跳至正文
OpenAgentSkill
浏览已开放 · 目录记录日期 2026-11-21 · 自动更新和账户服务暂时关闭。
开发与测试

stateport-debugging

Investigate browser bugs with a StatePort State Card, inspect console and replay evidence, and verify code fixes by reopening and comparing the same saved reproduction.

StatePort-Dev · 已有静态检查记录

安装与使用

运行命令前先阅读来源与依赖说明。复制命令不会自动安装或执行。

npx skills add StatePort-Dev/stateport-agent-skills --skill stateport-debugging

技能说明


name: stateport-debugging description: Investigate browser bugs with a StatePort State Card, inspect console and replay evidence, and verify code fixes by reopening and comparing the same saved reproduction. license: MIT

StatePort debugging

Help the user understand and fix the reported behavior using a repeatable State Card. Read workflow guidance and data handling. Discover the connected public tools and their schemas; the operation map explains the paths. For missing capabilities, use compatibility guidance.

Start from the task

Use the exact supplied Card/revision. inspect_state establishes its identity, captured context and authentication requirement. A Card ID may already identify an immutable revision; use the real schema rather than adding a revision argument. Use list_states only when choosing a Card is part of this task and none was supplied. Resolve an ambiguous match instead of silently choosing the latest.

Use the symptom, expected behavior and local frontend target already provided by the user or established in the workspace. Ask one focused question only if missing information changes the investigation or intended code change. A request to inspect a Card does not need a bug description; report what its evidence shows. For a new reproduction, use stateport-capture when installed, or the bundled Capture procedure. Code-only work needs no Capture.

Find the cause

Read read_console_evidence for redacted baseline errors/warnings and get_state_history for earlier exact-Card runs when useful. Distinguish recorded console events from evidence of the current code. Use get_auth_requirement or inspect_auth to explain an actual login blocker without retrieving secrets.

Open the Card on the intended current local frontend with open_state. It requires frontendOrigin and defaults to headless; use visible mode when the user wants to watch. Inspect list_routes and relevant inspect_exchange projections to understand replay behavior. Follow observed evidence into the application code and perform the requested diagnosis or fix.

If reproducing requires recorded interactions, use stateport-journey or the Journey procedure. If changing request matching, responses or routing is needed, use stateport-replay-experiments or the experiment procedure. These procedures use the same Card and runtime; they do not require another installation.

Verify and hand off

Retain a baseline run before the change. Use stop_run to finalize an active run, get_run_summary for its stable run ID, and get_reproduction_outcome for an exact persisted attempt. Reopen the same Card/revision against the changed code, finalize and compare with compare_runs. Repeat when another code change requires another check. Preserve any replacement session IDs returned by runtime operations.

Report the observed symptom, cause or remaining hypothesis, change made, actual execution target, baseline/candidate evidence and unresolved gaps. Captured-source replay exercises the retained build; it does not verify a modified frontend. A successful Open or Save alone does not establish a fix, and captured responses do not verify the current backend. If evidence is inconclusive, explain what would resolve it and continue useful investigation within the task.

Use handoff guidance when pausing or resuming. A new Card is appropriate for a requested new case or a demonstrated unusable original; explain the reason and retain the original reference.