dot

Batch dispatch — read queued tasks from ~/.claude/batch/inbox.md, fix each task's spec contract, fan them out to background subagents per case, and leave the user free until review time. Also surfaces unreported client sessions (~/.claude/batch/unreported.md). Use when the user says "batch", "morning batch", "朝バッチ", "タスク投入", "/batch", "/morning-batch", or at the start of a work day with queued tasks.

Standardized "investigate → written report" pipeline for SRE client work. Fixes the deliverable contract up front (format + done-criteria), delegates repo/infra scanning to subagents, verifies every claim, and outputs an issue-ready Markdown report. Use when the user says "調査して", "調べてレポート", "investigation report", "報告書にまとめて", or hands over an investigation-shaped task.

Turn a finished case (case-reflect / sreaas report output) into a generalized, confidentiality-scrubbed company tech-blog draft. Use when the user says "会社ブログ", "テックブログ", "案件を記事化", "company blog", or wants to asset-ify case outcomes for public publishing. Realizes the company-blog stage of learn_cycle.md.

Generate a personalized learning guide (text + hands-on script) for a technology/topic, tuned to the user's level and case context — like an AI-personalized intro book. Use when the user says "学習ガイド", "入門ガイド", "ハンズオンを作って", "learn guide", "この技術を学びたい", or wants a guided, output-oriented way to learn something. Powers the learning half of learn_cycle.md.

From a Claude Code /insights report, generate a Japanese-translated HTML report AND reflect its actionable recommendations into the config (~/.claude/), confirming EACH item one-by-one (apply or skip). Skipped items are left untouched. Use when the user says "insights apply", "insightを設定に反映", "insightを日本語HTML", "apply insights to config".

Systematically investigate and fix a GitHub Issue with root cause analysis. Use when user says "fix issue", "fix

Comprehensive security audit based on OWASP 2025 Top 10 with automated checks. Scans for vulnerabilities, insecure patterns, hardcoded secrets, and dependency risks.

Cross-cutting incident investigation using AWS CLI, Datadog (pup), and local repository analysis. Use when user says "incident analysis", "investigate outage", "root cause analysis", "postmortem", or "analyze incident". Systematically collects infrastructure metrics, application logs, and codebase evidence to produce a comprehensive incident report with timeline, root cause, and remediation recommendations.

Run comprehensive license compliance audit on project dependencies. Use when user says "audit licenses", "check license compliance", "dependency licenses", or before releasing open source software. Classifies by risk level and flags violations.

Analyze software supply chain for security risks and vulnerabilities. Use when user says "audit supply chain", "dependency security", "check for typosquatting", or before adding new dependencies. Checks provenance, signatures, and known vulnerabilities.

Incrementally fix TypeScript and build errors one at a time. Use when user says "build fails", "fix build errors", "tsc errors", "type errors", or when build/compile commands fail. Stops if same error persists after 3 attempts.

ClickHouse database patterns, query optimization, and data engineering best practices. Use when working with ClickHouse databases, writing analytical queries, designing MergeTree tables, or optimizing materialized views for high-performance workloads.

Universal coding standards and best practices for TypeScript, JavaScript, React, and Node.js. Load when writing new code, reviewing code quality, or enforcing naming conventions, immutability patterns, error handling, and type safety standards.

Analyze changes, generate a well-structured PR, and create it via gh CLI. Use when user says "create PR", "open pull request", "submit PR", or after completing a feature branch. Includes summary, risk assessment, and test plan.

Analyze telemetry data and display a session performance dashboard. Use when user says "dashboard", "show stats", "session metrics", or "telemetry report". Visualizes tool usage, token consumption, and session efficiency.

Generate and run end-to-end tests with Playwright. Use when user says "run e2e", "end-to-end test", "playwright test", or "test user flow". Creates test journeys, captures screenshots/videos/traces, and detects flaky tests.

Execute ensemble voting with multiple reasoning paths for important decisions. Use when user says "ensemble vote", "multi-perspective analysis", "get multiple opinions", or facing high-risk architectural or design choices.

