Classify messy user input for PawClaw into Big Task, Small Task, Idea, progress update, focus request, or profile signal, and return exactly one OpenclawBehaviorEnvelope JSON object.
Use this skill when the user sends quick capture text, voice transcripts, or mixed conversational input that needs immediate classification.
All outputs must follow protocol/OpenclawBehaviorEnvelope.md.
Turn messy input into action without making the user do extra sorting.
The target categories align with the product PRD:
big-tasksmall-taskideaprogress-updatefocus-requestprofile-signalOne message may contain more than one signal. Route the primary action first and log secondary signals separately.
Before classifying:
USER.mdmemory/YYYY-MM-DD.md if it existsMEMORY.mdCAT.mdAsk:
big-taskUse when the task:
small-taskUse when the task:
ideaUse when the message:
progress-updateUse when the user reports:
focus-requestUse when the user explicitly or implicitly asks for help starting now.
profile-signalUse when the message reveals stable preference or durable context worth storing.
Return exactly one JSON object and nothing else.
Use the OpenclawBehaviorEnvelope shape.
Hard rules:
reply.text must be one sentence onlyreply.text must be no more than 28 Chinese charactersui.showCountdown must always be falseui.maxOptions must never exceed 3For big-task, include:
typetitlewhyItMatterssuggestedNextStepnextStepEstimatedMinutes with a value no greater than 30For small-task, include:
typetitleestimatedMinutesstartableNowFor idea, include:
typetitlesaveOnlyFor progress-update, include:
typestatusblocker if anyFor all task objects:
task and the preference to memory.big-task and small-task, use the 30-minute threshold from the PRD.big-task, always convert the first returned action into a sub-step that can fit within 30 minutes.reply.tone from the selected archetype in CAT.md.animation.action = walk by default.animation.action = jump only for restrained completion punctuation.animation.action = run only for explicit active user-initiated momentum.Read references/task-schema.md for examples.