Execution briefs
Make the brief repeatable before the work
SmaugBrain provides plan access, Agent management and an account entry for opening the configured workspace. A reusable execution brief connects those capabilities to a bounded piece of storefront work. The brief, evidence and human sign-off make a run repeatable; repeated prompts alone do not.
On this page
Treat the brief as the unit of work
Choose one question with a testable finish: review a product page for a particular market, for example. Separate supplied facts, assumptions and requested changes. A broad instruction to improve the store makes completion difficult to judge.
Keep the brief with your project materials. The personal center manages access and Agents; it is not a project file library or a delivery approval system.
Reuse means keeping the question and acceptance criteria stable while replacing the inputs. Give each run a short identifier and note which sources changed since the previous review. If a new market introduces a different policy, claim or customer concern, revise the brief instead of treating it as another identical run.
Prepare the source pack
Collect only what this question needs, with an identifiable version.
Business context
Name the storefront, target market, product, intended customer and decision owner. State which claims or requirements are already approved.
Evidence
Provide dated page captures, product specifications, policy text or relevant export excerpts. Identify the source and any missing permissions or data.
Constraints
Specify the output language, allowed sources, scope exclusions and reviewer. Remove customer identifiers and credentials that the work does not require.
Acceptance
Describe the expected format and checks. Request references for findings, explicit assumptions, unresolved questions and a separate proposal for changes.
Resolve account dependencies in order
For a paid plan, follow these dependencies. Existing granted access still needs a valid plan; creating another Agent also requires available quota.
| Dependency | Where to check | What must be established |
|---|---|---|
| Payment | Your order | Use confirmed payment status, not a success return URL. Continue an existing pending checkout when available. |
| Plan | Account overview | The plan must be active and unexpired. Payment options depend on current site configuration. |
| Agent creation | Agent management | Create explicitly within the available quota. A paid order does not itself mean an Agent has been created. |
| Launch | The same Agent record | Open a ready Agent through its account entry. Launch checks ownership, runtime state and plan validity. |
| Delivery review | Your named reviewer | Compare the output with the brief. Publishing or deployment remains a separate human decision. |
Example: review one product page
Illustrative brief, not a customer case or a promised result.
Question: What prevents a German buyer from understanding this product page? Inputs: approved specifications; current page capture; shipping policy; existing German copy. Scope: review supplied evidence; do not invent performance or compliance claims. Requested output: source-linked issues, proposed copy, and unanswered questions. Acceptance: each finding names its evidence; each proposed claim has an approved source. Approval: the store owner reviews all changes before publication.
Recover the existing operation
If an order appears paid but access is not active, retain its reference and seek a status check rather than paying again. If creation times out, keep the existing Agent record and use its recovery action when offered.
An uncertain pause, resume or deletion should be reconciled through the same record. Do not create a replacement merely to force a different status. Runtime availability also depends on the configured provider.
Close the brief before reusing it
Ask the Agent to distinguish observations from proposed actions. Check quotations, calculations and claim sources yourself. Record what you accepted, rejected or could not verify, then retain the input version and reviewed output together.
For the next run, update changed inputs and reopen unresolved questions. Available file inputs and tools depend on the deployed workspace. Prepare text extracts where an input format is unsupported.
Use a small evidence ledger in your own notes: finding, source location, proposed change and review decision. It should be possible for another person to follow a finding without reconstructing the whole conversation. If the Agent cannot access a source or complete a check, keep that item open and name the missing input.
Questions about repeatability
Does reusing a brief schedule future runs?
No. Reuse here means an operator starts another run with updated inputs. The personal center has no task center or frontend task scheduler.
Does an active Agent mean the work is approved?
No. Agent status is a management signal. Check readiness and actual workspace access separately. An active label does not prove current reachability, output quality, permission to publish or a successful storefront change.
Check the Agent before starting the brief
Review your available quota and Agent state. Create or open the appropriate workspace only after access is confirmed.
Open Agent management