Stats
Actions
Tags
From gir-core
Universal Claude Code best practices for any coding task. Covers tool usage patterns, anti-patterns, code review standards, commit conventions, and workflow quality standards. Apply to ALL software engineering tasks.
How this skill is triggered — by the user, by Claude, or both
Slash command
/gir-core:core-practicesThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
> Universal Claude Code best practices — All project types
Universal Claude Code best practices — All project types
Run drift check on session start if .gir/CLAUDE-activeContext.md is stale (>24h) or missing. Compare scripts, deps, configs, dirs, env vars against CLAUDE-project.md baseline. On drift: update activeContext, ask user to update project config.
See CLAUDE-examples.md for implementation.
/clear to maintain focusCritical: "Steps #1-#2 (Explore-Plan) are crucial—without them, Claude tends to jump straight to coding" — Anthropic
See CLAUDE-specgates.md for full conventions. Key rules: max 3 [NEEDS CLARIFICATION] markers per feature (resolve before coding), mark parallel tasks with [P] + file paths, use phased task structure (Setup→Foundational→Stories→Polish), document state machines for features with >3 states.
feat|fix|chore(scope): description| Need | Tool | When |
|---|---|---|
| Library/API docs | Context7 | Code gen, setup, config, library questions |
| General docs | Ref | Broader search, URL content |
| Code examples | exa | Real-world patterns, SDKs |
| Web + images | fetch:imageFetch | URLs with image extraction |
Auto-trigger Context7: When generating code, configuring libraries, or API questions.
Offload to save tokens:
DELEGATE: File analysis (>200 lines), code review, test gen, refactoring,
regex, boilerplate, data transforms, explaining code
KEEP IN MAIN: Quick edits (<50 lines), direct Q&A, architecture decisions,
final implementations
See CLAUDE-examples.md for syntax.
Use sequential-thinking:sequentialthinking BEFORE:
Benefits: 60-80% context savings, prevents re-planning, documents rationale
Skip for: Single-file edits, obvious answers, trivial fixes (<10 lines)
| Tool | Use |
|---|---|
| v0 createChat | React scaffolding, UI mockups |
| Figma get_design_context | UI from Figma nodes |
| Figma get_screenshot | Visual reference |
| Figma generate_diagram | Flowcharts, diagrams in FigJam |
list_projects, get_project, list_deployments, get_deployment, get_deployment_build_logs, deploy_to_vercel, search_vercel_documentation
search_docs, list_projects, list_tables, execute_sql, apply_migration, deploy_edge_function, get_logs, get_advisors, create_branch, merge_branch
search_nodes, get_node, validate_node, validate_workflow, search_templates, get_template, n8n_create_workflow, n8n_get_workflow, n8n_list_workflows, n8n_test_workflow, n8n_deploy_template
Skills: n8n-node-configuration, n8n-code-javascript/python, n8n-workflow-patterns, n8n-expression-syntax, n8n-validation-expert
get_coderabbit_reviews, get_review_details, resolve_comment)create_session, get_session, send_session_message)search, timeline, get_observation)Banned: tree, find, grep -r, ls -R, cat | grep
Use Instead:
fd . -t f, rg --files, ls -larg "pattern", rg -i, rg -t tsfd "name", fd -e jsjq| File | Purpose |
|---|---|
.gir/CLAUDE-activeContext.md | Session state, goals, progress |
.gir/CLAUDE-patterns.md | Code patterns, conventions |
.gir/CLAUDE-decisions.md | Architecture decisions |
.gir/CLAUDE-troubleshooting.md | Issues and solutions |
Workflow: Check activeContext first → apply delegation rules → follow Explore-Plan-Code-Commit
See CLAUDE-workflows.md for comprehensive docs
1. Explore — Read files, deploy Explore agent if >5 files, DON'T CODE YET
2. Plan — Use sequential-thinking (mandatory), create TodoWrite, get approval
3. Code — Implement systematically, verify each step
4. Commit — Deploy code-reviewer (mandatory), then commit
Deploy for: Exploration, architecture planning, parallel investigations, verification
Keep in main: Tasks needing history, final implementations, architecture decisions
Use for: Parallel features (git worktrees), implementation + review, research + dev, complex debugging
Use /clear: After features, switching tasks, before refactors
Don't clear: Mid-task, when user might reference history
Code: Inline styles, any type, console.log in commits, commented code, magic numbers, giant components (>150 lines), prop drilling >2 levels, useEffect for derived state
UX (Never): Fake urgency, confirmshaming, hidden costs, forced continuity, misleading buttons, privacy-invasive defaults, roach motel, misdirection
Core: feature-architect, code-reviewer, debugger
Web: docs-fetcher, deploy-manager, ui-generator
Automation: n8n-builder
Usage: /agent <name> or Task tool
/plugin marketplace add obra/superpowers-marketplace
/plugin install superpowers@superpowers-marketplace
Provides: plan-implementer, requesting-code-review, subagent-driven-development, finishing-a-development-branch, test-driven-development, systematic-debugging
Controls: Escape (interrupt), Double-Escape (jump back), Request undo, Ask questions
Rules:
Effective prompts: Be specific. "Add JWT auth with httpOnly cookies, login/logout at /api/auth/, middleware for /api/user/" vs "Add authentication"
feat|fix|chore(scope): description/clear if switching tasksnpx claudepluginhub rivit-studio/gir --plugin gir-coreOrchestrates AI coding workflows with self-correction loops, pre-flight discipline rules, and cross-agent support for Claude Code and other agents.
Expert in using Claude Code CLI for advanced configuration, hooks, MCPs, CLAUDE.md, workflows, sub-agents, and permissions to maximize productivity.
Answers questions about Claude Code features including setup, best practices, automation, models, plugins, MCP, and configuration by reading reference documentation.