# 6 daily workflows Skills are the parts. Workflows are the machine. Each workflow below chains skills into a real job, with the exact words to type, the time it takes, and what "done" looks like. Every workflow works in claude.ai (with the skills uploaded) and in Claude Code. If you use ChatGPT or Gemini, swap each skill for its prompt in `PROMPTS.md`. How to call a skill: - **Claude.ai**: just ask in plain words ("triage my inbox"). Claude picks the skill from its description. To force it, name it: "Use the inbox-triage skill." - **Claude Code**: same, or type `/inbox-triage` directly. --- ## Workflow 1: Morning inbox (15 minutes instead of 60+) **Skills**: `inbox-triage` → `voice-writer` → `meeting-to-actions` (optional) **Where**: a Project called "Daily ops" with your personal-assistant instructions. 1. **(1 min)** Open a new chat in "Daily ops". Type: ``` Triage my inbox: unread since yesterday 6pm. Use my inbox rules. ``` With the Gmail connector on, Claude reads them. Without it, paste the messages (sender, subject, body). 2. **(5 min)** Read the "Reply now" drafts. Fill the `[BRACKETS]`. Edit only what's wrong. Send them yourself. 3. **(3 min)** For each "Deep work" item, block time in your calendar now. Don't answer them in the morning sprint. 4. **(2 min)** Copy the "Delegate" handoff notes into Slack/WhatsApp. 5. **(1 min)** Answer "Anything I sorted wrong?" Every correction becomes a rule. After 2 weeks, the triage is almost always right. 6. **Optional**: if yesterday had a call, paste the transcript: "Run meeting-to-actions on this." Send the recap before 10am. **Done when**: inbox has only Deep work left, each with a time block. **Usage tip**: start a new chat every morning. Yesterday's inbox is dead weight. --- ## Workflow 2: Write a post in your voice (20 minutes, once calibrated) **Skills**: `voice-writer` → `content-repurposer` (optional) **One-time setup (15 min)**: run voice-writer calibration with 3-5 real samples, save `voice-profile.md` (see the skill). 1. **(3 min)** Brain-dump the idea. Voice notes work: record 2 minutes on your phone, paste the transcript. Messy is good. Real details are what make it yours. 2. **(1 min)** Type: ``` Write a LinkedIn post in my voice from this. Audience: [WHO]. The one point: [POINT]. 150-200 words. ``` 3. **(5 min)** Read it out loud. Mark anything you'd never say. Reply with specific notes: "Line 3 is too polished. Cut the last paragraph. Add the number from my notes." 4. **(2 min)** Final pass yourself. Change at least one line by hand. You're the author. 5. **(optional, 5 min)** "Now repurpose this into an X post, a Threads post and a short video script." Done: 4 posts from one idea. 6. **(1 min)** If you made the same edit twice ("you always make my intro too long"), tell Claude: "Add that to my voice profile." **Done when**: you'd be comfortable if a friend asked "did you write this?" --- ## Workflow 3: Client proposal (45 minutes instead of half a day) **Skills**: `meeting-to-actions` → `research-brief` → `anti-generic-strategy` → `voice-writer` **Where**: your "[Business] HQ" Project (business-operator template) with your price list and 1-2 past proposals in Project knowledge. 1. **(5 min)** After the discovery call, paste the transcript: ``` Run meeting-to-actions. Then list everything the client said about their problem, budget, timeline and decision process, with quotes. ``` 2. **(10 min, optional)** If you need context on their market or competitors: ``` Quick research brief: [CLIENT]'s top 3 competitors and how they handle [PROBLEM]. Decision: what to emphasize in my proposal. ``` 3. **(15 min)** Draft: ``` Draft a proposal for [CLIENT] using anti-generic-strategy. Use their words from the call, my price list, and my past proposal as the format. Structure: their situation (in their words), the cost of not fixing it, what we'll do (3 phases max), timeline, price with 2 options, what we won't do, next step with a date. ``` Answer the questions it asks. Leave `[UNKNOWN]` where you truly don't know. 4. **(10 min)** Read the specificity report. Fix every `[assumption]` you can confirm. Make sure the price and dates are right. These are never Claude's call. 5. **(5 min)** "Write the cover email in my voice. Under 120 words. One clear next step." **Done when**: every section mentions something only this client said or has. Run the swap test: if you could send it to another client unchanged, it's not done. --- ## Workflow 4: Weekly review (20 minutes, Friday or Sunday) **Skills**: `weekly-review` → `decision-memo` (if a big choice comes up) **Where**: a Project called "Weekly reviews". Save each review into Project knowledge so next week's review can compare. 