Turns what you can't articulate into agents that work.
A meta-agent that designs agents. Say one vague sentence — it confirms the details with multiple-choice cards, then delivers a ready-to-use prompt, SOP, or a full agent asset package with built-in QA.
What this is: PromptDog isn't a website or an app — it's a skill package you install into your AI (Codex / Claude Code).
Download the folder, hand it to your agent, then just say "build me a WorkDog for weekly reports." This page is only a demo.
The problem
The hardest part of putting AI to work is spelling things out
PromptDog takes the spelling-out off your plate — it guesses first, you just pick. Why the current way is exhausting:
1
Too much to explain
Say “write my work summary” and it fires back: which highlights? what metrics? for whom? — typing all that out is more work than just writing it yourself.
2
Complex work gets one paragraph
Contract review, resume screening — process jobs — and it returns one professional-sounding paragraph. What got missed? Is it right? You can't tell.
3
Tuned-in experience doesn't stick
You finally got it dialed in; the next task starts from zero. Your know-how evaporates into chat history.
LIVE DEMO
One PromptDog, many WorkDogs
Meet the two roles first, then watch each one work — auto-plays · switch scenarios · replay.
🐕 PromptDog — the skill that builds agents
Say a vague need; it asks multiple-choice questions to lock the industry details, then generates a complete agent asset package. Demo ① is PromptDog itself at work.
🐾 WorkDog — the scenario-specific agent it produces
Drop the folder into your workspace and it's on duty — dedicated to one scenario, living in the kennel/, getting sharper with use. Demos ②③④ are three WorkDogs already on the job.
Your one line→🐕 multiple-choice confirm→generate asset package→drop into workspace→🐾 WorkDog on duty
Same model — why is a WorkDog better than a plain agent? A plain agent dumps one prompt in one shot and hopes. A WorkDog runs in stages, sets acceptance criteria first, has an adversarial reviewer hunt for flaws, and sharpens through retros & regression — quality is designed, not lucked into.
PromptDog WorkbenchDEMO
🐕 PromptDog🐾 WorkDogs
Auto-playing demo… (switch scenarios / replay)
Quality by design
It sizes the job before doing the job
Small ask
One great prompt — copy, paste, done. No questions asked.
Multi-step
An SOP with a checkpoint at every step.
Complex
A staged pipeline — clean context per stage, checks between stages.
Heavy
An AI team with divided roles that review each other.
Four safeguards on every delivery: acceptance criteria agreed up front · critical claims cross-checked · an adversarial reviewer before shipping · a hard stop-loss after three rounds. Quality is designed, not lucked into.
The kennel
Ready-to-work WorkDogs
WorkDog
What it does
📑 Bid Writing
Parses the tender, builds a scoring matrix, drafts per scoring point, checks every knockout clause, runs a mock evaluation.
⚖️ Contract Review
7-stage first-pass review with law-firm red lines; delivers report + clause annotations + structured JSON.
✍️ Content Marketing
"Trust through transparency" methodology; corpus-grounded — it never fabricates a number; tri-cut per platform.
Every WorkDog ships with launch commands, a retro flywheel and regression gates — the more you use it, the sharper it gets.
作者 Orcastao