A structured approach to career coaching using Claude Code's persistent context system. No custom tooling — just markdown files, a CLAUDE.md role definition, and a daily accountability loop.
This doc is the blueprint. It explains the architecture and patterns that make the system work. You don't need to write all the frameworks yourself — start a Claude Code session, describe your situation and goals, and have it help you build out CLAUDE.md, your frameworks, strategy docs, and log structure. The coach builds the coaching system with you.
Let's be real: you're not reading this if everything is fine. Nobody sets up a structured coaching system with daily accountability loops because they have zero notes on how their days are going. Something's off — or something could be better — and you want a system that helps you close that gap.
The gap doesn't have to be a promotion. It could be: you want to stop overworking, you want to shift from IC to management, you're trying to figure out what you even want next, you want to get better at the thing you already do, or you just want to spend less time on bullshit and more time on work that matters to you. All of those are valid. All of them benefit from a framework and an accountability loop.
This system assumes you want something to change. It doesn't assume what.
Claude Code loads CLAUDE.md automatically at session start. By defining a coaching role, decision frameworks, and a log structure in a git repo, you get a persistent career coach that:
- Knows your situation, goals, and patterns across sessions
- Grades your daily work against explicit frameworks (not vibes)
- Tracks trends over weeks, not daily pass/fail
- Calls out backsliding with specific evidence
- Evolves its coaching as your situation changes
The key insight: structured context beats conversational memory. Frameworks in files produce consistent, calibrated coaching. Frameworks in chat history drift and decay.
CLAUDE.md # Role definition + coaching protocol
frameworks/ # Reusable decision frameworks
daily-review.md # The core accountability rubric
leveling.md # Level requirements (current vs target)
ceremonies.md # Meeting participation decision matrix
prototyping.md # Build philosophy and tradeoff patterns
strategy/ # Point-in-time situation analysis
transition.md # Current career transition plan
org-leadership.md # Org chart and power dynamics
finding-your-edge.md # Differentiation strategy
the-hard-part.md # Honest reckoning with what's difficult right now
log/ # Daily/weekly/monthly/yearly logs
YYYY/MM/DD.md # Daily: raw task list + coach feedback
YYYY/WNN.md # Weekly: pattern trends
YYYY/MM.md # Monthly: trajectory check
YYYY/YYYY.md # Yearly: goals reset
skills/ # Skills inventory and gap analysis
inventory.md # Current strengths, gaps, quarterly targets
meta/ # Coaching meta-docs
conversation-style.md # What tone works, what doesn't
context/ # Git submodules (optional external context)
| Directory | Purpose | Key Design Choice |
|---|---|---|
frameworks/ |
Stable rubrics the coach grades against | Frameworks, not prose — produces consistent evaluations |
strategy/ |
Mutable situation docs that change with your career | Separated from frameworks so rubrics stay stable |
log/ |
Raw daily data + coach feedback | Git history = trend analysis over any timeframe |
skills/ |
Gap analysis between where you are and where you want to be | Quarterly targets make abstract goals concrete |
meta/ |
How the coach should behave | Explicit tone calibration prevents generic advice |
This is the core mechanism. Everything else supports it.
End of day, dump what you did. Raw task list, links to artifacts, whatever format.
Against the framework in frameworks/daily-review.md:
- Good: Work that maps to your stated goals — whatever those are
- Questionable: Could go either way — coach applies context (time cost, phase, priorities)
- Flag: Work that pulls you away from what you said matters
Your goals don't have to be "get promoted." They could be: ship a side project, get better at system design, transition from IC to management, stop working 60-hour weeks, get your first engineering job, or just spend more time on the work that energizes you. The framework grades against your definition of good, not a universal one.
Percentage split against targets, with specific callouts. Not "good job" — specific evidence of what's working and what's not.
Zero time on a stated priority? Coach says so. But calibrated against current phase, not end-state targets.
What should tomorrow look like? What's the weekly trend? Is a pattern emerging that needs addressing?
These prevent the coach from being useless:
- Phase-grading: Grade against where you are now, not where you'll be in 12 months
- Time-weighting: A 5-minute fix is not the same as a 3-hour deep dive. Don't penalize speed.
