Build · Create · Engineering
Goal decomposed into an executable task plan
A goal broken into ordered, tool-assigned steps with a resolved, acyclic dependency graph, so an agent (or a team) can execute it without ambiguity.
You receive: A structured plan (JSON: steps with id, prompt, parameters, assigned tool, dependencies) that passes the completeness + acyclic-graph checks.
Part of Ship MVP
Example
A sample of what this play produces. Your result is generated for your inputs.
Goal
Stand up STUD.com's content-led acquisition loop for the soft launch: publish play-grounded SEO pages and launch-community posts that convert readers into workspace signups, instrumented end to end, sized to the illustrative launch scenario of 200 signups.
Steps
- Id
- s1-fact-baseline
- Prompt
- Read the STUD workspace facts and extract the acquisition-relevant baseline: the one-liner, target customer (solo and early-stage founders, small product teams), pricing (prepaid credit packs, 1 credit is 4 cents, 500 credits for $20 up to 5,000 credits for $200, purchased credits never expire, most plays cost 500 credits or $20), the 265-play catalog across 14 playbooks, and the illustrative launch scenario (200 signups, 40 activated buyers month 1, illustrative blended CAC about $40). Output a single facts brief every later step consumes, with each figure tagged real or illustrative.
- Parameters
- Source
- workspace facts, refreshed 2026-08-13
- Tag provenance
- true
- Assigned to
- file_read
- Id
- s2-keyword-landscape
- Prompt
- Search for the query landscape where STUD's target buyers already look: queries like 'AI agent for market sizing', 'automate competitor analysis', 'AI task decomposition plan', and category queries against the named competitor set (general-purpose AI chat assistants, freelance marketplaces, static documentation tools). Return the 40 highest-intent queries with the searcher's job-to-be-done for each.
- Parameters
- Query count
- 40
- Competitor frames
- ai chat assistant
- freelance marketplace
- documentation tool
- Assigned to
- web_search
- Dependencies
- s1-fact-baseline
- Id
- s3-play-query-map
- Prompt
- Map each of the 40 queries to the single live play in the 265-play catalog that most directly does that job. Drop queries with no honest play match rather than stretching a claim; the page must only promise what the play's acceptance check actually enforces. Output query, play slug, playbook, and the honest one-sentence claim.
- Parameters
- Catalog size
- 265
- Drop unmatched
- true
- Assigned to
- llm
- Dependencies
- s2-keyword-landscape
- Id
- s4-page-template
- Prompt
- Draft the SEO page template in STUD's register: living-system framing first (STUD knows your business, does the work, learns from every result), the mapped play as the concrete answer to the query, a short intake walkthrough, and the acceptance check as a supporting mechanism detail, never the headline. Enforce the brand voice rules: plain factual register, no em dashes, no coax or discount framing.
- Parameters
- Register
- D-027 living-system
- Sections
- query answer
- play walkthrough
- how results compound
- pricing
- Assigned to
- llm
- Dependencies
- s1-fact-baseline
- s3-play-query-map
- Id
- s5-page-batch
- Prompt
- Author the first batch of 15 pages from the template, one per mapped query, each grounded in the facts brief: real facts stated as fact, illustrative figures (CAC about $40, the 200-signup scenario) labeled illustrative wherever cited. Every page ends with the play's real intake fields so the reader sees exactly what a run asks for.
- Parameters
- Batch size
- 15
- Label illustrative
- true
- Assigned to
- llm
- Dependencies
- s4-page-template
- Id
- s6-community-posts
- Prompt
- Write 5 launch-community posts (founder-forum register, first person from Dan Schmitz) that each walk one real play end to end: the goal, the filled intake, the deliverable that landed, and how a workspace fact stated once flowed into the run. No verification stamps or guarantee language; the artifact speaks for itself.
- Parameters
- Post count
- 5
- Voice
- founder first person
- Assigned to
- llm
- Dependencies
- s3-play-query-map
- s4-page-template
- Id
- s7-instrumentation
- Prompt
- Define and compute the funnel instrumentation: page view to signup to first run, attributed by landing query. Write the analysis script that reads the event export and reports signups per query cluster and blended cost per signup against the illustrative $40 CAC ceiling and the 200-signup launch target.
- Parameters
- Funnel
- view
- signup
- first_run
- Cac ceiling usd
- 40
- Signup target
- 200
- Assigned to
- python_exec
- Dependencies
- s5-page-batch
- s6-community-posts
- Id
- s8-freshness-check
- Prompt
- Fetch each published page and community post URL and verify it resolves, renders the current pricing (prepaid credit packs: 500 credits for $20, 1,000 for $40, 2,500 for $100, 5,000 for $200; most plays 500 credits), and carries no stale vocabulary (no 'hire', no 'objective'; plays are run). Output a defect list keyed by URL.
- Parameters
- Pricing literals usd
- 20
- 40
- 100
- 200
- Banned terms
- hire
- objective
- Assigned to
- http_fetch
- Dependencies
- s5-page-batch
- s6-community-posts
- Id
- s9-iterate-or-scale
- Prompt
- Read the s7 funnel report and the s8 defect list and decide the next batch: fix defects first, then either double down on the query clusters converting under the $40 illustrative CAC or re-map the underperformers to different plays. Output the ranked next-15-page queue with the reason for each inclusion, sized to the single founder-operator constraint.
- Parameters
- Next batch size
- 15
- Decision rule
- scale clusters under CAC ceiling, re-map the rest
- Assigned to
- llm
- Dependencies
- s7-instrumentation
- s8-freshness-check
Opens soon
STUD opens runs for everyone soon, on every play at once.
Scope of the check
What's verified: STUD verifies the submitted plan (JSON) is structurally complete and executable on paper: every step has a unique id, a prompt, a parameters object, and a tool assigned from your allowed tools, every dependency resolves to a real step with no self-reference, the dependency graph is acyclic, no step is orphaned, and the step count is within your max. STUD does NOT judge whether the steps are the right way to reach the goal or whether an agent could actually carry them out.
Get early access to STUD the day it goes live.