Automate freelance admin with a morning briefing agent
Posted on Apr 23 β’ Edited on Apr 23. Task: Posted on Apr 23 β’ Edited on Apr 23 OpenClaw Challenge Submission π¦ No Python. No Node. Three skill files, one Lobster workflow, and my consultancy now runs on autopilot. What I Built A fully autonomous freelance operations agent I call Claw-Ops that handles the three biggest time drains in running a dev consultancy: Morning dev briefing β PR triage from GitHub, action items pulled from Slack, delivered to Telegram at 8AM Client status drafting β every Friday, reads merged PRs + closed Linear tickets, writes the weekly update email, sends me the draft to approve with one tap Inbox zero on follow-ups β detects unanswered client emails older than 48 hours, drafts a polite follow-up, asks me on Telegram whether to send it Before this: ~9 hours a week on context-switching, manual status writing, and email chasing. After: I review a Telegram message in the morning and say yes or no. The Stack OpenClaw (gateway running on a Mac Mini, always on) Claude via Anthropic API (the brain) Telegram (my interface β DMs and one-tap approvals) Three custom skills + one Lobster workflow (no external services beyond what I already used) Zero new subscriptions. Zero new dashboards. Everything surfaces in the app I already have open: Telegram. Skill 1: pr-morning-digest Skill 2: slack-action-extractor Skill 3: client-status-drafter The Lobster Workflow That Chains Everything The Setup (Reproducible in Under an Hour) Step 1: Install OpenClaw Step 2: Create the skill files Step 3: Create the Lobster workflow Step 4: Connect Telegram Step 5: Start the gateway and test Total time from zero to running: 47 minutes. What Actually Happens in Production Monday 8:02 AM β Telegram: Friday 4:03 PM β Telegram: I reply "send." Done. The Result Honest Notes What worked immediately: The PR digest and Telegram integration were flawless out of the box. What needed iteration: The client status email tone took two rounds of refinement in agents.md. Adding "Write like a human who has a strong working relationship with this client. Warm, direct, no corporate filler." to the agent personality fixed it completely. Security note: I run with --sandbox disabled because I need full shell access for the gh and linear CLIs. This is a deliberate choice with understood risk. Start sandboxed, then selectively enable only the tools each skill genuinely needs. Fork It All three skill files and the Lobster workflow are in this GitHub Repository: https://github.com/FarooqShabbir/claw-ops The entire point of the skills format is that you're not copying code β you're copying intent. Edit the instructions to match your stack, and the agent adapts automatically. Submitted for the OpenClaw Writing Challenge on DEV β OpenClaw in Action prompt. Top comments (0) For further actions, you may consider blocking this person and/or reporting abuse; tools: OpenClaw. Human role: After: I review a Telegram message in the morning and say yes or no.; limitations: Security risk: The author disabled OpenClaw's sandbox to allow full shell access for gh and linear CLIs, a deliberate choice with understood risks. Requires existing GitHub, Slack, Linear, and Gmail accounts. Integration uncertainty exists for tone matching, which needed manual refinement. Proposed advice: Start sandboxed and selectively enable tools..
Tools
Published on Agent Work Atlas:
Overview
Posted on Apr 23 β’ Edited on Apr 23
Workflow
Posted on Apr 23 β’ Edited on Apr 23 OpenClaw Challenge Submission π¦ No Python. No Node. Three skill files, one Lobster workflow, and my consultancy now runs on autopilot. What I Built A fully autonomous freelance operations agent I call Claw-Ops that handles the three biggest time drains in running a dev consultancy: Morning dev briefing β PR triage from GitHub, action items pulled from Slack, delivered to Telegram at 8AM Client status drafting β every Friday, reads merged PRs + closed Linear tickets, writes the weekly update email, sends me the draft to approve with one tap Inbox zero on follow-ups β detects unanswered client emails older than 48 hours, drafts a polite follow-up, asks me on Telegram whether to send it Before this: ~9 hours a week on context-switching, manual status writing, and email chasing. After: I review a Telegram message in the morning and say yes or no. The Stack OpenClaw (gateway running on a Mac Mini, always on) Claude via Anthropic API (the brain) Telegram (my interface β DMs and one-tap approvals) Three custom skills + one Lobster workflow (no external services beyond what I already used) Zero new subscriptions. Zero new dashboards. Everything surfaces in the app I already have open: Telegram. Skill 1: pr-morning-digest Skill 2: slack-action-extractor Skill 3: client-status-drafter The Lobster Workflow That Chains Everything The Setup (Reproducible in Under an Hour) Step 1: Install OpenClaw Step 2: Create the skill files Step 3: Create the Lobster workflow Step 4: Connect Telegram Step 5: Start the gateway and test Total time from zero to running: 47 minutes. What Actually Happens in Production Monday 8:02 AM β Telegram: Friday 4:03 PM β Telegram: I reply "send." Done. The Result Honest Notes What worked immediately: The PR digest and Telegram integration were flawless out of the box. What needed iteration: The client status email tone took two rounds of refinement in agents.md. Adding "Write like a human who has a strong working relationship with this client. Warm, direct, no corporate filler." to the agent personality fixed it completely. Security note: I run with --sandbox disabled because I need full shell access for the gh and linear CLIs. This is a deliberate choice with understood risk. Start sandboxed, then selectively enable only the tools each skill genuinely needs. Fork It All three skill files and the Lobster workflow are in this GitHub Repository: https://github.com/FarooqShabbir/claw-ops The entire point of the skills format is that you're not copying code β you're copying intent. Edit the instructions to match your stack, and the agent adapts automatically. Submitted for the OpenClaw Writing Challenge on DEV β OpenClaw in Action prompt. Top comments (0) For further actions, you may consider blocking this person and/or reporting abuse
Reported result
The author claims time spent on admin tasks dropped from ~9 hours/week to ~20 minutes/week. PR digests and Telegram integration worked immediately, but client email tone required two iterations in agent personality settings. No independent verification of these metrics is provided; success is self-reported.
Human role
After: I review a Telegram message in the morning and say yes or no.
Limitations
Security risk: The author disabled OpenClaw's sandbox to allow full shell access for gh and linear CLIs, a deliberate choice with understood risks. Requires existing GitHub, Slack, Linear, and Gmail accounts. Integration uncertainty exists for tone matching, which needed manual refinement. Proposed advice: Start sandboxed and selectively enable tools.
Try it
I want to set up an OpenClaw workflow named 'claw-ops' to automate my freelance admin tasks. Inputs: [GitHub Repo URL], [Slack Channel Name], [Linear Project ID], [Telegram Bot Token], [Anthropic API Key]. Steps: 1. Install OpenClaw and configure the Anthropic API key. 2. Create three skill files: 'pr-morning-digest' (checks GitHub PRs >24h old or failing CI), 'slack-action-extractor' (finds tasks in Slack mentioning me), and 'client-status-drafter' (compiles merged PRs and closed Linear tickets into a draft email). 3. Create a Lobster workflow 'claw-ops-daily' scheduled for 8 AM Mon-Fri that runs the first two skills and sends a Telegram briefing. Add a Friday-only step for the status drafter. 4. Connect Telegram and start the gateway. Deliverables: A working Telegram bot that sends morning briefings and Friday email drafts. Quality Check: Run a dry test on a sample day. Verify the PR list matches GitHub status and the Slack extraction finds at least one known task. Safeguards: Do NOT send any emails or messages automatically. The agent must only draft and ask for approval. I will manually review the first three outputs before enabling any 'send' actions. If inputs are missing, ask me for the specific credentials or IDs before proceeding.
Source reports are not independently verified. Publication date is the catalog publication date, not the original article date.