Build SOPs for Every Repeatable Task

1. Open https://claude.ai and start a **new chat** (stay in this same thread through Order 4). 2. Paste this prompt and fill in the brackets: "I am writing ONE standard operating procedure (SOP) for a repeatable task. Do NOT draft numbered steps yet. Task name: [...] Who performs it (usually me / role): [...] Start trigger — first observable event that means this workflow begins (e.g. contract signed, folder received, calendar event ends): [...] Stop trigger — observable handoff when this workflow is done (e.g. invoice sent, welcome email delivered, asset handed to client): [...] Rough frequency (daily / weekly / monthly / ad hoc): [...] Tools already involved (names): [...] Definition of done in one sentence: [...] Reply with ONLY: (a) one paragraph that mirrors my scope back, (b) up to 5 ambiguity flags, (c) ZERO process steps or SOP draft." 3. Read Claude's mirror. If start/stop triggers sound like attitudes ("whenever needed", "client is happy"), reply: "Rewrite triggers as concrete events (if this file exists / after this email / when this status changes) — still no steps." 4. When you're satisfied, reply once in your own words: "Scope locked" so you don't accidentally widen the task later.

1. In the **same Claude chat** as Order 1, paste a chronological bullet list from a real run: meeting notes, Loom outline, support ticket trail, or memory bullets (aim for **at least 8** bullets, in order). 2. If the list is thin, add this line and send: "Ask me up to 6 yes/no or multiple-choice questions to fill gaps. Then output a single bullet list labeled RAW_RUN — chronological only. No numbering scheme yet. No best-practices filler." 3. Scan RAW_RUN: every line should name a **concrete** action, check, file, tool, or recipient. If anything is vague, ask Claude to rewrite just those bullets with a noun + verb.

1. In the **same chat**, paste: "Using RAW_RUN from above, produce SOP v1 with these **exact headings** in order: Purpose (max 2 lines) When this runs (copy the agreed start trigger) Inputs (bullets) Outputs / artifacts (bullets) Steps — numbered 1..N; under each step use a/b/c substeps where needed; after each numbered step add a line **Done when:** that states something checkable (file exists, field filled, message sent, status changed) If things go wrong — 3–5 if/then rows tied to MY tools and inputs above Time / owner — default owner me, give a rough time range Rules: cap at **12** numbered steps. If more work exists, add a Phase 2 note under Steps instead of inflating v1. Ban vague verbs without an object. Ban these phrases entirely: ensure quality, communicate clearly, stay professional, handle with care, be proactive. Every numbered step must mention a tool, file, template, or recipient by name." 2. Read the Steps section aloud once. If any **Done when** is not observable, ask Claude to tighten that step only. 3. Keep this block — you'll paste it into your doc in Order 5.

1. In the **same chat**, paste: "Red-team the SOP v1 above. Output four sections: (A) 3 missing-if branches **specific** to this workflow (e.g. missing asset, client ghosting, wrong template version) — each branch names an input from the Inputs section (B) 2 handoff risks — who owns it next and what artifact they must receive (C) 1 step that is too brittle (too many decisions in one line) and should be split (D) For each item in A–C, give a **short patch** — text to paste under If things go wrong or as a substep — do NOT rewrite the whole SOP" 2. Merge patches into your v1 mentally; if a patch is good, ask Claude for a single consolidated **If things go wrong** block you can paste over the old one. 3. Name your scariest real-world case out loud — confirm you can find it reflected in the doc.

1. In **Google Docs** (or your chosen Notion page — same idea), create a new doc titled: `[SOP] <Task name> — v1 — <YYYY-MM-DD>`. 2. Paste the full SOP v1 from Order 3, with patches from Order 4 merged in. 3. At the very top, add a short callout box or bold lines: **Owner:** you (or role) | **Last reviewed:** today's date | **Review cadence:** after 3 real runs or 30 days, whichever comes first. 4. Optional — open Claude once more in a **new** chat and paste: "Here is my finalized SOP. Give (a) a 3-line changelog entry for v1, (b) a 5-bullet read-aloud I can use to train myself in 60 seconds." Paste the changelog under a **Changelog** heading at the bottom. 5. Grab the shareable URL or confirm the doc lives in your standard folder so you can find it next week.