TASTESTUDIO user guide

TASTESTUDIO connects project editing, remote tests, failure evidence and approved AI-assisted fixes. The repository/package identifier remains tastedev-studio.

Open a project

Use Project Manager to register an existing local folder. Open its Workspace, connect the intended directory and grant access when prompted. In Explorer, open a UTF-8 text file, edit it and explicitly save. Opening a project or saving Protocol YAML does not execute a test. Review dirty-tab and deletion warnings; cancel preserves your edits.

UI languages match tastedev-web: Korean, English, German, Spanish, French, Italian, Brazilian Portuguese, Japanese, Simplified and Traditional Chinese. System language is the default; unavailable or unsupported languages fall back to English. Manual selection persists. Source, raw logs, user data and AI responses retain their original content. Monaco built-in menus may remain English.

Desktop and Web

The current native desktop implementation targets Windows. Native file operations, terminals/PTY and Git use its Rust runtime and installed tools. Web supports available folder access/editing and connected remote workflows, but local process/PTY and native Git require the desktop runtime. Check actual download metadata for supported release platforms; Agent Windows/Linux support does not imply a Linux Studio installer.

Connect Core and Agent

Core manages work; the independent Rust Agent executes it. Obtain the Core address and Studio authentication configuration from the operator. Verify Agent connection, idle/busy status, OS and runtime/browser capabilities before submission.

For prepared development sources, the existing commands are pnpm core:start and tastedev-agent --config agent-config.json. Core defaults to127.0.0.1:4340. Configure CORE_AGENT_TOKEN and CORE_STUDIO_TOKEN securely; the Agent's TASTEDEV_AGENT_TOKEN must match its Core Agent credential. Agent configuration includes endpoint (ws://127.0.0.1:4340/agent), name, workspaceRoot and tokenEnv. Use the actual executable name supplied by the package. Remote origins, LAN access and TLS require an operator-approved configuration. These commands are not an automatic service installation workflow.

Define and run tests

The .tastedev folder contains required project.yml and optional environments.yml, tasks.yml and tests.yml. Declare executable and arguments separately, save files and resolve Invalid or Dirty states before submission. Protocol declares what to execute; Core selects an eligible Agent and tracks the Run. A local Source edit is not automatically transferred to the Agent.

Select a saved Task/Test, submit to Queue and assign if the current workflow requires it. Inspect the Run's Agent, source revision or snapshot identity, stage logs and final result. The sequential pipeline is source → install → build → start → healthcheck → test → cleanup; optional stages run only when declared. Test is required. Cancellation, timeout, Agent unavailability and queued state are not PASS.

Browser tests use the Agent Playwright/Chromium runner and the project's Playwright tests. Review reports, console/page errors, network failures, screenshots and traces when available. Evidence can be partial or size-limited. Text masking does not guarantee visual redaction of screenshots; use safe test data and review images before sharing.

Analyze and fix

Open a failed Run and choose Analyze Failure. Review the Provider and masked minimum context; explicitly approve external transmission. Review root-cause candidates against actual evidence and files. Codex ChatGPT account connection and the separate OpenAI API provider use different account/authentication paths. For development login use pnpm ai:login and complete the official sign-in flow.

Review Proposed Diff, affected files, rationale and suggested tests. Save or cancel dirty buffers. Approve the exact proposal/scope and apply it, then review actual disk/Git changes. Execute approved saved Protocol validation or remote retest. A new proposal requires a new approval. The current GUI applies a selected proposal file; service multi-file support does not grant project-wide authority.

Revert AI Changes restores the pre-AI baseline, preserving earlier user changes. Further user edits or a stale base produce PATCH_CONFLICT instead of forced overwrite. The default retry cap is3 attempts. A model's claim is not fix success: verify the actual Test result. There is no unrestricted AI shell or automatic commit/push.

Uncommitted fixes can use small text Workspace Snapshots: up to100 files/24KiB source content, with an additional RPC limit. Secrets, credential paths, dependencies, build/cache and binaries are excluded. Required excluded files and oversize snapshots fail explicitly. Secret injection and arbitrary full-project transfer are not provided.

Issues and schedules

Create an Issue Candidate from a failed Run, review target/title/body and references, run duplicate search, then link an existing Issue or approve creation. An uncertain create response must be reconciled before retrying. Issue candidates are not already published. No automatic close/comment or Git writes occur.

Sync saved Protocol, select Project/Test/trigger/timezone, save and explicitly enable a schedule or Run Now. Manual, cron, interval and one-time triggers are supported. Core must remain alive; missed times are not backfilled and overlapping Project/Test jobs are limited. Cron/interval minimum60 seconds, active Job cap16 and maximum100 schedules apply. Scheduled failure does not automatically authorize AI fixes or GitHub publication.

Troubleshooting

Access Required: connect the intended folder and grant browser access. Cancelled picker/AbortError: selection did not complete; retry the folder selection. Invalid Protocol: repair saved YAML. Queued: compare Agent capabilities and availability. Desktop required: use desktop or remote execution. PATCH_CONFLICT: review current Source and generate a fresh proposal. Snapshot failure: inspect limits, exclusions and checksums. Provider errors: distinguish authentication/rate/context/timeout from a Source test failure.

Core execution history and registered Protocol remain memory-based. Schedule definitions persist, but Protocol must be resynced after restart. Full durable recovery, automatic service setup, team roles and cloud synchronization are planned enhancements, not current product promises. Confirm release availability and installed-platform qualification separately.