1. **(1 min)** New chat in the Project: "Weekly review." 2. **(7 min)** Answer the 6 questions. Bullets. Don't polish. 3. **(5 min)** Read the scorecard and patterns. If it says something slipped 3 weeks in a row, believe it. 4. **(3 min)** Push back on the top 3 if they're wrong. Ask: "Make #2 smaller so it fits in one morning." 5. **(2 min)** Put the calendar blocks in your calendar now. 6. **(2 min)** Save the review: download it as a file and add it to the Project knowledge (Claude Code: it saves `reviews/YYYY-MM-DD.md` for you). 7. **If a hard choice surfaced** ("should I drop this client?"): start a separate chat and run decision-memo. Don't mix it into the review. **Done when**: next week's top 3 are in your calendar with a day each. --- ## Workflow 5: Research → decision (1 hour instead of a week of tabs) **Skills**: `prompt-sharpener` → `research-brief` → `decision-memo` Example: "Should I switch from [TOOL A] to [TOOL B]?" / "Should I raise my prices?" / "Which market do I launch in first?" 1. **(5 min)** Sharpen the question: ``` Sharpen this into a research question tied to a decision: "[YOUR MESSY QUESTION]" ``` 2. **(15-25 min)** Run the research, with web search on: ``` Research brief, deep: [SHARPENED QUESTION]. Decision: [DECISION]. ``` 3. **(5 min)** Read "What's uncertain". If one unknown would flip the decision, get that one fact (a call, a trial, a quick test) before going further. 4. **(15 min)** New chat (keeps the research noise out). Paste only the brief's bottom line, findings and table: ``` Decision memo using this research: [PASTE]. My criteria ranked: [1, 2, 3]. ``` 5. **(5 min)** Write the kill criterion in your calendar on the date it names. **Done when**: you've decided, you know what would make you change your mind, and there's a date to check. **Usage tip**: research is the heaviest workflow (web search + long sources). Do it on the right model, in its own chat, once. --- ## Workflow 6: Ship a small feature with Claude Code (1-2 hours) **Skills**: `prompt-sharpener` → (plan mode) → `code-reviewer` **Setup**: a CLAUDE.md in your repo (start with `/init`, then merge the coding-project template). 1. **(5 min)** Start clean: `cd your-project && claude`. If you were working on something else: `/clear`. 2. **(5 min)** Write the ask like a ticket. Use the sharpener if it's fuzzy: ``` /prompt-sharpener Add a "duplicate invoice" button to the invoice page ``` A good ask has: what the user sees, where in the code, edge cases, how to verify ("the e2e test for invoices passes and a new test covers duplication"). 3. **(10 min)** Press **Shift+Tab** to switch to plan mode. Paste the ask. Read the plan. Push back now (it's cheap): "Don't add a new API route, reuse the existing createInvoice action." 4. **(20-60 min)** Approve. Let it work. If it goes the wrong way, press **Esc** right away and correct it. Use `/rewind` to go back to a checkpoint if needed. 5. **(5 min)** Make it verify: "Run typecheck, lint and the invoice tests. Show me the results." Don't accept "it should work". 6. **(10 min)** Review: ``` /code-reviewer ``` Fix every 🔴 Blocker. Decide on the 🟠s. 7. **(5 min)** Look at it yourself in the browser. Then: "Commit with a clear message." Push when you're happy. 8. **(1 min)** If Claude got something wrong that it will get wrong again, add one line to "Known gotchas" in CLAUDE.md. **Done when**: tests pass, the review has no blockers, and you've seen the feature work with your own eyes. **Usage tip**: one feature per session. `/clear` before the next one. --- ## Make your own workflow Pattern behind all six: **capture → structure → draft → you decide → save what you learned.** 1. Pick a task you do every week that takes 30+ minutes. 2. Do it once with Claude in one chat, narrating what you want. 3. At the end: "Turn what we just did into a skill. Name, description with when to use it, numbered steps, output format, rules." (Or run `prompt-sharpener` and ask it to make a skill.) 4. Save it in `~/.claude/skills/` (Claude Code) or zip and upload it (claude.ai). 5. Next week, run it and fix one thing. After 3 runs it's yours.