Five agents, three artifacts, scorer-gated quality.
One pass through the JD ships your cover letter, your custom HTML proposal page, and your reply-scoring loop. The scorer blocks anything below operator quality before it leaves your dashboard.
Paste the JD
You paste the Upwork job description. Closeline parses budget, posted-time, buyer signals, scenario type. Decides if it's worth your connects in 8 seconds.
Five agents argue
Buyer Psychology reads the JD for power posture and hidden fears. Specifics-Density Scorer counts named brands and methods. Fingerprint Scrubber catches 30+ LLM-tells. Bookend Validator locks the opening and sign-off. Fact Guard blocks unverified claims. They argue. They lock a strategy.
Sonnet 4.5 drafts
With the locked strategy as a system prompt + the verified-facts whitelist, Sonnet drafts the cover letter at temperature 0.45. Prompt caching keeps the cost down — most calls hit cache.
Scorers gate the draft
Specifics density must be ≥9 per 100 words. ≥18 total specifics. Opening 10-22 words anchored on the buyer's noun. Byte-exact sign-off. Zero blocking fingerprint hits. If any fail, the system auto-refines ONCE with the failure reasons injected. If it still fails, it blocks.
Custom HTML proposal page
In parallel, Closeline scratch-generates an HTML proposal page tuned to the buyer — their stack, their pain, their decision trigger. Four layout recipes, hash-keyed to the buyer so two clients in the same archetype never see the same page.
Reply scoring + typed memory
Every buyer reply gets scored 0-100 on close-probability. Stage detected: cold → interested → vetting → ready_to_buy. Typed memory locks pricing frames, tone, hook archetypes that worked. The next draft starts smarter.
See it on your own JD.
One email when the first cohort opens — bring a JD to the call, we'll run it live.