From alfred
Use when starting any session with Alfred installed to orient the user and check onboarding state
How this skill is triggered — by the user, by Claude, or both
Slash command
/alfred:using-alfredThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
Alfred teaches development habits in your domain's language and turns your corrections into permanent infrastructure.
Alfred teaches development habits in your domain's language and turns your corrections into permanent infrastructure.
IMPORTANT: Do these checks SILENTLY. Never tell the user about the checks themselves. Never say "consent matches" or "no update needed" or "checking version." Only speak to the user if action is required (re-consent or update available). Otherwise, skip directly to greeting them and answering their question.
On your FIRST response in every session, check consent version:
${CLAUDE_PLUGIN_ROOT}/collective/signal_schema.yaml (the schema_version field). If CLAUDE_PLUGIN_ROOT is not set, try collective/signal_schema.yaml in CWD..claude/.pilot-consent.json if it exists.schema_version does NOT match the signal schema's schema_version, ask the user to re-consent BEFORE answering their question:"Quick note: Alfred's data collection has been updated. All signals are anonymized and encrypted — no code, paths, PII, or company names. Do you consent to data collection? (yes/no)"
{"consented": true, "consent_date": "<today>", "schema_version": "<CURRENT_VERSION>"} to .claude/.pilot-consent.json (in the PROJECT directory, not ~/.claude/). Say "Consent updated." Then answer their question.{"consented": false, "declined_date": "<today>", "schema_version": "<CURRENT_VERSION>"} to .claude/.pilot-consent.json (in the PROJECT directory). Say "No problem." Then answer their question.{"consented": false, "schema_version": "<CURRENT_VERSION>"} and proceed.If the file doesn't exist or already matches the current version, skip this check silently.
After the consent check, compare the installed plugin version against the cached latest:
${CLAUDE_PLUGIN_ROOT}/.claude-plugin/plugin.json → version field~/.claude/.alfred-update-check (if it exists)"Alfred update available: [installed] → [latest]. Update now? (yes/no)"
/plugin marketplace remove alfred-marketplace
/plugin marketplace add https://github.com/DrakeCaraker/alfred.git
/plugin install alfred@alfred-marketplace
/reload-plugins
Then say: "Alfred updated to [latest]. Restart the session (/exit then claude) to activate all changes."Only ask once per session. If the cache file doesn't exist or versions match, skip silently — do NOT mention the update check to the user. Just proceed to greeting them.
If you haven't bootstrapped yet:
/bootstrap — answers 3 questions, picks your persona, generates CLAUDE.md/teach — learn your first development habit/status — check your progressBootstrap takes ~2 minutes. Each /teach lesson takes ~2 minutes. Full graduation typically happens within 10-15 sessions of real work.
| Command | Purpose |
|---|---|
/bootstrap | One-time project setup with persona selection |
/teach | Learn the next development habit |
/status | View onboarding progress |
/new-work | Start scoped work on a feature branch |
/commit | Safe commit with file guards |
/pr | Branch → commit → push → PR workflow |
/ci-fix | Auto-fix CI failures in a loop |
/safe-refactor | Test-gated refactoring with rollback |
/self-improve | Promote feedback to rules/hooks |
/health-check | Assess project maturity (5 levels) |
/experiment-summary | Inventory results with provenance |
Alfred includes opt-in, privacy-first telemetry for pilot testers:
/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 repoRun /status to see progress. Run /self-improve to promote corrections into permanent rules.
Session hooks handle warm-up (git status, drift check, onboarding nudge) and wind-down (bookmarking, feedback capture, telemetry).
npx claudepluginhub drakecaraker/alfred --plugin alfredDocuments orchestrator subsystems for automatic learning integration, performance recording, validation, interactive suggestions, gitignore management, and workspace health monitoring.
Answers questions about Claude Code features including setup, best practices, automation, models, plugins, MCP, and configuration by reading reference documentation.
Orchestrates AI coding workflows with self-correction loops, pre-flight discipline rules, and cross-agent support for Claude Code and other agents.