by /
- 1What was actually wrong
- 2How the 1–5 bar is scored
- 3The line, named once
The Task Library was a flat list of 240 SOPs. Agents opened one, did it, and treated the next as a new chat. The Content Factory is a line. This run put the line on the dashboard, scored every task 1 to 5 from evidence, and fleshed the stations that make or break ads.
243tasks in the rebuilt library
123still needs-work or gap (inventoried, not sampled)
75importance-5 stations (ads, gates, every-run spine)
1playbook for Claude-only or Grok-only
What was actually wrong
The files were not empty. Almost every skill already had Inputs, Steps, and a Definition of done. Complete vs needs-work was mostly whether a nine-requirement definitive article existed. Agents still failed the factory because three things were missing from the dashboard they actually open.
| Gap | What an agent did | What we shipped |
|---|---|---|
| No priority | Picked the next yellow row, or the one with a short name | A 1–5 volume bar on every row. Score is max(frequency, revenue, gating). A pixel install is a 5 because it unblocks spend. |
| No chain | Ran Jennifer, or posted to WordPress, as if that were the job | Four-phase diagram on the dashboard: Produce → Process (Descript) → Post → Promote, with Digital Plumbing as Gate. Each row names Before and After. |
| Vendor fork | Waited for Claude, or invented a Grok-only SOP | run-content-factory-on-any-engine. Single-engine is first-class. Multi-engine is optional throughput on the same line. Handoff is files, never one vendor’s memory. |
How the 1–5 bar is scored
Not vibes. build/factory.py takes the max of three evidence axes:
- Frequency — every factory cycle and weekly MAA beat a yearly SOP audit.
- Revenue — Dollar a Day, boosts, remarketing, and the pixel that makes them attributable are 5. Publishing the asset ads will amplify is 4.
- Gating — a small get-access task is a 5 if skipping it blocks the chain. Search Console is the example we already paid for: 16 articles shipped to a site with no GSC, six never indexed, weekly reports that said “needs more time.”
Distribution after the rebuild: 75 at 5, 36 at 4, 100 at 3, 32 at 2, none at 1. Filter the dashboard on “5 only” or click a phase. (An early pass scored 83 fives because the keyword author fired inside authoritative — hyphen-token matching closed that.)
The line, named once
Three zoom levels, one factory. Task Library column names may still say Gate/Plumbing or MAA; the factory line itself is only Produce → Process → Post → Promote. Plumbing and Perform sit before and after the factory — not as factory stages. Agents were treating them as competing diagrams.
| Task Library | Factory line (Produce → Process → Post → Promote) | 6-stage assembly |
|---|---|---|
| Gate / Plumbing | Plumbing (onboarding before the factory) | (before capture) |
| Produce (gather) | Produce | Capture |
| Process (Descript) | Process | Transcribe → hub → atomize |
| Post | Post | Distribute |
| Promote (ads) | Promote | Boost |
| MAA | Perform (MAA after the factory) | (read the numbers) |
The public Content Factory article now carries that mapping so the pack skill and the Task Library cannot drift again.
What we fleshed, not just flagged
123 incomplete skills are listed in INCOMPLETE-INVENTORY.md with star, phase, and lane. Highest-value incomplete files got operational depth this run: failure modes, real examples, and which engine can run them.
| Station | Why it was a 5 |
|---|---|
| GSC + GA4, GTM, Meta pixel, Meta Business Manager | Gates. Ads without a pixel is spend you cannot retarget. Articles without GSC is content you cannot measure. |
| Jennifer (A- terminator) | The pack already had the rubric. The Task Library station now calls it, stops at A-, and flags a wrong WordPress author as a B+ cap. |
| Dollar a Day creatives, switch-boost, cold→warm→conversion, remarketing | Highest money. Boost winners only, one variable at a time, 1-day abandoner window. |
| Content Library, Descript clips, Facebook page post, WP author | The handoff files. If the work only lives in a chat, the next engine starts from zero. |
Single-engine is not a consolation prize. If you only have Grok, or only have Claude, you still run Produce → Process → Post → Promote. Optional local Qwen overnight writing is the same Process station with cheaper tokens. It is not a second factory.
Where the live dashboard now lives
The WordPress page at /task-library-dashboard/ is a shell. The app is GitHub Pages from Local-Service-Spotlight/task-library — a repo this org can write. The iframe is pointed at that host so the next score or chain change does not wait on a personal GitHub account. Build: python3 build/build.py. Tests: python3 build/test_factory.py.
123 skills are still yellow or red. That is not a secret. Complete means a nine-requirement definitive article plus a worked example, not a fleshed SOP. The inventory is the work order. Next agent: keep walking the 5-star incomplete list; do not re-sample.
Run the factory, not a chat
243 skills, a 1–5 bar on every row, the four-phase line on the page, and a single-engine path.
Open the Task Library Content Factory hub How the library was built Persistent agentsDennis Yu
Dennis Yu is the CEO of Local Service Spotlight, a platform that amplifies the reputations of contractors and local service businesses using the Content Factory process. He is a former search engine engineer who has spent a billion dollars on Google and Facebook ads for Nike, Quiznos, Ashley Furniture, Red Bull, State Farm, and other brands.
Dennis has achieved 25% of his goal of creating a million digital marketing jobs by partnering with universities, professional organizations, and agencies. Through Local Service Spotlight, he teaches the Dollar a Day strategy and Content Factory training to help local service businesses enhance their existing local reputation and make the phone ring.
Dennis coaches young adult agency owners serving plumbers, AC technicians, landscapers, roofers, electricians, and believes there should be a standard in measuring local marketing efforts, much like doctors and plumbers must be certified. He has appeared on 353 podcasts with 619 credited episodes — see the full list of his podcast appearances.