Weapp acceptance

Keep planning and code edits in the host agent. Weapp Agent executes checks and collects evidence; it does not need another model or API key.

1. Call weapp_project_inspect when available, or wv accept <project> --inspect --json. Limit acceptance claims to supported weapp-vite native/Wevu projects on WeChat.
2. Maintain acceptance.scenarios in weapp-acceptance.config.json for the requested behavior. Read scenarios.md when creating or changing a scenario. Preserve unrelated checks and user changes. Choose assertions from the requested outcome, not from whichever output the current code produces.
3. Start weapp_acceptance_start, retain its jobId, and query weapp_acceptance_status until it is no longer running. Use CLI wv accept <project> --json if MCP is unavailable; CLI waits for completion. Do not launch duplicate jobs to poll progress.
4. Read weapp_acceptance_report and relevant named artifacts. Repair the cause of failures within the user's task, then run a new acceptance job. Repeated environment failures require resolving the stated prerequisite, not repeated navigation or input.

Honor existing host authorization. A tool's action_required result does not authorize silently adding --trust. Explain the concrete scripts/configuration that need review. Scenario edits invalidate the stored trust decision. Never lower required checks, delete assertions, or replace actual values with expected values to obtain a pass.

Reports use version: 2. Claim acceptance passed only when passed is true and snapshot.stale is false; state the report's requiredChecks. Build-only acceptance is not runtime verification. verify --json is the legacy command report and is not a full acceptance verdict.

Report failures, missing evidence, and environment blockers distinctly. Screenshots support inspection but do not prove visual correctness. runtime: wechat-devtools describes the developer-tool simulator, not a physical device. Interrupted tasks are not replayed: inspect the last recorded step and current state before starting a fresh job.

When reporting completion, provide the job ID, checks performed, failed/unverified categories, and relevant screenshot/log artifacts. Do not claim a runtime action occurred unless the report records it.