Claude Skills by kookr-ai

Plan phase for oss-pr-lessons playbook — load state, determine next batch of PRs to analyze for any repository

Threshold verification for oss-pr-lessons playbook — determine if distillation should trigger for any repository

Decide where to place a new rule, skill, hook, KB entry, agent, CLAUDE.md rule, memory, or context artifact. Use when choosing among user, project, plugin, hook, memory, and knowledge-base surfaces, especially before saving a behavioral rule or feedback memory.

Create or revise Kookr playbooks and scheduled orchestration, including parameters, project identity, idempotency, model policy, and launch integration

Playwright E2E testing patterns. Use when playwright, E2E test, toBeVisible, route mock, modal, flaky test, or selector issues.

Patterns for excellent open-source PR contributions, distilled from analyzing real PRs across repositories

Triage, fix, and resolve PR review comments end-to-end — fetch comments, assess relevance, implement fixes, push, and resolve GitHub threads. Used by post-push when feedback arrives.

Self-review checklist before creating a PR — repo checks, diff review, reviewer specialists, and gate state creation before gh pr create.

Process lifecycle and graceful shutdown - signal handling, connection draining, health probes, resource cleanup ordering, worker management, zero-downtime deploys

Keep realtime UI state consistent with snapshot+delta sequencing, source tagging, dedup layers, and safe reconciliation.

Writing precise, testable, structured requirements - SHALL/SHOULD/MAY phrasing, INVEST stories, acceptance criteria, ubiquitous language, prioritization

Spawn an isolated judge subagent to evaluate reviewer predictions against real human reviews

Meta-mutation — improve the mutator or judge skills when the reviewer improvement loop stalls

Spawn a mutator subagent to improve the reviewer skill based on judge feedback

Spawn an isolated reviewer subagent to blindly predict PR review comments

Prepare isolated context files and fetch ground-truth reviews for selected PRs

Select merged PRs with substantive human reviews for reviewer distillation experiment

Iterative RFC drafting workflow — draft in worktree, run parallel critic subagents, incorporate feedback over N rounds, present to user before any action

Pre-push checklist for rust-lang/rust PRs — catches naming, verification, description, and convention issues before they reach reviewers

Test writing conventions for rust-lang/rust contributions — naming, headers, directives, verification, PR descriptions

Safe refactoring patterns - code smell detection thresholds, baby-step refactoring, characterization tests, Expand-Migrate-Contract, Strangler Fig, primitive obsession, value objects

Build a Ralph-like sequential Kookr task chain where each task completes one independent unit, records durable state, and spawns the next task with the same continuation contract. Use for issue batches, queue drains, staged migrations, or other long runs that should proceed one task at a time without relying on conversation memory.

Reflect on a mistake or unnecessary user correction — root-cause why it happened, then propose and implement structural fixes (skills, hooks, CLAUDE.md, memory, scripts). Use after the user corrects behavior, repeats guidance, or points out an avoidable error.

Shell subprocess safety patterns - execFileSync over execSync, array arguments, shell feature replacements, static analysis guards. Use when spawning child processes, running git/docker/system commands, or reviewing code that calls execSync.

State machine discipline, saga compensation, event-sourced persistence, workflow orchestration. Use when state transitions, workflow steps, saga, compensation, event sourcing, or state machine testing.

Per-task self-reflection workflow triggered by user thumbs-up/down on a completed Kookr task. Reads a feedback bundle, walks the project Persistence Mechanism Picker, and proposes ONE remediation (skill update, CLAUDE.md edit, or hook addition) for thumbs-down OR ONE reinforcement edit for thumbs-up. Two human gates: confirm direction before editing, confirm before opening a PR. Memory frontmatter gate blocks type: feedback.

Anytime task reflection workflow triggered from a live or terminal Kookr task. Reads an immutable snapshot bundle, analyzes what went well and what went wrong, then asks the user to validate the diagnosis and choose positive reinforcement, negative reinforcement, both, or observation only before making any changes.

Implement features using TDD workflow with requirements-first approach. Use when implementing new features, adding functionality, or when user asks to build something.

Choose and implement unit/integration/E2E tests that catch real regressions in this monorepo.

Reduce token waste — avoid redundant reads, duplicate searches, unnecessary tool calls. Use when planning tool usage or when context is getting large.

TypeScript type safety patterns - unknown vs any, generics golden rule, null narrowing, inference discipline, import type, discriminated unions, error typing. Use when writing or reviewing TypeScript code for type correctness.

Before implementing any user-facing UI/visual change, proactively offer design mockup variants for the user to pick from — via a single multiple-choice prompt. Use whenever a task asks to add, change, redesign, restyle, or improve any UI element (panel, button, badge, layout, dashboard, indicator, color, spacing, look-and-feel). Ask once (do mockups? how many? what fidelity?), build a self-contained HTML gallery from the app's REAL CSS tokens, let the user choose a variant, then implement it....

WebSocket patterns for Hono/Bun dashboard connections, connection manager, per-client state, browser lifecycle, back-pressure, rate limiting, message versioning