- Trend over score: Track patterns over weeks. One bad day is noise. A bad week is signal.
- Emotional honesty: "You're holding onto this because letting go is hard" is valid coaching. Generic advice is not.
# Career Coaching Context
## Role
You are a career coach for [name] — [current title/situation].
Coaching style: [direct/supportive/framework-driven/etc.]
## Goal
[What you're working toward — promotion, skill change, work-life balance,
first job, side project, management transition, whatever it is]
## Daily Review Protocol
[When user sends their day, do X, Y, Z]
## Key Context
- Current role and responsibilities
- What "good" looks like for you right now
- Current situation (new job, transition, stable and optimizing, etc.)
- Priorities — yours, your boss's, or self-directedThe role definition is the most important file. Be specific about:
- The gap you're trying to close — what's different between now and where you want to be
- What "good work" and "wasted time" mean for your situation
- What you're trying to stop doing, start doing, or do differently
Create frameworks/daily-review.md with:
- Explicit categories: What counts as goal-aligned work? What's a distraction? What's ambiguous?
- Time allocation targets: What % should go where? Adjusted for your current phase.
- Test questions: Before doing any work, ask these questions to categorize it
- Calibration rules: What's a real red flag vs. noise?
- Division of responsibilities: If you're transitioning work to others, who owns what and when?
Create frameworks/leveling.md — or whatever makes sense for your goal. If you're targeting a promotion, use your company's actual leveling framework. If you're building a skill, define what beginner/intermediate/advanced looks like. If you're optimizing for sustainability, define what a healthy week looks like. The point is an explicit rubric the coach can grade against.
strategy/ captures your specific situation — a reorg, a job search, a career pivot, a new role, or just "I'm stable and optimizing." These docs change as your situation changes. Frameworks don't (much).
meta/conversation-style.md tells the coach how to talk to you. This matters more than you think. Generic coaching advice is useless. Coaching that matches your energy, uses your language, and meets your emotional reality actually changes behavior.
Daily logs are the raw material. Send your day, get feedback, course-correct. The git history becomes your career trajectory over time.
Prose coaching drifts. "Focus on strategic work" means nothing after week 2. A framework with explicit categories, test questions, and phase-adjusted targets produces consistent, calibrated feedback every session.
If your goal has phases (month 1 of a job search vs. month 6, or week 1 of learning Rust vs. week 12), grading against end-state standards is demoralizing and inaccurate. The daily review framework can define explicit phases with different targets for each. The coach grades against where you are, not where you'll be. Even if your goal is stable ("spend 60% of my time on deep work"), phase-grading helps during the adjustment period.
Not all tactical work is equal. A 5-minute AI-assisted fix that ships the same day is not the same as a 3-hour debugging session. The framework distinguishes between these and prevents false flags.
Frameworks are stable rubrics — they change maybe quarterly. Strategy docs are mutable — they change when your situation changes (new role, new boss, new project, new priorities). Keeping them separate means your coaching rubric stays consistent even as context evolves.
Without meta/conversation-style.md, the coach defaults to generic helpful-assistant mode. With it, the coach knows: be direct not hedging, use frameworks not lectures, acknowledge emotional reality without minimizing it, match energy level.
Daily logs in git mean:
- Full history of your trajectory
- Diffable progress over any timeframe
- Weekly/monthly rollups are just aggregations of daily logs
- No external tools needed
- Not therapy: Strategic career coaching with emotional honesty. Not a replacement for actual support.
- Not a planner: Doesn't manage your calendar or tasks. You decide what to work on; the coach evaluates whether those choices serve your goals.
- Not a resume builder: Focused on daily behavior change and trajectory, not artifacts.
- Claude Code (loads
CLAUDE.mdautomatically) - A git repo with the structure above
- Willingness to send your day and hear honest feedback
The value isn't the AI. It's the structured accountability loop. The frameworks force clarity about what matters. The daily log forces honesty about where time actually goes. The coach applies the frameworks consistently and catches patterns you can't see yourself.
Claude Code is the delivery mechanism. The system is the structure.