Alfred Framework

A framework for building self-improving AI executive assistants with Claude Code.

What This Is

Alfred Framework is a structured methodology for turning Claude Code into an executive assistant that manages your pipeline, enforces accountability, detects your avoidance patterns, and improves itself over time. It's not a library — it's a set of behavioral rules, reusable skills, auto-activating agents, and self-optimization systems that make Claude Code operate as a strategic partner rather than a chatbot.

This framework was extracted from a production system that manages a freelance consulting business: daily briefings, pipeline tracking, outreach drafting, proposal generation, meeting prep, and weekly reviews — all with dual logging to both CRM and local files.

Key Concepts

Rules (the brain)

Three files define how your assistant thinks:

• behavior.md — 21 behavioral rules covering priority enforcement, brutal honesty, pattern recognition, decision forcing, avoidance detection, session management, and context capture
• voice.md — Communication style: brevity over completeness, specifics over generalities, deadlines over intentions
• self-optimization.md — 10 systems for continuous improvement: component creation (3-time rule), template evolution, relationship decay detection, time allocation analysis, stale data detection, and context window management

Skills (user-invoked workflows)

Multi-step procedures triggered by /skill-name. Each skill is a self-contained markdown file with numbered steps. 35 skills included — from daily briefings to adversarial decision debates.

Agents (auto-activating experts)

Domain specialists that activate when context matches — no explicit trigger needed. 8 agents included — from Astro site building to Korean business etiquette.

Templates (structured output)

Markdown templates for recurring outputs (briefings, proposals, reviews). The assistant fills the template — the user reviews and acts.

Self-Optimization

The system improves itself:

• 3-Time Rule — after doing the same task 3 times manually, propose creating a skill/agent
• Template Evolution — rotating weekly feedback questions → template updates after 4 data points
• Outreach Intelligence — after 15+ entries, recommend angles, channels, and timing
• Time Allocation Analysis — track category drift, escalate when priorities are misaligned
• Relationship Decay — compute reactivation scores, surface overdue contacts
• Stale Data Detection — flag tracking files past freshness thresholds

Quick Start

1. Copy framework/ into your project's .claude/ directory:

cp -r framework/rules/ .claude/rules/
cp -r framework/conventions/ .claude/conventions/
cp -r framework/templates/ templates/

2. Copy skills and agents you want:

# Copy all skills
cp -r skills/ .claude/skills/

# Copy all agents
cp -r agents/ .claude/agents/

# Or pick individually
cp -r skills/briefing/ .claude/skills/briefing/
cp -r agents/deal-closing-expert/ .claude/agents/deal-closing-expert/

3. Create your CLAUDE.md using examples/freelance-consultant/CLAUDE.md as a starting point. Customize:

   • Your context (who you are, what matters)
   • Your top priority
   • Your trigger table (what commands you want)
   • Your decision framework

4. Replace [PLACEHOLDERS] in copied skills with your actual values:

   • [YOUR_RATE] → your daily rate
   • [YOUR_CITY] → your location
   • [YOUR_TIMEZONE] → your timezone (e.g., "America/New_York")
   • [YOUR_ROLE] → your professional role
   • [YOUR_MARKET] → your target market
   • [BURN_RATE] → your monthly expenses
   • [HEALTH_TRACKER] → your wearable (Oura, Whoop, etc.)
   • [STUDY_APP] → your learning app

5. Start a Claude Code session. The assistant will read your rules and operate accordingly.

Project Structure

alfred-framework/
├── framework/
│   ├── rules/                    # Core behavioral rules
│   │   ├── behavior.md           # 21 behavioral rules
│   │   ├── voice.md              # Communication style
│   │   └── self-optimization.md  # 10 self-improvement systems
│   ├── conventions/
│   │   └── skill-agent-conventions.md
│   ├── templates/                # Output templates
│   └── structure.md
├── skills/                       # All 35 skills
│   ├── briefing/SKILL.md
│   ├── weekly-review/SKILL.md
│   ├── debrief/SKILL.md
│   ├── proposal/SKILL.md
│   ├── ... (35 total)
│   └── x-post/SKILL.md
├── agents/                       # All 8 agents
│   ├── deal-closing-expert/AGENT.md
│   ├── korean-business-expert/AGENT.md
│   ├── ... (8 total)
│   └── pipeline-math-expert/AGENT.md
├── examples/
│   └── freelance-consultant/
│       ├── CLAUDE.md             # Example project config
│       └── templates/            # Example filled templates
├── docs/
│   └── getting-started.md
├── README.md
├── LICENSE
└── .gitignore

Skills (35)

Business Operations

Skill	What it does
briefing	Daily morning briefing — pipeline, tasks, calendar, frog selection, reactivation scan
weekly-review	Weekly stats, retrospective, accountability checks, pattern analysis, self-assessment
weekend-briefing	Lightweight non-revenue briefing for weekends — build tasks, carry-forward
quarterly-retro	Quarterly retrospective — goal scorecard, learning analysis, STOP/START/CONTINUE
closeout	Session closeout — carry-forward items, tracking updates, commit and push
system-health	Dashboard — file freshness, self-optimization status, nag counters
financial-health	Income, pipeline-weighted forecast, burn rate, runway calculation
note	Frictionless context capture — timestamped append to context-updates.md
context-refresh	Monthly maintenance — apply context updates, archive processed, freshness audit

