content-engine-cron

A weekly content pipeline for solo creators. One Anthropic API call a week plans your posts, drops them straight into your phone calendar, and measures what worked. No servers, no auto-posting, no databases.

You write the posts. The pipeline tells you when, what about, and how it's doing.

---

How it works

Sunday  →  weekly_planner.py     (Claude + web_search → next week's cues)
        →  build_calendar.py     (cues + your schedule → .ics file)
        →  publish.sh            (.ics → public Gist)
                ↓
          iPhone calendar subscription refreshes ~12h
                ↓
Mon–Sat →  Each slot fires as a calendar event with a cue + template ref
        →  You write the post, hit publish
                ↓
Sunday  →  Drop analytics CSV in data/, run score.py, read what worked

Five posts a day, six weekdays + a Sunday block. All times, slots, and themes live in schedule.json — change them however you like.

---

Quickstart

git clone https://github.com/<your-username>/content-engine-cron.git
cd content-engine-cron

bash setup.sh

setup.sh is idempotent. It:

• Copies each *.example.* file to its real name (only if missing — won't clobber)
• Creates a Python venv at .venv and installs requirements.txt
• Prints next-step instructions

Then:

# 1. Edit configs
$EDITOR .env             # paste your ANTHROPIC_API_KEY
$EDITOR profile.json     # who you are, your voice, your topics
$EDITOR schedule.json    # slot times + weekly themes (defaults are sensible)

# 2. Smoke test — no API call, no cost. Confirms your configs parse and
#    shows the prompt the planner will send to Claude.
source .venv/bin/activate
python3 weekly_planner.py --dry-run

# 3. Authenticate the GitHub CLI for gist publishing
gh auth login

# 4. Run the pipeline
bash sunday_run.sh

The script prints a Gist URL on first run. On iPhone: Settings → Calendar → Accounts → Add Account → Other → Add Subscribed Calendar → paste the URL → Save. Re-runs update the same Gist; iOS auto-refreshes within ~12h.

---

File map

.env.example                # Placeholder — copy to .env, add ANTHROPIC_API_KEY
profile.example.json        # Identity + voice + search topics + rules
schedule.example.json       # Slot times, Sunday block, weekly themes
score_config.example.json   # Pillar/hook keywords for auto-tagging
templates.example.md        # Reusable post templates
tracker.example.csv         # Header-only template

setup.sh                    # One-command bootstrap (run this first)
weekly_planner.py           # Claude + web_search → cues for next week
build_calendar.py           # Cues + schedule → content_schedule.ics
publish.sh                  # Push .ics to a public Gist
sunday_run.sh               # plan + build + publish (run weekly)
score.py                    # Analytics CSVs → tracker → weekly_report.md
requirements.txt            # Just `anthropic`

.github/workflows/sunday.yml   # Optional: GitHub Actions weekly cron

data/x_raw/                 # Drop X analytics CSV exports here
data/yt_raw/                # Drop YouTube Studio CSV exports here

*.example.* files are committed templates. Your live configs (profile.json, schedule.json, etc.) and all generated artifacts are gitignored.

---

Customizing for your niche

You only need to edit two files to make this yours:

profile.json — who you are

• name, handle, bio_paragraph — what the planner says about you in its system prompt
• domain_descriptor — e.g. "DeFi and AI-agent", "climate-tech", "open-source ML"
• slot_topics — what each of your 5 daily slots is about
• search_queries — the planner uses these as web search seeds, in order
• rules — non-negotiable constraints the planner must follow when writing cues

schedule.json — when and what

• daily_slots — five slots a day (time, duration, summary, description). Slots 2/3/4 receive an AI cue; slots 1 and 5 are static.
• sunday_events — the Sunday block (thread / batch / review).
• weekly_themes — array of themes to rotate through. Each has a name, a Sunday-thread suggestion, and template cues for Mon–Sat that fall back if the planner is offline.

score_config.json and templates.md (optional)

Defaults work out of the box. Edit the keywords in score_config.json to match your domain so the scorer can auto-tag posts into pillars (e.g. gm, personal, builder, your-thing) and hook types (contrarian, proof_of_work, etc.).

