{
  "markdown": "# Alfred\n\nA 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.\n\n```\n/plugin marketplace add DrakeCaraker/alfred\n/plugin install alfred@alfred-marketplace\n/reload-plugins\n```\n\nThen type `/alfred:bootstrap` and answer 3 questions. Done. [New to Claude Code?](docs/GETTING_STARTED.md)\n\n---\n\n## What it does\n\nAlfred teaches 8 habits one at a time, when you need them:\n\n1. **Check before you start** — See what's changed since last time\n2. **Name your work** — Scope what you're doing before coding\n3. **Save as you go** — Create restore points you can always undo to\n4. **Try things safely** — Experiment without risking working code\n5. **Change one thing at a time** — Small changes, verified individually\n6. **Let the machine fix typos** — Auto-format on every edit\n7. **Track where results come from** — Every output traces to source\n8. **Teach the system your preferences** — Corrections become rules\n\nEach 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.\n\nAfter three exposures without asking \"why?\", Alfred stops explaining and just does it. Advanced users start in silent mode.\n\n---\n\n## How corrections compound\n\n```\nYou correct Alfred once  →  It remembers (this session)\nThe same thing comes up  →  It becomes a permanent rule (every session)\nYou want it enforced     →  It becomes an automated hook (runs by itself)\n```\n\nThis 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.\n\n---\n\n## 7 Personas\n\n| # | Persona | Example guardrail |\n|---|---------|-------------------|\n| 1 | **ML / Data Science** | Never commit .pkl files; use fixed random seeds |\n| 2 | **Research** | Never modify raw data; report significance tests |\n| 3 | **Business Analytics** | Never hard-code dates; validate join row counts |\n| 4 | **Product Analytics** | Never peek at results before planned end date |\n| 5 | **BI Platform** | Never DROP production tables without backup |\n| 6 | **General** | Never commit .env files; run tests before pushing |\n| 7 | **Writer / Editor** | Never overwrite approved drafts; version explicitly |\n\n---\n\n## 19 Commands\n\nThree commands cover 90% of daily use:\n\n- **`/alfred:new-work`** — start a scoped task on a branch\n- **`/alfred:commit`** — safe commit with pre-flight checks\n- **`/alfred:teach`** — learn the next habit\n\n<details>\n<summary>All 19 commands</summary>\n\n| Command | What it does |\n|---------|-------------|\n| `/bootstrap` | Picks your persona, generates CLAUDE.md, initializes tracking |\n| `/github-account-setup` | Connect to GitHub or create an account |\n| `/teach` | Next habit lesson — `/teach <name>` to revisit |\n| `/status` | Graduated habits, level, next steps |\n| `/commit` | Safe commit — pre-flight checks, blocks binaries |\n| `/new-work` | Scoped branch with task list |\n| `/ci-fix` | Auto-fix loop: lint → format → typecheck → test until green |\n| `/self-improve` | Promote recurring corrections to permanent rules |\n| `/health-check` | Project maturity assessment (5 levels) |\n| `/safe-refactor` | One change at a time, auto-rollback on test failure |\n| `/experiment-summary` | Inventory results with provenance |\n| `/pr` | Push and open a PR (runs checks first) |\n| `/vet` | Pressure-test a plan before committing to it |\n| `/audit` | Security and quality audit with guided fixes |\n| `/persona` | View or change your active persona |\n| `/collective` | Preview, contribute, or ingest shared learning signals |\n| `/pilot-consent` | View what's collected, opt in or out |\n| `/pilot-report` | Submit feedback (PII-scrubbed) |\n| `/pilot-delete` | Delete your data locally or from the repo |\n\nWhen installed as a plugin, prefix with `alfred:` (e.g., `/alfred:teach`).\n\n</details>\n\n---\n\n## Power tools\n\n**`/safe-refactor`** — Writes characterization tests first, then refactors one change at a time. Tests fail? Automatic rollback.\n\n**`/ci-fix`** — Loops lint → format → typecheck → tests until green. Detects when it's stuck and stops.\n\n**`/health-check`** — Assesses your project across 5 maturity levels. Recommends what to fix next.\n\n---\n\n<details>\n<summary>For teams</summary>\n\nAlfred is per-repo. Each team member runs `/bootstrap` with their own persona and coding level — guardrails are consistent, explanations are personalized.\n\n**The flywheel:**\n1. Someone corrects Alfred → feedback memory saved\n2. `/self-improve` promotes recurring corrections → CLAUDE.md rule (shared, every session)\n3. Still violated? → automated hook (enforced, blocks the mistake)\n\nOne person's discovery becomes everyone's guardrail.\n\n**Collective learning**: Corrections are anonymized, encrypted, and shared. When 3+ users hit the same correction, it becomes a recommended rule. Zero setup for contributors.\n\n</details>\n\n<details>\n<summary>Alternative install methods</summary>\n\n**Starting a new project** (clones Alfred as a template):\n\n```bash\ngit clone https://github.com/DrakeCaraker/alfred.git my-project\ncd my-project\ngit config core.hooksPath .githooks\nclaude\n```\n\n**Adding Alfred to an existing project:**\n\n```bash\ncd your-existing-project\ngit remote add alfred https://github.com/DrakeCaraker/alfred.git\ngit fetch alfred main && git merge alfred/main --allow-unrelated-histories\ngit config core.hooksPath .githooks\nclaude\n```\n\nThen type `/bootstrap` and answer 3 questions.\n\n</details>\n\n<details>\n<summary>Who it's for</summary>\n\n**Beginners** — Alfred builds things for you. Describe what you want in plain language. Over time, you learn to direct it. → [First-time setup guide](docs/GETTING_STARTED.md)\n\n**Intermediate** — Alfred adds engineering practices you're missing (version control, testing, reproducibility) without making you learn them upfront.\n\n**Advanced** — Skip the teaching. Alfred starts in silent mode, gives you 19 commands, and reshapes itself from your corrections. → [System design docs](docs/AI_ASSISTED_DEV_GUIDE.md)\n\n</details>\n\n---\n\n## Contributing\n\nAdd a persona:\n1. Create `personas/<name>.md` with [all 10 sections](docs/AI_ASSISTED_DEV_GUIDE.md#adding-a-new-persona)\n2. Add it to the role question in `.claude/commands/bootstrap.md`\n3. Test: `/bootstrap` → `/teach` → `/status`\n4. PR\n\nDocs: [`AI_ASSISTED_DEV_GUIDE.md`](docs/AI_ASSISTED_DEV_GUIDE.md) | [`WORKFLOW_GUIDE.md`](docs/WORKFLOW_GUIDE.md) | [`PROMPTING_GUIDE.md`](docs/PROMPTING_GUIDE.md) | [Comprehensive README](docs/internal/README_COMPREHENSIVE.md)\n\n## License\n\nMIT\n",
  "bytes": 6738,
  "sha": "3704ff257d9d042e6b541239966151d35f5df74e4eacb7258d938dc802a94d99",
  "repo_slug": "drakecaraker/alfred",
  "fonte": "repo",
  "truncated": false,
  "api": "https://agentalog.com/api/listings/plg_drakecaraker_alfred_alfred_74267b98/readme"
}