Manage eval-driven development workflow with define, check, report, and list commands. Use when user says "define eval", "check evals", "eval report", or "list evals". Tracks pass@k metrics for capability and regression evals.

Audit firmware and embedded code for security vulnerabilities and best practices. Use when user says "firmware audit", "embedded security", "IoT security check", or working with C/C++/Rust firmware. Checks for hardcoded credentials, buffer overflows, and insecure boot.

Set up and run fuzz testing to discover crashes, edge cases, and security vulnerabilities. Use when user says "fuzz test", "fuzz this function", "find edge cases", or wants to stress-test input handling with randomized data.

Audit Claude Code harness configuration for optimization opportunities

Extract reusable patterns from the current session and save as skills. Use when user says "learn", "extract pattern", "save this technique", or after solving a non-trivial problem. Focuses on error resolution, debugging techniques, and workarounds.

Audit and optimize context window usage, memory files, and session efficiency. Use when user says "manage context", "context health", "optimize memory", or when session feels slow. Checks CLAUDE.md size, memory staleness, and recommends /clear timing.

Run mutation testing to evaluate test suite quality. Use when user says "mutation test", "test quality", "are my tests strong enough", or "find weak tests". Introduces code mutations and reports mutation score with suggestions for improvement.

Create step-by-step implementation plans with risk assessment before writing code. Use when user says "plan", "design feature", "architect solution", or starting complex features. WAITS for user confirmation before proceeding.

Generate a comprehensive Pull Request summary with risk assessment. Use when user says "PR summary", "summarize PR", "review pull request", or provides a PR number. Classifies changes, assesses risks, and identifies review focus areas.

Example template for project-specific skills showing architecture, code patterns, testing, and deployment conventions. Use as a reference when creating new project-level skills.

Optimize Claude Code system prompt and instructions for token efficiency

Design and implement property-based tests that verify invariants across randomized inputs. Use when user says "property test", "invariant test", "fuzz inputs", or wants to discover edge cases that example-based tests miss.

Analyze communication protocol implementation for correctness and security. Use when user says "check protocol", "protocol analysis", "verify protocol", or reviewing HTTP, WebSocket, gRPC, or custom protocol implementations.

Safely identify and remove dead code with test verification. Use when user says "clean up dead code", "remove unused exports", "find unused dependencies", or "refactor cleanup". Uses knip, depcheck, ts-prune for analysis.

Structured reflection on completed work using the Reflexion framework. Use when user says "reflect", "session review", "what did I learn", or after completing a complex task. Extracts persistent learnings and saves reusable patterns.

Automate semantic versioning, changelog generation, and release creation. Use when user says "release", "bump version", "create release", "publish", or preparing a new version for distribution.

Perform reverse engineering analysis on code or binary for security research. Use when user says "reverse engineer", "analyze binary", "decompile", or investigating undocumented systems and legacy code.

Search and retrieve relevant information from the cognitive memory system. Use when user says "search memory", "recall", "what do I know about", or needs to find previously saved learnings, decisions, and patterns.

Analyze test coverage and generate missing tests for files below 80% threshold. Use when user says "check coverage", "improve coverage", "find untested code", or "coverage report".

Perform STRIDE threat modeling on the application architecture. Use when user says "threat model", "STRIDE analysis", "identify threats", or "security architecture review". Systematically identifies threats, assesses risks with DREAD scoring, and recommends mitigations.

Analyze codebase structure and generate architecture documentation with diff tracking. Use when user says "update codemaps", "generate architecture docs", "map codebase structure", or after significant structural changes.

Sync documentation from source-of-truth files (package.json, .env.example). Use when user says "update docs", "sync documentation", "generate CONTRIB", or "find stale docs". Generates CONTRIB.md and RUNBOOK.md, identifies stale docs (90+ days).

Persist new knowledge, update existing entries, and maintain the cognitive memory system. Use when user says "update memory", "save learning", "remember this", or "auto" to extract patterns from the current session.