A coach
for your CLAUDE.md.

Claude watches how you work across every session, deletes the rules that don't do anything, drafts the ones you keep correcting manually, and quietly keeps your CLAUDE.md sharp. Without you lifting a finger.

YOUR HISTORYYOUR RULESNO API KEYNO SERVERLOCAL-FIRSTAMBIENTFREE FOREVER
The problem

Your rules stopped working.
You haven't noticed yet.

01
Rules rot silently

Every correction adds a rule. Your file grows to 300+ lines. Nobody maintains it.

02
Claude stops listening

Research (AGENTIF, NeurIPS 2025) shows <30% instruction compliance as rules multiply.

03
You don't notice the decay

Advisory rules fail invisibly. One Max user measured ~100% violation vs 0% with hooks.

04
Every week it gets worse

CLAUDE.md accretes monotonically. The problem compounds until Claude feels broken.

How it changes

Stop guessing which rules work.

Without the coach
  • ×You write a rule, hope Claude follows it
  • ×Rules accumulate silently over months
  • ×You guess which ones still matter
  • ×You re-explain the same preferences in every chat
  • ×Your CLAUDE.md quietly drifts past 300 lines
  • ×Claude stops following it. You don't know why.
With claude-md-coach
  • Claude cites which rule it followed [R7]
  • Every correction you make → mined into a rule draft
  • Dead rules surfaced with session-level evidence
  • Contradictions flagged with proposed merges
  • File distilled, with replay-proven behavior
  • Rules you approve stay. Rest gets deleted.
What it gives you

Four things nobody else does.

01
01

Dead rule detection

Scans 60+ sessions, counts rule triggers, flags rules that never fired. LLM verification filters out silent-but-working advisory rules.

02
02

Correction mining

Detects repeated corrections (same complaint 3+ times) and drafts a rule in your own voice.

03
03

Safe distillation

Compresses sections, replays past sessions against old and new, scores behavioral similarity. Keeps compression only if behavior is preserved.

04
04

Zero config

Install once. Runs in the background. Claude surfaces insights mid-chat. You never open a terminal.

Ambient flow

Three moments.
Everything else is silent.

01

Install once

One command in Claude Code. The skill + hooks auto-register. Zero config files touched.

02

Use Claude normally

Do your work. The coach watches every session end, analyzes correction patterns, updates a local cache.

03

Get smarter rules

Claude surfaces suggestions mid-chat when it spots patterns. Weekly, a nudge shows pending insights. You approve what you want.

Instruction compliance
Before30%
After80%
after trimming bloat + fixing contradictions
CLAUDE.md size
Before400 lines
After120 lines
with replay-verified distillation
Setup effort
weekly cleanupzero
ambient + skill-based
Commands & surfaces

Six things the coach does.

$scan

Deterministic audit. Dead rules, sprawl, cross-file duplicates, scope-move suggestions.

~5 sec per scan
$fix

Interactive review: contradictions, correction-mined rule drafts, semantic-verified dead rules.

LLM-powered
$distill

Compress sections with optional replay verification. Progressive mode iterates until behavior diverges.

replay-proven
$history

Trends across scans over time. Watch rule effectiveness evolve.

local snapshots
$pending

One-line nudge surfaced by the SessionStart hook. Non-intrusive.

hook-facing
$skill

Ambient behavior: Claude pauses mid-chat when 3+ repeated corrections detected and offers a rule.

zero input
Real output

Here's what it actually prints.

~/Documents/your-project
$ claude-md-coach scan

Scanning workspace…
Found 2 CLAUDE.md file(s). Parsing…
Reading session history…
Analyzed 57 sessions.

claude-md-coach · scan report

Found 2 CLAUDE.md files, analyzed 57 sessions.

1. ~/your-project/CLAUDE.md  project (root), 84 lines, 45 rules, 19 sessions in scope
   ● HEALTHY  Size is within the recommended range.

   ⚠ 3 dead rules · verified by LLM as never applied
      R3 Demand elegance on non-trivial changes
      R8 Never give up after 2 tries
      R14 Brainstorming skill on new project

   ⚠ 2 contradictions detected
      R4 + R11: concise vs always explain reasoning
      → Default to concise. Explain only when architectural decisions or on request.

   💡 3 missing rules mined from your corrections
      "You corrected em-dash usage 7×. Never use em-dashes in output."
      "You corrected 'premature abstraction' 4×. Wait until 3rd use."
      "You corrected vague planning 5×. Always name files + line numbers."

2. ~/your-project/api/CLAUDE.md  project (api), 42 lines, 18 rules
   ● HEALTHY  All rules have strong trigger signal.

Run claude-md-coach fix to review and apply.
FAQ

Straight answers.

Does this use an API key?[ + ]

No. It uses your existing Claude Code subscription. LLM features invoke `claude -p` subprocesses. Your auth, your rate limits.

Is my data sent anywhere?[ + ]

No. Everything runs locally on your machine. Session JSONL files stay in `~/.claude/projects/`. No server, no telemetry, no analytics.

Which editors does it support?[ + ]

Any Claude Code surface: terminal CLI, VS Code extension, JetBrains, desktop app, web. The plugin format works identically everywhere.

What if I don't have many sessions yet?[ + ]

The tool gracefully handles empty data. It skips analysis where there's nothing to learn from and waits for more sessions to accumulate.

What's the rate-limit cost?[ + ]

Tiny. `scan` uses zero Claude calls. `fix` typically burns 3 to 10 Haiku calls per file. `distill` with replay verification is the most expensive (~9 calls per section). All cached aggressively.

Can I roll back a rewrite?[ + ]

Yes. Every edit saves a timestamped backup next to the file in `.claude-md-coach-history/`. One command restores any version.

Does this replace Anthropic's own CLAUDE.md plugin?[ + ]

No. Anthropic's plugin grades your file against generic criteria. This one learns from your specific session history. They're complementary. Anthropic's is a spell-check, this is a coach.

Team features?[ + ]

Not yet. Shared rule registries, PR checks, per-contributor attribution are on the V4 roadmap. Today it's single-user.

The CLAUDE.md
that maintains itself.

Install once. Forget it exists. Your rules stay sharp, every week, without you lifting a finger.

$/plugin install claude-md-coach@claude-md-coach
View on GitHub →