template
Distribution Autopilot Kit
Agents that run your site — publish, measure, fact-check, and flag runs that wrote nothing.
✓ = we've run it on that platform · others load the same SKILL.md format but aren't run-tested yet
Nine agent plays that publish what you wrote, measure what happened, catch when the world moves under your claims, and supervise each other. Built around one rule: a run that produces none of its declared artifacts is a failure, however cleanly the agent exited. Extracted from a three-brand fleet where three scheduled jobs reported green for days while writing nothing.
distribution-autopilot/ ├── README.md · INSTALL.md · RUN-UNATTENDED.md · LICENSE ├── plays/ ← 9 agent skills │ ├── ops-supervisor/ daily — watches the other eight │ ├── content-shredder/ daily — one page → per-platform drafts │ ├── demand-watch/ daily — what people actually asked │ ├── tool-watch/ daily — what moved in your space │ ├── freshness-check/ weekly — what changed under your claims │ ├── weekly-analytics/ weekly — snapshot → what to do next │ ├── aeo-weekly/ weekly — fix the most wrong page │ ├── visibility-tracker/ weekly — are you cited, and by whom │ └── reddit-scout/ 2×/week — drafts, never posts ├── ops/ ← collect + publish workers (hold the keys) ├── templates/ ← distribution.yaml, CI workflows, webhook └── verify.py ← "is this actually working?"
Why it's different
- ▸ Credentials never reach a scheduled agent — collectors hold keys, plays read sanitized JSON from git
- ▸ Auto-posts to dev.to and Bluesky; LinkedIn, Twitter and HN go to a Telegram approve/skip queue
- ▸ Reddit subreddit posting is human-only by design — bans are by domain and permanent
- ▸ Guards refuse rather than degrade: over-limit posts, missing canonical URLs, and failed sends all retry instead of silently passing
- ▸ Works with zero credentials — you get drafts in git, and each key turns on one more channel
Runs unattended
The recurring skills in this kit ship with a scheduling procedure — not a guide for you to follow, but a file your agent reads and executes. Paste one line; it installs to the right directory for your platform, creates the schedule, and proves the first run actually wrote something.
Read RUN-UNATTENDED.md in this folder and put this kit's plays on a schedule for my platform, verifying each step.
| Platform | Runs with your laptop closed | How |
|---|---|---|
| Claude Code | Yes — Anthropic's cloud | Routines, hourly or slower |
| Hermes | Only while its host stays up | cron, any interval |
| Codex | Needs a scheduler you own | codex exec + launchd or CI |
Every scheduled run has to name the file it must write, and records a result either way. A run that finishes cleanly having produced nothing is recorded as a failure — which is the difference between automation you can leave alone and automation that has quietly stopped working.
Questions, answered
Do I need Claude specifically?+
No. The plays are SKILL.md files — hand the kit to Claude Code, Codex, Hermes, or anything that reads them. INSTALL.md and RUN-UNATTENDED.md are written for the installing agent, and cover scheduling on Claude Routines, Hermes cron, codex exec, and GitHub Actions.
Is this the same as the AI Search Visibility Kit?+
No — they are opposite halves. That kit BUILDS a programmatic site. This one RUNS a site that already exists: publishing, measurement, accuracy checks, and a supervisor. They compose, and neither requires the other.
What if I have no analytics?+
Set the provider to none and Search Console becomes your audience signal. The kit ships adapters for Cloudflare D1 and GA4, and it refuses to pretend a configured-but-broken source is working — that distinction is most of the point.
Does it post to Reddit?+
To your own profile, optionally. Never to a subreddit. Reddit bans by domain rather than by account and it is not reversible, so the Reddit play drafts comments and a human posts them. That is not configurable.
What does it cost to run?+
The plays run on the agent subscription you already pay for. Each optional key (SerpAPI, Composio, GA4, Telegram…) unlocks one more channel and every one has a free tier — with zero keys you still get drafts committed to git.
Kits that pair with this one
Each workflow kit teaches a different job. They share the same skill standard, so they stack.
Prompt Ops: 150 Production Prompts
150 production prompts, annotated — the failure modes documented for every one.
$29$19
VIEW
Skill Forge
43 production agent skills — plus the framework and test harness to write your own.
$89$59
VIEW
Outreach Machine
Cold outreach your agent researches and drafts — enrichment → sequence → reply triage.
$69$49
VIEW