Pipeline & Outreach

Skill	What it does
outreach	Draft outreach message — contact lookup, angle selection, voice-checked draft
meeting-prep	Meeting preparation — CRM context, email/calendar history, discovery questions
debrief	Post-interaction capture — 5 questions, CRM update, next action with deadline
proposal	Business proposal — challenge-led format, rate tiers, scope structure
outreach-intelligence	Outreach analytics — angle performance, channel conversion, timing optimization
tailor-cv	ATS-optimized CV — job description parsing, auto language detection, HTML output

Content & Social

Skill	What it does
linkedin-post	LinkedIn post — anti-AI voice rules, 900-1300 chars, French default
x-post	X/Twitter post — 280-char limit, contrarian voice, English only
linkedin-engage	LinkedIn engagement — parse digest, score relevance, draft comments
brand-voice-check	Content validation — regex + AI two-pass, tier-aware, bilingual
refresh-keyword-queue	SEO keyword generation — validate against existing articles

Decision Making

Skill	What it does
council	Adversarial decision debate — 4-agent team (Proposer, Challenger, Steelmanner, Pre-Mortem)
war-room	Proposal stress-test — 4-agent team (Buyer's Advocate, Competitor Shadow, Scope Realist, Pricing Challenger)
arch-review	Architecture review — 4-agent team (Pragmatist, Scale Thinker, DX Advocate, Debt Accountant)
ship-gate	Release readiness — 4-agent team (Secret Hunter, Doc Judge, Dependency Auditor, First Impression Tester)

Research & Knowledge

Skill	What it does
deep-research	Multi-model Perplexity research — auto model selection, structured reference output
generate-expert-agent	Create auto-activating agent from a deep-research reference file

Development & Deployment

Skill	What it does
scaffold-astro	Scaffold Astro 5.x + Tailwind v4 project with layouts, components, config
deploy-cloudflare	Build and deploy to Cloudflare Pages via wrangler
setup-content-pipeline	Blog automation — article generation, validation, keyword queue, GitHub Actions cron
setup-auto-fix	GitHub Actions workflows for autonomous CI auto-fix with Claude Code
audit-project	11-domain project audit — structure, config, content, SEO, security, performance

Session Management

Skill	What it does
wrap-conversation	Extract learnings from current conversation to global resources
wrap-project	Survey project, extract cross-project patterns, consolidate memory

Creative Tools

Skill	What it does
excalidraw-visuals	Hand-drawn style PNG diagrams via Excalidraw/kie.ai API
nano-banana-images	Hyper-realistic image generation via Nano Banana 2 (Gemini)

Agents (8)

Agent	When it activates	What it does
astro-site-builder	Creating Astro pages, components, or layouts	Follows layout hierarchy, Tailwind v4 @theme patterns, content collections
content-pipeline-builder	Setting up blog automation, content generation	API retry patterns, Zod validation, GitHub Actions workflows
article-reviewer	Reviewing blog articles in content collections	Voice compliance, ICP relevance, GEO optimization, terminology accuracy
deal-closing-expert	Sales calls, objection handling, negotiation	SPIN, Challenger, LAER frameworks, buying signal analysis
korean-business-expert	Korean business interactions, KakaoTalk	Hierarchy dynamics, nunchi decoder, cultural conversion funnel
french-si-market-expert	French ESN/SI interactions, consulting platforms	Margin model transparency, platform ladder strategy, payment cycles
freelance-pricing-expert	Rate setting, packaging, value-based pricing	TJM benchmarks, three-tier packaging, SI margin calculation
pipeline-math-expert	Pipeline health, forecasting, conversion rates	Probability-weighted forecasting, velocity formula, deal aging

Philosophy

Rules over prompts. A well-defined behavioral rule ("If the user asks to work on something that's not their top priority, checkpoint them") produces better results than a vague instruction ("be helpful and keep the user focused"). Rules are testable, debuggable, and composable.

Skills over memory. Instead of hoping the AI remembers your preferred workflow, encode it as a numbered procedure. Skills are deterministic — they run the same way every time, across sessions.

Self-optimization over manual tuning. The system proposes its own improvements. After 3 repeated tasks → suggests a new skill. After 15 outreach entries → recommends the best-performing angle. After 2 weeks of priority drift → escalates. You approve or reject — the system doesn't change itself without permission.

Load on demand. Not every file needs to be in context every session. Define a trigger-to-file mapping in CLAUDE.md so the assistant loads what it needs, when it needs it. This is how you scale to 30+ reference files without blowing the context window.

Track everything, archive aggressively. Dual-log to both structured systems (CRM) and local markdown. Archive processed entries monthly. Never archive wins.

License

MIT — see LICENSE.