agent

Use when the user asks how to build with OpenAI products or APIs, asks about Codex itself or choosing Codex surfaces, needs up-to-date official documentation with citations, help choosing the latest model for a use case, latest/current/default-model prompting guidance, or model upgrade and prompt-upgrade guidance; use OpenAI docs MCP tools for non-Codex docs questions, use the Codex manual helper first for broad Codex self-knowledge, and restrict fallback browsing to official OpenAI domains.

Install Codex skills into $CODEX_HOME/skills from a curated list or a GitHub repo path. Use when a user asks to list installable skills, install a curated skill, or install a skill from another repo (including private repos).

Create and scaffold plugin directories for Codex with a required `.codex-plugin/plugin.json`, optional plugin folders/files, valid manifest defaults, and personal-marketplace entries by default. Use when Codex needs to create a new personal plugin, add optional plugin structure, generate or update marketplace entries for plugin ordering and availability metadata, or update an existing local plugin during development with the CLI-driven cachebuster and reinstall flow.

Split the current git worktree into one or more meaningful commits with clean staging boundaries and commit hygiene. Use when the user asks to commit current changes, wants related changes grouped into multiple commits, or needs help preparing commit messages without disturbing unrelated local modifications.

Code context using Exa. Finds real snippets and docs from GitHub, StackOverflow, and technical docs. Use when searching for code examples, API syntax, library documentation, or debugging help.

Search for research papers and academic content using Exa advanced search. Full filter support including date ranges and text filtering. Use when searching for academic papers, arXiv preprints, or scientific research.

Guide for creating effective skills. This skill should be used when users want to create, update, review, and debug native Zsh completion scripts built on the modern completion system (`compsys`). Use when Codex needs to create a new `_command` completion function, extend subcommand, option, or argument completion, fix completion parity against a CLI's current help output, migrate legacy `compctl` behavior to `#compdef` or `compdef`, or diagnose matcher, style, tag, or context issues in existing Zsh completion code.

Code context using Exa. Finds real snippets and docs from GitHub, StackOverflow, and technical docs. Use when searching for code examples, API syntax, library documentation, or debugging help.

Use when Codex needs to drive a real browser through the local `agent-browser` CLI for website navigation, form filling, screenshots, downloads, scraping, login reuse, tab or session management, or DOM inspection. Apply it to browser automation tasks that should use `agent-browser` instead of hand-written shell flows, and prefer it over broader page-cloning workflows when the goal is to interact with or inspect a live site.

Continue executing the active plan end-to-end, reprioritize remaining work as needed, and keep working until the planned tasks are complete. Use when the user wants the current task carried forward without stopping after the next milestone.

Use when Codex needs to inspect, manage, troubleshoot, or optimize Google Cloud resources with the Google Cloud CLI (`gcloud`). Apply it for Google Cloud tasks involving Compute Engine, Cloud SQL, GKE, Dataproc, Cloud DNS, Cloud Run, App Engine, authentication, configuration, or other project-scoped resource operations, especially when the work requires choosing between product-specific tools and raw `gcloud`, clarifying project or location identifiers, and using local gcloud defaults safely.

Apply version-aware modern Go idioms and standard-library features that are supported by the target module's Go version. Use when writing, reviewing, refactoring, or modernizing Go code and the correct choice depends on whether the repository targets Go 1.18 through Go 1.26 features.

Split work into meaningful commits and attach comparable benchmark evidence for performance-sensitive changes, especially Go code with `Benchmark...` tests. Use when the user asks for benchmark-backed commits, the diff affects benchmarked code, or relevant commit messages should include raw performance comparison output.

Summarize the current planned or remaining work as a prioritized number and checkbox list. Use when the user asks what tasks are planned, what remains to do, wants TODO items ordered by importance, or requests the plan in English or Japanese for a repository, feature, or target component.

Investigate and drive major performance work using repository-local planning, LSP-assisted code navigation, and benchmark evidence. Use when the user asks for performance optimization planning, wants dramatic latency, throughput, memory, CPU, I/O, or cost improvements, or invokes `/perf-plan`-style work with an LSP, focus, or scope.

Create and maintain repository-local ExecPlan documents for complex features and long-horizon tasks. Trigger only when the user explicitly invokes /execplan, asks to create or revise an ExecPlan, asks for a PLANS.md-compliant execution plan, or requests a design-to-implementation plan that must include milestones, progress tracking, validation commands, decision logs, and recovery guidance.

Create and execute structured ExecPlans — self-contained design documents that guide coding agents through complex implementations. Use this skill whenever the user's task involves multi-file changes, significant refactors, new features with multiple components, architectural changes, or any work that would benefit from a structured plan-then-execute approach. Also trigger when the user explicitly asks for an "execplan", "execution plan", or "structured plan". Even if the user doesn't mention planning, use this skill when the task is clearly too complex for ad-hoc implementation — e.g., tasks touching 3+ files, requiring research before coding, involving migrations, or needing coordinated changes across multiple packages or services. When in doubt about whether a task needs an ExecPlan, it probably does.