agent-capability-core

Use when the task involves reading, creating, or editing `.docx` documents, especially when formatting or layout fidelity matters; prefer `python-docx` plus the bundled `scripts/render_docx.py` for visual checks.

Use when a user gives a clear, generalizable correction about how a local skill should trigger or behave, or when a novel multi-step workflow is completed and the user approves continuing with it. Update the owning skill or create a narrow new skill, and record durable rules plus avoided paths under ~/.codex/skills.

Use whenever working inside any source repository or project workspace, especially if the task may create helper scripts, generated artifacts, scratch notes, exported reports, screenshots, debug logs, temporary data files, or test/validation run waste. Keeps the repository clean by routing auxiliary files outside the repo, using external workspace paths by default, cleaning accidental repo clutter immediately, and sweeping disposable test artifacts when runs stop.

Convert screenshot or image sequences into deterministic MP4 review clips. Use when Codex needs to merge screenshots or still-image frames into video from a folder, glob, numbered pattern, or JSON manifest, especially for sparse frame-step captures, per-sequence review clips, or sidecar report generation.

Use when the user explicitly asks for a desktop or system screenshot (full screen, specific app or window, or a pixel region), or when tool-specific capture capabilities are unavailable and an OS-level capture is needed.

Convert a user-trained multi-step workflow into a reusable Codex skill or a stable workflow template. Use when the user asks to 固化 workflow, 抽象模板流程, 沉淀流程与提示, 把当前流程做成技能, 将多轮修正变成可复用框架, or quickly freeze an orchestration pattern for future tasks.

Use when the user wants local files or folders in Codex output to appear as clickable blue links, or when a response includes local paths the user is likely to open. Formats local filesystem references as absolute Markdown links that open the target file or directory in the Codex app.

Maintain manifest-driven shared/runtime memory frameworks that use file-backed shared source plus runtime working copies and SQLite sidecar indexes. Use when Codex needs to add or refactor memory families, validate/render/promote/query/compact memory stores, enforce shared-vs-local boundaries, or tune lifecycle rules such as dedupe, supersede, provenance, and temporal validity.

Verify whether OpenAI API access is still usable. Use when Codex needs to test an OpenAI secret key or admin key, confirm chat, responses, image, or usage endpoints still answer, or distinguish quota, auth, scope, and model-permission failures from local client bugs.

Safely scope, commit, rebase, and push Git changes when the user wants minimal collateral impact, explicit conflict policy, or mixed superproject/submodule updates. Use for requests like 提交commit, 重新提交, push到远端, 合并主分支后再提交, or 只提交这些文件. Prefer mainline changes when integrating main into local work, prefer local recent changes when replaying local commits onto a moved remote branch, and verify untouched files are not altered.

Use when a user needs a new skill or script and the preferred workflow is to search the internet for existing skills, scripts, repositories, or examples first, then download or vendorize them locally, adapt them to the local environment, and only generate from scratch when no suitable reference exists.

Use when a user wants to audit local skills for portability across different models or agent environments, or wants non-portable skills rewritten so other people can reuse them outside the current model-specific setup.

Audit incoming contributor merges that add or modify skills, scripts, references, or workflow assets. Use when Codex needs to review a branch, PR, or local merge for near-duplicate skills, overlapping triggers, ambiguous ownership, redundant helper scripts, or reusable logic that should be unified before merging.

Automates browser interactions for web testing, form filling, screenshots, and data extraction. Use when the user needs to navigate websites, interact with web pages, fill forms, take screenshots, test web applications, or extract information from web pages.

Use whenever modifying code in any repository or project workspace unless the user explicitly asks for defensive programming, graceful degradation, compatibility fallbacks, or tolerant handling of invalid upstream state. Enforces a fail-fast style: prefer strict contracts, explicit preconditions, clear errors, and immediate failure over fallback branches, silent clamping, optional-path auto-detection, retries, or guard-heavy recovery code.

Use when the user asks to generate or edit images via the OpenAI Image API (for example: generate image, edit/inpaint/mask, background removal or replacement, transparent background, product shots, concept art, covers, or batch variants); run the bundled CLI (`scripts/image_gen.py`) and require `OPENAI_API_KEY` for live calls.

Export Markdown documents (.md) to clean PDF files through HTML rendering with Chromium, Edge, Chrome, or Playwright. Use when Codex needs to convert notes, letters, resumes, reports, or other Markdown documents into print-friendly PDFs, especially when browser-added headers and footers must be removed.

Use when tasks involve reading, creating, or reviewing PDF files where rendering and layout matter; prefer visual checks by rendering pages (Poppler) and use Python tools such as `reportlab`, `pdfplumber`, and `pypdf` for generation and extraction.

Persistent browser and Electron interaction through `js_repl` for fast iterative UI debugging.

Presentation creation, editing, and analysis. When Claude needs to work with presentations (.pptx files) for creating new presentations, modifying content, working with layouts, adding speaker notes, or any presentation tasks.

Create and edit presentation slide decks (`.pptx`) with PptxGenJS, bundled layout helpers, and render/validation utilities. Use when tasks involve building a new PowerPoint deck, recreating slides from screenshots/PDFs/reference decks, modifying slide content while preserving editable output, adding charts/diagrams/visuals, or diagnosing layout issues such as overflow, overlaps, and font substitution.

Use when the user asks to generate, edit, extend, poll, list, download, or delete Sora videos, create reusable non-human Sora character references, or run local multi-video queues via the bundled CLI (`scripts/sora.py`); includes requests like: (i) generate AI video, (ii) edit this Sora clip, (iii) extend this video, (iv) create a character reference, (v) download video/thumbnail/spritesheet, and (vi) Sora batch planning; requires `OPENAI_API_KEY` and Sora API access.

Use when the user asks for text-to-speech narration or voiceover, accessibility reads, audio prompts, or batch speech generation via the OpenAI Audio API; run the bundled CLI (`scripts/text_to_speech.py`) with built-in voices and require `OPENAI_API_KEY` for live calls. Custom voice creation is out of scope.

Toolkit for styling artifacts with a theme. These artifacts can be slides, docs, reportings, HTML landing pages, etc. There are 10 pre-set themes with colors/fonts that you can apply to any artifact that has been creating, or can generate a new theme on-the-fly.

Transcribe audio files to text with optional diarization and known-speaker hints. Use when a user asks to transcribe speech from audio/video, extract text from recordings, or label speakers in interviews or meetings.