---

Weekly ritual

# Sunday, 15 min:
# 1. Export X analytics CSV → drop in data/x_raw/
# 2. Export YouTube Studio CSV → drop in data/yt_raw/  (optional)
# 3. Score
python3 score.py
# 4. Read weekly_report.md, mark 1–2 winners
# 5. Plan + publish next week
bash sunday_run.sh

---

Virality score

score = impressions
      + 20 * reposts
      + 15 * bookmarks
      + 10 * replies
      + 25 * new_follows
      +  5 * profile_visits

Optimizes for reach × quality × audience growth, not vanity likes. Tune the weights in score.py:score_row if you want a different signal.

---

Automating the Sunday run

Pick one. All three do the same thing — run sunday_run.sh weekly.

A. macOS launchd

mkdir -p ~/Library/LaunchAgents
cat > ~/Library/LaunchAgents/com.content-engine-cron.plist <<PLIST
<?xml version="1.0" encoding="UTF-8"?>
<!DOCTYPE plist PUBLIC "-//Apple//DTD PLIST 1.0//EN" "http://www.apple.com/DTDs/PropertyList-1.0.dtd">
<plist version="1.0">
<dict>
  <key>Label</key><string>com.content-engine-cron</string>
  <key>ProgramArguments</key>
  <array>
    <string>/bin/bash</string>
    <string>-lc</string>
    <string>cd \$HOME/projects/content-engine-cron && bash sunday_run.sh >> sunday_run.log 2>&1</string>
  </array>
  <key>StartCalendarInterval</key>
  <dict>
    <key>Weekday</key><integer>0</integer>
    <key>Hour</key><integer>18</integer>
    <key>Minute</key><integer>0</integer>
  </dict>
</dict>
</plist>
PLIST
launchctl load ~/Library/LaunchAgents/com.content-engine-cron.plist

Stop with launchctl unload ~/Library/LaunchAgents/com.content-engine-cron.plist.

B. Linux / cron

0 18 * * 0  cd $HOME/projects/content-engine-cron && bash sunday_run.sh >> sunday_run.log 2>&1

C. GitHub Actions (no machine needed)

.github/workflows/sunday.yml runs the pipeline on GitHub's runners every Sunday. Setup:

1. Fork this repo.
2. Settings → Secrets and variables → Actions → New repository secret. Add:
   • ANTHROPIC_API_KEY — your Anthropic API key
   • PROFILE_JSON — full contents of your profile.json
   • SCHEDULE_JSON — full contents of your schedule.json
   • GH_PAT_GIST — a Personal Access Token with gist scope (the default GITHUB_TOKEN can't edit gists)
3. First run: trigger the workflow manually from the Actions tab. It creates a public gist and prints its ID in the run logs.
4. Add that ID as a fourth secret named GIST_ID so subsequent runs update the same gist instead of creating new ones.

Subscribe iOS to the gist's raw URL exactly as in step 4 of Quickstart.

---

Privacy

What gets sent off your machine, ever:

• One Anthropic API call per week (your system prompt + a web-search request).
• The generated .ics file gets pushed to your own public Gist for iOS to subscribe to.

Gitignored, never published:

• .env (your API key)
• profile.json, schedule.json, score_config.json, templates.md
• tracker.csv (your post-level analytics)
• data/*/raw/*.csv, current_week.json, weekly_report.md, content_schedule.ics, .gist_id

Forks have their own everything. There is no shared backend.

---

Cost

One Anthropic API call per week:

• Claude Opus 4.7 with adaptive thinking + web_search
• 16k output tokens cap (usually less)
• Cents to a couple of dollars per run depending on how much search/reasoning happens

Pricing: https://www.anthropic.com/pricing

---

Not in scope

• Auto-posting. The pipeline drafts and measures. You hit publish.
• API ingestion. CSV-export → drop-in-folder is the cheap path. Add X / YouTube API ingestion if rate-limit and OAuth complexity feel worth it to you.

---

License

MIT. See LICENSE.