Three agents that run daily on GitHub Actions across every Bizzal channel that touches a viewer (YouTube, Instagram, and the three blogs + their X accounts).

Advisory only — none of them ever edit or publish your content. They read, measure, and critique; you (or a PR) act on it.

They're deliberately ordered: scout gathers opportunities, measure pulls real numbers, then the reviewer critiques. Everything degrades gracefully — a dead feed, a missing key, or one bad property is reported, never fatal.

scout/     feeds.yml → fetch_feeds → rank (Claude) → opportunities/*.json + digest issue
measure/   collect_youtube (Data API) + collect_instagram (published file) → metrics/*.json + digest
reviewer/  targets.yml → collect → judge (Claude + rubrics) → per-repo issues + digest

Scores every property against a per-channel rubric and files the critique as GitHub issues. Targets live in reviewer/targets.yml.

deferred = no free way to read it (login/bot walls). The property is still reviewed on its other sources, and the judge is told not to score criteria it couldn't actually see — an unavailable source never drags the score down.

Instagram is fully reviewed. IG can't be read for free, and personal-account secrets don't cross repos — so Bizzal-Games-YT-PUB (where the IG token already lives and posts) collects insights into its data/metrics/instagram.json, and reviewer/instagram.py reads that published file. No IG secret is needed in this repo.

How it reports

• a detail issue per property in its own repo (via BIZZAL_REVIEW_PAT), so findings sit next to the code that produced them — one stable issue, appended daily;
• a digest issue here, authored by github-actions[bot] via the default GITHUB_TOKEN — that authorship is deliberate: GitHub never emails you about your own PAT-authored actions, so the bot-authored digest is the one that actually reaches your inbox. It links out to every detail issue.

Rubrics live in reviewer/rubrics/: youtube · instagram · blog · blog_growth

blog_growth is an alternate lens — it ignores editorial quality and scores purely on clicks (CTR) and follower conversion. Apply any rubric to any run with the rubric input (see below).

Pulls real numbers so review stops being opinion and starts being evidence. Writes a time-series to measure/metrics/<channel>.json (committed daily) and files a metrics digest issue.

Video IDs come from the uploads playlist via the Data API, not the public RSS feed — the feed 404s intermittently even from Actions.

Not yet wired: YouTube retention/CTR/watch-time (needs the Analytics API + OAuth) and blog analytics (GitHub Pages has none; GoatCounter is the plan).

Unifies content discovery. Where the reviewer critiques what's published, the scout finds what to make next.

• scout/feeds.yml — one registry of every RSS/Atom/social feed, tagged by topic (ttrpg / crypto / cyber). A feed shared across beats (e.g. The Hacker News → crypto + cyber) is fetched once and routed to every channel that wants it.
• Per-channel opportunities — ranked by Claude into concrete title/hook ideas with a why-now and a source link.
• Cross-channel plays — one story that feeds 2+ channels (a supply-chain hack for crypto-security and cyber), with the per-channel spin. This is the point of unifying: the four pipelines were previously blind to each other.

Output: machine-readable queues at scout/opportunities/<channel>.json (for a pipeline's create step to consume) plus a daily digest issue. Advisory today — not yet wired into the pipelines' generation step.

Repo → Settings → Secrets and variables → Actions. bizzal70 is a personal (non-org) account, so secrets do not carry over from your other repos — each must be added here, even if the same value already exists elsewhere.

Secrets

No Instagram or scraper key is needed here — IG metrics arrive via the games repo's published file, and all other collection is key-free.

Variables (optional)

Unset Actions variables arrive as an empty string, not as absent — the code coalesces blanks to the defaults above with or, not dict.get(k, default).

Reviewer inputs

• only — one property id or a comma-separated list; blank = all. Ids: bizzal-games-youtube, bizzal-games-instagram, itsalreadywritten, itsalreadypriced, itsalreadywhen
• rubric — override the rubric for every reviewed property; blank = each property's default. e.g. blog_growth for a clicks/followers pass.

# score just the crypto blog on the growth lens
gh api -X POST repos/bizzal70/Audit_User_Agent/actions/workflows/daily-review.yml/dispatches \
  -f ref=main -f "inputs[only]=itsalreadypriced" -f "inputs[rubric]=blog_growth"

Local debug

REVIEW_ONLY=itsalreadypriced python -m reviewer.run
python -m scout.run
python -m measure.run