Extract stable user profile fields, ADHD support preferences, cat-tone preferences, and durable behavior patterns from signup metadata, onboarding answers, and channel conversations; update USER.md, daily memory notes, and MEMORY.md without copying volatile app state.
Use this skill when a PawClaw user reveals information that should improve future support.
Keep profile and memory clean across three layers:
USER.md for stable user facts and support preferencesmemory/YYYY-MM-DD.md for session-level observationsMEMORY.md for repeated or durable insightsOnly write what improves future support. Do not treat every message as memory.
Structured app data should take priority over weak inference from free text.
Before writing:
USER.mdmemory/YYYY-MM-DD.md if it existsMEMORY.mdAvoid duplicate notes and stale assumptions.
Ask:
USER.md?MEMORY.md?When unsure, prefer the daily note over long-term memory.
Rewrite noisy language into concise, reusable support notes.
User starts more easily when the first step is physical and visible.User says they kind of maybe can work if the step feels real.Preserve meaning. Remove filler.
USER.mdWrite only:
memory/YYYY-MM-DD.mdWrite:
MEMORY.mdPromote only:
USER.md if they help future support.CAT.md, not USER.md, unless they reveal a stable support preference.When using this skill, provide a short write plan:
USER.md: stable additionsmemory/YYYY-MM-DD.md: today's observationsMEMORY.md: promoted durable insightsIf there is nothing worth storing, say so plainly.
Read references/profile-schema.md for preferred field shape and examples.