--- name: open-claw description: Run AutoPod blogs and podcasts on full daily autopilot. Use when the user wants to operate one or many AutoPod blogs hands-free — propose 3 fresh article topics per blog each day, accept the user's number-only choices (or skips), kick off generation, then file a 2-hour follow-up review of what shipped or broke. version: 1.0.0 --- # Open Claw — AutoPod Daily Autopilot Open Claw is the operator skill for [AutoPod](https://autopod.co). It turns Claude into your daily content manager: every day you get 3 suggested article titles per blog, you reply with numbers (or "skip"), Claude triggers generation, and 2 hours later it tells you what shipped — and what, if anything, didn't. ## When to use Use this skill when the user: - Wants their AutoPod blog(s) on full autopilot. - Asks you to "run today's blogs", "do today's autopod", "what should I write today on \", or anything similar. - Asks you to check on results of articles you triggered earlier today. - Triggers you via a daily scheduled task that mentions Open Claw / AutoPod. Do NOT use this skill for general writing assistance unrelated to AutoPod — it only operates against the AutoPod REST API at `https://autopod.co/api/v1`. ## One-time setup the user has to do Before you can run the daily flow, the user needs to give you an API key. Store it in their environment as `AUTOPOD_API_KEY`: - **Tenant key** (`apod_…`) — manages exactly one blog. Get it from Dashboard → API tab → "Tenant API Key" inside that blog. - **Account key** (`apoa_…`) — manages every blog the user owns from one Claude. Recommended for owners with multiple blogs. Get it from Dashboard → API tab → "Account API Key (multi-blog)". Both require an active subscription (Starter / Growth / Scale all work). If the user hasn't provided a key yet, ask once for the key and the model they prefer (account or tenant). Don't keep asking — just remind them on the next turn if it's still missing. ## Recommended schedule (the user sets this up in Claude) Open Claw is meant to be triggered by a Claude scheduled task — Claude itself isn't always-on. Suggest these to the user during setup: 1. **Daily morning task** (e.g. 09:00 user-local): "Run today's Open Claw suggestions for AutoPod." → Claude executes the **Daily flow** below. 2. **Follow-up task scheduled by Claude** at "now + 2 hours" after the user confirms picks → Claude executes the **2-hour review flow**. If your runtime supports scheduling new tasks programmatically, schedule the 2-hour review yourself the moment the user finishes picking. Otherwise, at the end of the daily flow, instruct the user to set the 2-hour reminder. --- ## Daily flow Run these steps in order. Stop and report immediately on the first hard error. ### 1. List blogs the key can manage ```bash curl -sS -X POST https://autopod.co/api/v1 \ -H "Authorization: Bearer $AUTOPOD_API_KEY" \ -H "Content-Type: application/json" \ -d '{"action":"listBlogs"}' ``` Expected response: ```json { "success": true, "principal": { "kind": "account" }, "blogs": [ { "id": "tenantA", "name": "Visual Field Test", "slug": "vft", "plan": "scale", "status": "active", "niche": "eye health", "articlesUsedThisMonth": 12, "articlesLimitThisMonth": 40, "articlesRemainingThisMonth": 28, "apiAccessEnabled": true } ] } ``` If `success` is false or the call returns 401/403, stop and tell the user exactly what the API said. Do not retry blindly. Skip any blog where `apiAccessEnabled` is `false` or `articlesRemainingThisMonth <= 0` — explain to the user that those are quota-blocked or inactive. ### 2. For each remaining blog, prepare 3 fresh suggestions For each blog (parallelizable), do the following. **2a.** Pull the most recent published titles for novelty checks: ```bash curl -sS -X POST https://autopod.co/api/v1 \ -H "Authorization: Bearer $AUTOPOD_API_KEY" \ -H "Content-Type: application/json" \ -d '{"action":"listRecentArticles","tenantId":"","limit":20}' ``` Note: `tenantId` is required only when using an account (`apoa_…`) key. Tenant keys ignore it. **2b.** Look at any pending suggestions already queued for today: ```bash curl -sS -X POST https://autopod.co/api/v1 \ -H "Authorization: Bearer $AUTOPOD_API_KEY" \ -H "Content-Type: application/json" \ -d '{"action":"listSuggestions","tenantId":"","limit":10}' ``` **2c.