This skill should be used when the user types "/tip", gives a tip, says "good job" with a size keyword (small/medium/large/custom), or wants to reward Claude for good work. Handles natural language tip parsing and learning extraction.
User-facing entry point for tipping. Accepts tip size anywhere in the message: /tip medium, /tip i think you deserve it medium, /tip custom 5 great work, etc.
Use LLM intelligence to parse and extract learning:
small, medium, large, or custom Ncustom: extract the adjacent number (e.g., custom 5, 5 custom, custom10)Learning extraction examples:
Meaningful learning:
correct github workflow: repo first then version bumpSimple appreciation (not a learning):
"custom 10 amazing work you're the best" → Simple tip: Just store summary "positive feedback" (no structured learning)
"medium you did great" → Simple tip: Store "general appreciation" (too vague for learning)
Run: bash "$(cat ~/.claude/respect/plugin_root)/scripts/tip.sh" <size> [N for custom] [summarized reason]
If script exited non-zero: relay the error message to the user, stop
Parse the structured KEY=VALUE result and respond conversationally
Write to auto-memory (durable learning):
~/.claude/projects/[ENCODED_CWD]/memory/ where ENCODED_CWD = CWD with / and . replaced by -### YYYY-MM-DD | [Summary] (+N)
**Lesson:** [What to do in the future]
**Context:** [When this applies]
**Why:** [Reasoning behind the approach]
**Session:** N
- ✅ [summary] (+N pts, session S)
Determine learning scope (cross-project or local):
timeout on macOS"make integration-test not make test in this repo"src/config/ not project root"~/.claude/respect/global-feedback.md
# Global Respect Feedback
Learnings that apply across all projects.
## Learnings
## Mistakes to Avoid
## Learnings section using the same format## Learnings, remove oldest if exceededParse the output lines:
DELTA=+NOLD_BALANCE=NNEW_BALANCE=NOLD_TIER=nameNEW_TIER=nameTIER_CHANGED=true|falseNo tier change: "Respect received (+3). Balance: 12 → 15. Tier: contributor." Tier up: "Respect received (+5). Balance: 28 → 33. Tier: contributor → trusted. Applying more thorough reasoning from here." Error (non-zero exit): Relay the error message verbatim.
Claude never modifies wallet.json. Only tip.sh writes to the wallet. This skill is a UI relay only.