update time entry 98765 to 2.25 hours and notes Bug triage and hotfix review; do not change the date, project, or task.”

These prompts help the agent choose the right Harvest tool instead of guessing from a broad request.

Harvest Automation skill FAQ

Is Harvest Automation for Workflow Automation?

Yes. Harvest Automation for Workflow Automation is most valuable when Harvest is part of a repeatable operational process: daily time logging, weekly timesheet review, project setup, task maintenance, and invoice preparation. It does not replace your billing judgment, but it can reduce manual navigation and copy-paste work.

How is this different from an ordinary prompt?

A normal prompt can tell you how to use Harvest. The Harvest Automation skill can call Harvest-related tools through the Rube MCP connection, provided your account is authenticated and the tool permissions are available. That makes it suitable for real actions such as listing, creating, retrieving, and updating Harvest records.

Is it beginner-friendly?

It is beginner-friendly if you already know your Harvest workspace structure. New Harvest users may struggle because project IDs, task IDs, client relationships, and date formats matter. Start with read-only listing requests, learn the exact IDs, and then move to create or update actions.

When should I not use it?

Do not use the skill when you cannot verify the target client, project, task, or time entry. Avoid broad commands like “fix all my timesheets” unless you first ask for a proposed change list. For sensitive billing changes, invoice creation, or large batch updates, require confirmation before execution.

How to Improve Harvest Automation skill

Improve Harvest Automation results with better context

The most useful context is operational, not verbose. Include the exact date range, workspace terms, record IDs, billing intent, and whether the request is read-only or write-enabled. If you do not know an ID, ask the skill to find candidates first rather than forcing it to infer from names.

Example: “Find active projects matching Acme and list their tasks with IDs. Do not create or update anything yet.”

Common failure modes to prevent

The main risk is ambiguity: similar project names, task names reused across projects, unclear dates like “last Friday,” or duration versus start/end-time account settings. Another common issue is asking for invoice-ready summaries before verifying that entries have accurate notes and task assignments.

Prevent this by making the skill show the records it plans to touch and by using absolute dates.

Iterate after the first output

After the first response, refine the workflow instead of starting over. If the skill lists entries, ask it to filter missing notes, group by project, or prepare a confirmation table. If a create request fails because a task is not assigned to a project, ask it to list valid tasks for that project before retrying.

A good iteration prompt is: “From the entries you just listed, show only records over 8 hours or missing notes, then suggest corrections without applying them.”

Repository improvement ideas for maintainers

The Harvest Automation skill would be stronger with example prompts for each supported action, read-only versus write-action guidance, and a small troubleshooting section for MCP authentication, invalid project-task pairings, and date formatting. A short safety checklist for destructive or billing-sensitive updates would also make installation decisions easier for teams evaluating the skill.