** Decide whether the existing pending suggestions are diverse enough versus the last 20 published titles. If you can pick 3 strong, distinct suggestions from what's already pending, do that. Otherwise call: ```bash curl -sS -X POST https://autopod.co/api/v1 \ -H "Authorization: Bearer $AUTOPOD_API_KEY" \ -H "Content-Type: application/json" \ -d '{"action":"generateSuggestions","tenantId":"","count":5}' ``` The server already deduplicates against the last 50 published titles + last 50 pending suggestions, but YOU are the final judge. Reject any suggestion that is essentially a rewording of a recent title and pull another from the returned list (or generate again — at most once). End up with exactly 3 suggestions per blog: keep their `id`, `title`, `prompt`. ### 3. Present the menu to the user Show one block per blog. Use this exact shape so the user can answer with just numbers: ``` Visual Field Test (28 articles left this month) 1. How LASIK Affects Long-Term Night Vision 2. The Real Difference Between Glaucoma Tests at the Optometrist 3. Why Reading on a Phone Makes Eyes Feel "Stuck" Cooking With Less Salt (3 articles left this month) 1. ... 2. ... 3. ... ``` End with one line: > Reply with the number you want for each blog (e.g. `vft: 2, salt: skip`), > or just say `skip all` / `1, 1, 2` if blogs are listed in order. Accept any of these answer styles: - `1, 2, skip` — positional, in the order you listed blogs. - `vft: 2, salt: 1` — named, using slug or short name. - `2` — single blog only. - `skip ` or `skip all` — skip explicitly. - A natural-language answer; map it carefully and confirm before acting. If the user's answer is ambiguous, ask one clarifying question — don't act. ### 4. Trigger autopilot for each picked blog For each `(blog, suggestionId)` the user picked: ```bash curl -sS -X POST https://autopod.co/api/v1 \ -H "Authorization: Bearer $AUTOPOD_API_KEY" \ -H "Content-Type: application/json" \ -d '{"action":"runAutoPilot","tenantId":"","suggestionId":""}' ``` Capture the returned `jobId` for each blog. Keep them in your scratchpad as: ``` [ { "blogId": "tenantA", "blogName": "Visual Field Test", "title": "How LASIK …", "jobId": "abc123" }, ... ] ``` If `runAutoPilot` returns 429, the blog is out of quota — surface that and move on. Don't retry. ### 5. Confirm + schedule the 2-hour review Reply to the user with a short summary: ``` Triggered 2 articles, skipped 1. ✓ Visual Field Test → "How LASIK …" ✓ Cooking With Less Salt → "..." - Sourdough For Beginners → skipped I'll check back in 2 hours and report what shipped. ``` Then either: - Schedule a follow-up task for `now + 2h` that re-invokes this skill with the saved jobId list (if your runtime supports it), OR - Ask the user to set a 2-hour reminder, and persist the jobId list where the next invocation can find it (note in the conversation, memory, etc.). --- ## 2-hour review flow Triggered 2 hours after step 5 (or whenever the user asks "how did today's articles go?"). For each blog you triggered today: ```bash curl -sS -X POST https://autopod.co/api/v1 \ -H "Authorization: Bearer $AUTOPOD_API_KEY" \ -H "Content-Type: application/json" \ -d '{"action":"getJobStatus","tenantId":"","jobId":""}' ``` If you don't have the jobId list (e.g. starting fresh in a new session), fall back to: ```bash curl -sS -X POST https://autopod.co/api/v1 \ -H "Authorization: Bearer $AUTOPOD_API_KEY" \ -H "Content-Type: application/json" \ -d '{"action":"listJobs","tenantId":"","sinceMinutes":180}' ``` Each job response contains both a top-level `status` and (for API-triggered jobs) a `publish` block with the downstream pipeline: ```json { "success": true, "job": { "id": "abc...", "status": "completed", "articleId": "art123", "articleSlug": "how-lasik-affects-long-term-night-vision", "publishError": null, "publish": { "articleStatus": "published", "imageError": null, "translationsExpected": 5, "translationsCompleted": 5, "translationErrors": [], "audioExpected": 1, "audioCompleted": 1, "audioErrors": [] } } } ``` Map each job using BOTH fields: - `status: completed` + `publish.articleStatus: "published"` + zero errors → fully shipped. Note `articleId` + `articleSlug`. - `status: completed` + `publish.articleStatus: "published_no_image"` → shipped but image generation failed; surface `publish.imageError`. - `status: completed` + `publish: null` → an older job not from the API path (dashboard-triggered draft). Skip it from Open Claw reporting. - `publish.translationsCompleted < publish.translationsExpected` AND no errors → translations still running, re-check in 15-30 min. - Any non-empty `publish.translationErrors` / `publish.audioErrors` arrays → list each `lang` + `error` to the user verbatim. - `publish.audioCompleted < publish.audioExpected` → podcast still being generated (or failed if `audioErrors` populated). - `status: processing` / `pending` — still generating the article itself. Don't treat as failure unless it's been > 90 minutes. - `status: failed` — surface the `error` field verbatim. - `publishError` — finalization broke (article generated but couldn't publish). Surface verbatim. Compose a single status report for the user. Use checkmarks per pipeline stage so they know exactly what shipped vs what's still running: ``` Today's Open Claw run — review ✓ Visual Field Test "How LASIK Affects Long-Term Night Vision" — published. article: ✓ image: ✓ translations: 5/5 ✓ podcast: ✓ Slug: how-lasik-affects-long-term-night-vision ⏳ Cooking With Less Salt "5 Ways To Replace Salt In Soup" — published, still finishing up. article: ✓ image: ✓ translations: 3/5 ⏳ podcast: ⏳ I'll re-check in 30 minutes. ✗ Mountain Biking Blog Generation failed: "OpenAI API Error: 429 rate_limit_exceeded". You can retry from the dashboard, or I can re-trigger it with the same suggestion when you say so. Nothing else queued today. ``` If anything is in `failed` state, or any `publish.*Errors` array is non-empty, **always** tell the user — never silently swallow it. If the user asked you to be 99.9% hands-free, the only way that works is by reporting failures loudly. --- ## API reference (quick) Endpoint: `POST https://autopod.co/api/v1` Auth: `Authorization: Bearer ` where the key is either `apod_…` (tenant) or `apoa_…` (account). | action | body fields | notes | |----------------------|--------------------------------------------------------|-------| | `listBlogs` | — | Tenant key returns 1 blog. Account key returns up to 50. | | `listRecentArticles` | `tenantId?`, `limit?` (default 20, max 100) | Published only. | | `listSuggestions` | `tenantId?`, `limit?` (default 5, max 20) | Most-recent first. | | `generateSuggestions`| `tenantId?`, `count?` (default 5, max 10) | Server already dedups vs last 50 articles + 50 suggestions. | | `runAutoPilot` | `tenantId?`, `suggestionId?` | Picks the most recent suggestion if `suggestionId` omitted. Server-side autopilot: generate text → image → publish → translations → audio → Buzzsprout. Returns `jobId`. | | `getJobStatus` | `tenantId?`, `jobId` | Returns `status` + `publish` block (articleStatus, imageError, translationsExpected/Completed, audioExpected/Completed, error arrays). | | `listJobs` | `tenantId?`, `sinceMinutes?` (default 180, max 10080), `limit?` (default 50, max 200) | Used for the 2h review. Summary includes total, completed, failed, inProgress, published, translationsExpected/Completed, audioExpected/Completed. | `tenantId` is **required** for every action except `listBlogs` when using an account key. With a tenant key, `tenantId` is ignored. ## Failure modes you must surface - `401` → "Invalid API key. Did the user revoke it in the dashboard?" - `403` "Tenant subscription is not active" → tell the user that blog needs a renewed/active plan. - `429` "Monthly article limit reached" → tell the user which blog. - `404` "No suggestions available" on `runAutoPilot` → run `generateSuggestions` first, then retry once. - 5xx → tell the user the call failed, include the response body, do not retry more than once. ## Environment - `AUTOPOD_API_KEY` — required. The user's `apod_…` or `apoa_…` key. - `AUTOPOD_BASE_URL` — optional override of `https://autopod.co/api/v1`, useful for testing against `https://autopod-test-env.web.app/api/v1`. If `AUTOPOD_BASE_URL` is set, use it everywhere instead of the prod URL.