Alfred
A Claude Code plugin that teaches development habits in your domain's language — then fades as you learn. Your corrections become permanent rules; your rules become automated hooks.
/plugin marketplace add DrakeCaraker/alfred
/plugin install alfred@alfred-marketplace
/reload-plugins
Then type /alfred:bootstrap and answer 3 questions. Done. New to Claude Code?
What it does
Alfred teaches 8 habits one at a time, when you need them:
- Check before you start — See what's changed since last time
- Name your work — Scope what you're doing before coding
- Save as you go — Create restore points you can always undo to
- Try things safely — Experiment without risking working code
- Change one thing at a time — Small changes, verified individually
- Let the machine fix typos — Auto-format on every edit
- Track where results come from — Every output traces to source
- Teach the system your preferences — Corrections become rules
Each habit is explained using your domain's language. An ML scientist hears "checkpoint your experiment." A researcher hears "sign a lab notebook page." A writer hears "seal a draft." Same concept — framing that sticks.
After three exposures without asking "why?", Alfred stops explaining and just does it. Advanced users start in silent mode.
How corrections compound
You correct Alfred once → It remembers (this session)
The same thing comes up → It becomes a permanent rule (every session)
You want it enforced → It becomes an automated hook (runs by itself)
This is the core loop. You're not configuring a tool — you're training an environment. Every correction makes the system better. Over time, good practices become the default and mistakes require effort.
7 Personas
| # | Persona | Example guardrail |
|---|
| 1 | ML / Data Science | Never commit .pkl files; use fixed random seeds |
| 2 | Research | Never modify raw data; report significance tests |
| 3 | Business Analytics | Never hard-code dates; validate join row counts |
| 4 | Product Analytics | Never peek at results before planned end date |
| 5 | BI Platform | Never DROP production tables without backup |
| 6 | General | Never commit .env files; run tests before pushing |
| 7 | Writer / Editor | Never overwrite approved drafts; version explicitly |
19 Commands
Three commands cover 90% of daily use:
/alfred:new-work — start a scoped task on a branch/alfred:commit — safe commit with pre-flight checks/alfred:teach — learn the next habit
All 19 commands
| Command | What it does |
|---|
/bootstrap | Picks your persona, generates CLAUDE.md, initializes tracking |
/github-account-setup | Connect to GitHub or create an account |
/teach | Next habit lesson — /teach <name> to revisit |
/status | Graduated habits, level, next steps |
/commit | Safe commit — pre-flight checks, blocks binaries |
/new-work | Scoped branch with task list |
/ci-fix | Auto-fix loop: lint → format → typecheck → test until green |
/self-improve | Promote recurring corrections to permanent rules |
/health-check | Project maturity assessment (5 levels) |
/safe-refactor | One change at a time, auto-rollback on test failure |
/experiment-summary | Inventory results with provenance |
/pr | Push and open a PR (runs checks first) |
/vet | Pressure-test a plan before committing to it |
/audit | Security and quality audit with guided fixes |
/persona | View or change your active persona |
/collective | Preview, contribute, or ingest shared learning signals |
/pilot-consent | View what's collected, opt in or out |
/pilot-report | Submit feedback (PII-scrubbed) |
/pilot-delete | Delete your data locally or from the repo |
When installed as a plugin, prefix with alfred: (e.g., /alfred:teach).
Power tools
/safe-refactor — Writes characterization tests first, then refactors one change at a time. Tests fail? Automatic rollback.
/ci-fix — Loops lint → format → typecheck → tests until green. Detects when it's stuck and stops.
/health-check — Assesses your project across 5 maturity levels. Recommends what to fix next.
For teams
Alfred is per-repo. Each team member runs /bootstrap with their own persona and coding level — guardrails are consistent, explanations are personalized.
The flywheel:
- Someone corrects Alfred → feedback memory saved
/self-improve promotes recurring corrections → CLAUDE.md rule (shared, every session)- Still violated? → automated hook (enforced, blocks the mistake)
One person's discovery becomes everyone's guardrail.
