code-yeongyu/lazycodex

Ast-Grep performs AST-aware code search and rewrite, letting you match code by structure rather than text. Use it for deterministic codemods like rewriting console.log to logger.info, stripping `as any`, or migrating require to import across a codebase.

Bug Fix Contribution debugs a concrete LazyCodex or Codex defect, implements the smallest tested fix in a fresh temporary workspace, and delivers it as a verified-fix issue for LazyCodex-owned repos or a fork PR for upstream openai/codex.

Codex Rules injects static project instructions and file-specific rules into Codex sessions automatically. Use it to configure supported rule file locations, matching behavior, and environment variables for Codex.

Coding Agent Sessions locates, reads, and reconstructs coding-agent sessions across products such as Codex, Claude, OpenCode, Senpi, and others. Use it when asked to find, search, or export local agent transcripts, session IDs, token usage, or subagent histories.

Comment Checker helps Codex understand and respond to automatic comment-checker feedback emitted after successful edit-like PostToolUse hooks. Use it when a patch triggers a warning and Codex must fix or explain the flagged comment before moving on.

Data Scientist processes and analyzes data files using DuckDB and Polars for high-performance workloads. Use it when you need to query, filter, join, aggregate, visualize, or clean CSV, Parquet, JSON, or DuckDB datasets.

Debugging applies a hypothesis-driven workflow to find root causes of crashes, silent failures, wrong responses, memory leaks, and async misbehavior in any language or binary, using runtime evidence and locking fixes with failing tests. Use it when attaching debuggers, reverse engineering binaries, or reproducing flaky or CI-only failures.

Frontend routes frontend/web UI/UX/visual requests to four rulesets for design taste, performance perfection, UI/UX lookup, and design operations, plus lazyweb real-app research. Use it when building, styling, redesigning pages/components, React setup, performance audits, or visual QA.

Git Master handles Git history and commit tasks: atomic commits, rebase/squash/fixup, blame, bisect, reflog, and log search. Use it when a task requires a commit or investigation of who/when/why code changed.

Init Deep generates hierarchical AGENTS.md files across a repository, scoring subdirectories by complexity and deciding where guidance is warranted. Use it to initialize or refresh an agent-readable knowledge base for a codebase.

LSP connects Codex to language servers for diagnostics, definitions, references, symbols, and rename safety checks. Use it when you need type-aware feedback or to verify edits in the current workspace.

LSP Setup configures a Language Server for a specific language so editor and agent tooling (diagnostics, go-to-definition, find-references, rename) works. Use it to install servers, choose between alternatives like basedpyright vs pyright, or fix 'no LSP server configured' errors.

LazyCodex Bug Reporter creates high-signal bug issues or PRs routed to the repository that owns the defect, using source-backed root cause, reproduction steps, and fix guidance. Use it when a user asks to report, file, or triage a LazyCodex or upstream Codex bug.

LazyCodex Doctor inspects a local LazyCodex and Codex CLI installation and compares it against the latest sources to produce a PASS/WARN/FAIL report. Use it when Codex behaves oddly after an install, update, or config change, or when you want a local install audited.

Programming enforces strict typed, test-driven coding standards across Python, Rust, TypeScript, and Go. Use it before writing or editing code in those languages to load language-specific references and apply the shared seven axioms.

Refactor performs intelligent, deterministic refactoring with codebase awareness, mapping code, assessing risk, planning, and verifying each change. Use it to safely restructure files, modules, or projects by triggering /refactor and specifying a target, scope, and risk strategy.

Remove AI Slops cleans AI-generated code smells from branch changes or explicit file lists, locks behavior with regression tests first, runs categorized cleanup via parallel agents, and verifies with quality gates. Use it when asked to remove slop or clean AI-generated patterns from recent changes.

Review Work orchestrates a post-implementation review by launching five parallel sub-agents for goal verification, code quality, security, QA execution, and context mining; all must pass before a PR handoff. Use it when you need to review completed work.

Start Work executes a Prometheus work plan in Codex with Boulder state, evidence ledger updates, worktree discipline, parallel subagents, and Stop-hook continuation. Use after planning when the user says start work, execute plan, continue plan, resume plan, or asks to run a .omo/plans plan.

Teammode runs a named team of cooperating Codex workers with durable, script-managed state. Use it when you need to coordinate parallel agents or threads on non-isolated work with clear goals.

ULW Loop decomposes work into systematic, evidence-bound steps using ultrawork mode and manual QA. Use it for durable goal execution, checkpointed long-running delivery, or when evidence-led work and subagent delegation are required.

ULW Plan turns a vague or large request into a decision-complete work plan by exploring the codebase, asking only unresolved forks, and waiting for approval before writing plan artifacts under .omo/. Use it when you explicitly request ulw-plan or ask for a plan before coding.

ULW Research orchestrates maximum-saturation research using parallel explorer and librarian swarms across codebases, official docs, web sources, and open-source repositories, then delivers a cited synthesis with charts and visual QA. Use it for explicit deep-research or ultra-precise investigation requests.

Ultimate Browsing escalates web access when normal browsing or fetching is blocked, using layered techniques from headless extraction and WAF bypass to platform-specific readers and a stealth Chromium for interactions. Use it to scrape blocked pages, Chinese and social platforms, login-gated content, or to click, fill forms, screenshots, and play media.

Ultrawork activates a rigorous, outcome-first coding workflow when a Codex prompt contains 'ultrawork' or 'ulw', requiring tier triage, RED-to-GREEN evidence, and real-surface manual QA through HTTP, terminal, browser, or computer-use channels before work is declared done.