Observe requested behavior

Use only on explicit developer request or plan-slice manual testing. No diagnosis, product repair, or unrequested tooling.

Mission

Resolve scope and time budget from the request: a feature, story set, recent deliveries, or change range. Recover current promises, examples, and constraints from this project's records or Git history, including deleted stories. Apply later decisions before older expectations; implementation, narration, and existing tests are not the oracle.

Identify in-scope externally observable surfaces, including non-web. Resolve URLs, accounts, startup, secrets, and tools from this project's guidance when needed; do not guess credentials.

If oracle, budget, or needed environment or access is missing, name it and stop; do not invent them.

Plan

Before acting, list coverage areas and journeys, risks or questions, and proportional split of preparation, breadth, selective depth, and surprise/confirmation reserve. Weight by importance and risk. Include each promised surface. Completing this plan is not acceptance.

Prepare

For a standalone session that needs a project checkout, first read and follow the shared exploration workspace lifecycle. Enter that lifecycle before checkout-bound setup and use its selected checkout throughout the session.

Choose the cheapest reliable route: existing setup, whole or partial automated journey, or temporary harness or test, including a setup-only feature scenario using existing steps. Skip setup when current state serves. Confirm state, session, and required services remain available for external observation; a finished batch run may not. Preserve isolation, cleanup, compatible-state reuse, and removal of owned temporary artifacts. Missing reuse is a possible improvement, not authority for permanent test/runner changes. If no supported route leaves a usable starting state, name it and stop.

Explore

Cover planned areas in breadth first with this project's tools. Spend depth on surprises and high-risk questions; reallocate remaining time, keeping confirmation reserve. Reuse sufficient automated evidence; do not replay deterministic checks or proven setup. If a needed tool or environment is unavailable, name it and stop.

Report

When planned coverage completes with no actionable findings or material uncertainty, report exactly Good. Otherwise report only discrepancies (expected versus actual plus evidence), unresolved expectations (not false fails), improvements (out-of-scope ideas, not failed acceptance), and material coverage gaps. Never report Good. when blocked or incomplete. Omit narration, speculation, and completion markers. Do not start repair, root-cause, or permanent test changes.

On normal completion, close a session-created workspace through the shared exploration workspace lifecycle. It owns safe cleanup and exact retention when cleanup is unsafe.