claude-coach
Session coaching, reflection, live Sonnet advisor, and 133 curated tips for Claude Code.
    
- Session reflection — auto-extracts learnings, proposes memories, tips, CLAUDE.md patches, and skill patches
- Sonnet advisor — reads your transcript, injects session-specific coaching
- 133 curated tips — sourced from Boris Cherny, Anthropic docs, and community best practices
- Frustration coaching — detects frustration, coaches Claude to diagnose and course-correct (local regex, zero latency)
<details> <summary>Yo dawg</summary> <img src="xzibit.png" alt="Yo dawg, I heard you like Claude" width="300"> </details>
Install
claude plugin marketplace add kam-l/claude-coach
claude plugin install claude-coachNo API keys required. The Sonnet advisor uses the claude CLI directly (your existing Pro/Max subscription).
Quick Start
# Inside Claude Code — one command for everything
/claude-coach:setup install # first-time setup
/claude-coach:setup refresh # re-apply tips after adding commands/skills
/claude-coach:setup customize # list tips, add custom tips, explain plugin
/claude-coach:setup uninstall # remove all traces
# Restart Claude Code after install/refresh/uninstallFeatures
Data flow: 1. Tips (always on) — 133 curated tips rotate in the statusline during tool calls. Zero cost, passive reinforcement. 2. Advisor (opt-in) — Sonnet reads the transcript every N minutes, produces contextual tips (ℹ️). Strong advice (⚠️) is injected directly into Claude's context via additionalContext on your next prompt. 3. Reflection (automatic) — after each session, a detached Sonnet call extracts corrections, confirmations, and feedback patterns. Results are routed to the right target and queued — never auto-applied. Review and redistribute them using /reflect command. 4. Frustration coaching (always on) — local regex detects expletives, blame, "still broken". Injects a coaching directive that makes Claude pause, name the mistake, and state a corrected approach. Zero latency, no API calls.
How It Works
Sonnet Advisor (opt-in)
A detached Sonnet worker reads your session transcript and produces 1-3 tips grounded in what you're actually doing:
ℹ️ Run tests before committing the auth middleware changes
ℹ️ Use /fix — methodical debugging beats trial and error here
⚠️ The retry logic in api.js needs a backoff — ask Claude to add oneWhen the advisor has strong advice (⚠️), it's injected directly into Claude's context via additionalContext. Claude acts on the coaching without you having to relay it.
Session Reflection
After each session ends, a detached Sonnet call analyzes the transcript for:
- Corrections — user told Claude it did something wrong
- Confirmations — user validated a non-obvious approach
- Feedback patterns — "don't do X", "always Y", "I prefer Z"
Each learning is routed to the right target:
| Signal type | Routes to | Example | |-------------|-----------|---------| | Durable fact or preference | Memory file | "User prefers single PRs for refactors" | | Actionable hint | Spinner tip | "Use /fix for methodical debugging" | | New invariant or convention | CLAUDE.md patch | "Statusline aggregator wraps any tool" | | Workflow step or anti-pattern | Skill patch | "Add convergence check to /refine" |
💭 2 pending reflections — /reflect to reviewRun /reflect to review, accept, or discard — human-in-the-loop, never auto-applied.
Frustration Coaching
When the enrichment hook detects frustration signals (expletives, blame, "still broken", "not what I asked"), it injects a coaching directive via additionalContext:
1. Name what went wrong — quote the specific mistake
2. Explain why — wrong assumption? Misread the code?
3. State corrected approach in one sentenceNo apology walls. No repeating the failed approach. No unnecessary clarification questions.
Configuration
// ~/.claude/settings.json
{
"env": {
"CLAUDE_COACH": "1",
"CLAUDE_COACH_INTERVAL": "5"
}
}| Variable | Default | Description | |----------|---------|-------------| | CLAUDE_COACH | 0 | Enable Sonnet advisor + hook injection | | CLAUDE_COACH_INTERVAL | 5 | Minutes between advisor calls | | CLAUDE_COACH_COSTS | 0 | Enable advisor cost in statusline ([$0.05]) |
Advisor cost: ≤$0.05/call (hard-capped via --max-budget-usd). Pro/Max users spend rate-limit budget, not dollars.
Or run /claude-coach:setup install for guided setup — it wires all of this automatically.
Statusline Compatibility
/setup install detects your existing statusline and adapts:
| Your statusline | What happens | |-----------------|-------------| | None | Registers statusline-tips.js (tips + pending reflections counter) | | ccstatusline, claude-hud, or other tool | Generates statusline-aggregator.js — wraps original command, appends coach output | | Custom script already integrating coach | Detects existing integration, no changes | | Prior claude-coach version | Updates path to current version |
Tip Categories
| Category | # | Examples | |----------|---|---------| | Workflow | 28 | Plan mode, /rewind, fan-out, "choose and commit", action-explicit phrasing | | Context | 24 | 200-line limit, /compact at 50%, data-at-top/query-at-bottom, quote grounding | | Agents | 19 | Test time compute, "say use subagents", pipeline gates, curb overuse | | Hooks | 12 | exit 2 feedback, route permissions to Opus, Stop hook to nudge | | Quality | 30 | Positive framing, add WHY, self-correct loops, "grill me — no PR until I pass" | | Performance | 20 | /sandbox (84% fewer prompts), CI budget caps, Opus with thinking |
User-Facing Commands
| Invocation | Type | What it does | |------------|------|-------------| | /claude-coach:setup | Skill | Install, uninstall, refresh tips, or customize | | /claude-coach:reflect | Command | Review pending session reflections |
Sources
- Boris Cherny's tips — community-curated best practices from the creator of Claude Code + team (primary source for tips)
- Anthropic prompting best practices — official prompt engineering guide (positive framing, WHY behind instructions, self-correct loops)
- Anthropic strengthen guardrails — hallucination reduction, consistency, quote grounding
- Anthropic skill best practices — skill authoring patterns, progressive disclosure, eval-driven dev
- trigger.dev 10 CC tips — CI safety caps, effort slider, session forking
- Anthropic Claude Code docs
Known Issues
Statusline tips may intermittently show … instead of the tip text. This is a Claude Code TUI rendering bug affecting multi-line statusline output on Windows (#32917). Tips render correctly — CC's display occasionally drops them between refreshes.
License
MIT
