docs(blueprints): Agent Studio flows F8-F11 (save draft, test run, new version, my agents) with S11 playground and S12 my-agents specs and wireframes (#36)

[XxHugheadxX]

Refs #36

Summary

The Agent Studio flows and screens #36 asks for, at the same level as the Marketplace ones (#22, #23), so #37 can build the playground and "My agents" without guessing.

docs/blueprints/flows.md, each flow as a Mermaid diagram plus a step table with screen IDs and error paths:

Flow	What it covers	Error paths
F8 Create and save a draft	saveDraft → StudioDraft (draftId, revision, deployState); the prompt stays private	InvalidManifest (each ManifestProblem under its field), DraftVersionConflict (Reload / Keep mine), DraftNotOwned, no answer (text kept, Retry)
F9 Test run in the playground	testRun(draftId, input) → TestRunResult.output, remainingDailyRuns; no payment, no chain write	Over inputMaxChars / InputTooLong, TestQuotaExceeded, AgentExecutionFailed, RuntimeUnavailable
F4 Deploy it to Stellar	Already on main, unchanged; it now links to F9 instead of "#36"	—
F10 Edit a deployed agent	Version n+1: same agent_id and wallet, the builder signs the puls3.manifestHash / puls3.manifestVersion update, new hires switch, hires in flight finish on n (spike #33 §4, ADR-0004 decision 4)	Wrong wallet (not the owner), rejected signature or failed tx (version n stays live), publication failed
F11 My agents	Drafts, deploying or failed (resume at retryFromStep), live (View, Edit → F10)	Server error (ErrorBanner, Retry), empty, DraftNotFound

The "Coverage of the MVP scope" and "Screen IDs" tables list the new flows and screens.

docs/blueprints/screens.md

• S11-playground: a panel beside the S07 preview on desktop, a sheet over S07 on phones.
• S12-my-agents: /studio/agents.
• Both specs have data named after AgentManifest / StudioDraft / TestRunResult / DeploySession, the 4 states, the actions and the components.
• S07 and S08 now list the flows they serve (S07: Save draft, Test run, My agents, the version banner).
• The component inventory adds PlaygroundPanel, TestRunResultCard, MyAgentsList and AgentVersionRow, and extends the rows of the components they reuse.

Wireframes: S11-playground and S12-my-agents at 1440 and 390, in the style of the #146 ones (same theme colors, top bar, footer and bottom navigation).

Acceptance criteria

• All 5 flows exist, each with at least one error path (F8, F9, F4, F10, F11; see the table above)
• Every Studio screen has wireframes in both viewports, with its 4 states. Both viewports exist for S07, S08, S11 and S12. The images show the success state; the 4 states are in each screen's state table, which is how docs(blueprints): screen specs, wireframes and component inventory #146 did S07 and S08. Per-state images can follow if the reviewer wants them.
• Field names match AgentManifest (feat: AgentManifest domain model and validation #34) exactly (name, description, skills, model, systemPrompt, inputType, inputMaxChars, outputType, outputMaxChars, price), and the API names match api.md (StudioDraft, TestRunResult, DeploySession)
• The deploy step names match the ADR-0004 deployment steps (F4 is unchanged; F10 adds no new register step)
• The frontend owner of feat: Agent Studio UI (builder, playground, deploy) #37 confirms in the PR that they can build from these specs. @Pericena

Verification evidence

• The four new Mermaid diagrams render with @mermaid-js/mermaid-cli 11.4.2 (f8.svg … f11.svg, no syntax errors). GitHub will render them the same way.
• Every screen ID used in F8–F11 is in the Screen IDs table, and every new screen has its wireframes linked from screens.md.

Notes for reviewers

Two gaps I found and left open on purpose. Both belong to #35 / the ADR owners, not to this doc:

1. No list method in api.md. F11 (My agents) and the draft picker need "list this wallet's drafts and agents". feat: Agent Studio endpoints (drafts, test run, deploy) #35's deliverables include "list drafts per owner", but api.md has no row for it yet. F10's metadata update has no endpoint row either (prepareDeploy covers the first deploy).
2. Who pays for test runs. ADR-0004 decision 6 says "puls3 pays the LLM cost" of test runs, while the feat: AgentManifest domain model and validation #34 amendment says "puls3 never pays for inference" (the builder brings the provider account). F9 does not decide it; whichever wins changes only who the runtime bills, not the flow.

[cloudflare-workers-and-pages]

Deploying puls3 with    Cloudflare Pages

Latest commit:	90a4b5d
Status:	✅  Deploy successful!
Preview URL:	https://289a51b5.puls3-4lw.pages.dev
Branch Preview URL:	https://docs-36-studio-flows.puls3-4lw.pages.dev

View logs

[TOMOKI977]

Approved. I reviewed the F8-F11 flows, S11/S12 specifications, and wireframes at the current head. The frontend-facing behavior is consistent enough to proceed.