원클릭으로
web-works
Fast website understanding and execution skill for LLM agents with strict verification and optional JSON task contracts.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
메뉴
Fast website understanding and execution skill for LLM agents with strict verification and optional JSON task contracts.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
| name | web-works |
| description | Fast website understanding and execution skill for LLM agents with strict verification and optional JSON task contracts. |
Use this skill when user gives website work and expects fast, accurate execution.
This skill standardizes:
Activate for requests like:
Two valid input modes are supported.
Minimum required:
Use:
.github/skills/web-works/web-task.schema.json.github/skills/web-works/web-task.template.json.github/skills/web-works/PROMPTS.mdIf JSON is provided, validate first and execute exactly by steps.
Use execution_profile from task JSON:
balanced: default mode (understand -> execute -> verify)deep: force deeper analysis before risky/ambiguous stepsturbo: after understanding is clear, reduce tool calls for speedIf task becomes ambiguous or fails repeatedly, switch to deep behavior temporarily even when profile is balanced or turbo.
Use file-backed checkpoints so a new chat can continue from where previous chat stopped.
resume.state_file from task JSON..playwright-mcp/resume-state.json.task_id matches, resume from last_completed_step_id + 1.start_url.about:blank).task_idlast_completed_step_idcurrent_urlupdated_atstatus (running/blocked/done)status=done and final evidence summary.For each step:
Never claim done without evidence.
Honor these JSON toggles when provided:
reasoning.understand_firstreasoning.auto_escalate_deep_researchreasoning.max_retries_per_stepresponse_quality.fix_grammarresponse_quality.strict_formatresponse_quality.concise_step_reportsRun this before heavy actions:
If task includes files/commands, run this before edits/execution:
Do not run write commands before this pass.
Use selector fallback order:
data-*, id, name)Avoid generic getByText("Search") as first selector on dynamic apps.
Use concrete checks:
Require explicit confirmation before:
For messaging apps, draft first and ask before send.
When site_profile is whatsapp-web:
https://web.whatsapp.com.Use web-task.template.json and set:
site_profile to whatsapp-webstart_url to https://web.whatsapp.comside_effect_policy.auto_send_allowed to false unless user explicitly requests auto-sendUse:
Action:Tool:Verification:Next:Also include:
Understanding: one-line state summary before first action.Done only when: