Coach Loop

A personal career coach you can run in Claude Desktop, Claude Code, Conductor, or ChatGPT.

Set goals → plan your week (Monday) → check in (Friday) → repeat. Personality assessments personalize the coaching. Pattern detection surfaces drift before it becomes a crisis.

What you get

• 6 coaching flows — career-profile, enneagram, emotional-intelligence, goals, week-plan, coach-checkin
• Local data, not a SaaS — your goals, systems, and session log live in plain Markdown files on your machine
• Pattern detection — catches goal proliferation, goal decay, system decay, missed cadence, and more
• Systems lifecycle tracking — every practice tagged untested → working → not working → retired

Pick your platform

Platform	What you need	Persistence	Setup difficulty
Claude Code	the claude CLI	direct filesystem reads/writes	★
Conductor	Conductor app	direct filesystem (via Claude Code under the hood)	★
Claude Desktop	Claude Desktop + Filesystem MCP + (optional) Google Calendar MCP	direct via MCP	★★
ChatGPT	ChatGPT Plus	upload at session start, paste output back at end (or use sync.sh)	★★★

Then follow the matching INSTALL guide:

• adapters/claude-code/INSTALL.md
• adapters/conductor/INSTALL.md
• adapters/claude-desktop/INSTALL.md
• adapters/chatgpt/INSTALL.md

How it works

core/                          ← source of truth (platform-agnostic)
  flows/                       ← the 6 coaching flows
  rules/                       ← coaching rules, pattern detection, schemas
  question-banks/              ← Enneagram + EI statements and reference data
  templates/                   ← starter files for fresh users

adapters/
  claude-code/                 ← thin shims that map abstract verbs → Claude Code tools
  conductor/                   ← points to claude-code (Conductor wraps it)
  claude-desktop/              ← .skill bundle + MCP config example
  chatgpt/                     ← Custom GPT instructions + knowledge files + sync.sh

scripts/
  build.sh                     ← regenerates desktop bundle and chatgpt knowledge from core/
  validate.sh                  ← lints adapters for drift and missing references

The core/ directory is the only thing you should edit. Run ./scripts/build.sh after changes to refresh the per-platform bundles.

The core loop

1. career-profile (once) — personality + strengths assessment, sets coaching cadence
2. goals — define SMART goals (3-month, 1-year, 3-5 year)
3. week-plan (Monday) — alignment review of last week, then forward planning with calendar blocks
4. coach-checkin (Friday) — commitment review, retrospective on misses, systems classification, new commitments
5. Repeat

Files in your data directory

File	Purpose
goals.md	SMART goals
systems.md	Repeatable practices, lifecycle-classified
career-profile.md	Personality results, cadence, session count
wins.md	Accomplishments + "what worked" notes
not-now.md	Deliberately parked ideas
coaching-log/<date>-<flow>.md	One file per session
coaching-log/summary.jsonl	Machine-parseable session index

See core/rules/data-schema.md for exact formats.

Privacy

Your coaching data lives only on your machine. The flows never send it anywhere except to the LLM you've chosen to talk with. ChatGPT users: review your data-controls settings before pasting goals or wins, since OpenAI may use chat content for model improvement depending on your plan.

License

Apache 2.0 — see LICENSE.