SKILLS

Create premium Accenture-branded PowerPoint presentations following Tufte data visualization principles and Dieter Rams design philosophy. Use when creating slides, presentations, or decks for Accenture corporate settings that require (1) Accenture brand identity (purple A100FF chevron logo clean layouts) (2) Minimal data-focused design (Tufte less but better) (3) Executive-ready aesthetics for boardroom presentations client pitches or internal reports

Route proposed additions to the global Codex control plane by deciding whether a request belongs in `~/.codex/AGENTS.md`, a global workflow or reference doc under `~/.codex/docs`, `~/.codex/bin/workflow.py`, or an enforceable non-prose surface. Use when asked to "add this to global AGENTS", "promote this workflow to global", "add a new global trigger", or decide whether something should be a trigger, workflow, reference, hook, lint rule, or runtime guard.

Add, remove, serve, test, and operate a server-backed in-page agent feedback capability for HTML artifacts, static pages, or local app builds. Use when the user wants Browser-style annotation inside the artifact/app itself, marker-local chat, comment-triggered agent work, queued marker processing, deterministic install/remove scripts, preflight/closeout checks, browser acceptance evidence, or progressive-disclosure access to marker context.

Audits any agent harness codebase against 13 meta-principles of quality, scores each 0 to 10 with code citations, generates a self-contained HTML report for operators, and writes an agent-readable AGENTS.md improvement directive with file-level architecture recommendations to reach 9.5 or better. Use when the user wants to audit or rate an agent harness, evaluate harness quality, score an agent framework, get a roadmap for improving a harness, run agentharness-audit, audit this harness, rate Hermes, audit Codex, or audit a Claude Code harness. Not for individual skill audits or general code review.

Use when controlling ChatGPT Atlas through Codex Computer Use, especially browser tasks where context efficiency matters: fewer screenshots, fewer UI snapshots, fewer clicks, and cleaner browser state.

Audit project against golden principles, produce scorecard. Use for /audit, "check project quality". NOT for system setup health (/harness-audit) or eval scoring (/eval-score).

Autonomously optimize any Claude Code skill by running it repeatedly, scoring outputs against binary evals, mutating the prompt, and keeping improvements. Based on Karpathy's autoresearch methodology. Use when: optimize this skill, improve this skill, run autoresearch on, make this skill better, self-improve skill, benchmark skill, eval my skill, run evals on. Outputs: an improved SKILL.md, a results log, and a changelog of every mutation tried.

Apply the Autonomous Agent Builder design system to the autonomous-agent-builder codebase. Use when building or restyling Builder dashboard pages, adding or auditing primitive components, generating theme presets, wiring status language, motion hooks, or validating a screen against the locked Builder design system.

Allows you to view the user's screen as well as several hours of history. Use when the user makes a reference to their recent work, for which it'd be helpful to see the screen. This skill MUST be used whenever you need to resolve ambiguity in a user request, where the user hasn't specified enough context to do the task. Examples include disambiguating the specific user/app/document/error the user is referring to. You must also use this skill if the user asks about any question regarding Chronicle or asks what you can see from the screen.

Use when configuring apps to use z.ai GLM 4.7 proxy instead of standard Anthropic API. Works with any frontend (Next.js, Vite) or backend (Python, Node.js) framework.

Design or review CLIs so both coding agents and humans can use them reliably: dual-audience output, non-interactive paths, layered help, machine-readable data, predictable flags, safe mutations, and actionable errors. Use when building a CLI, adding commands, writing --help, or when the user mentions agents, terminals, automation-friendly CLIs, JSON output, or headless usage.

Use for read-only code quality, correctness, security, maintainability, observability, and operational audits that produce a ranked PROGRESS.md backlog in an isolated audit worktree.

Audit a project and recent Codex session behavior to recommend Codex productivity setup across AGENTS.md, skills, hooks, MCP/plugins, subagents, local environment actions, automations, memories, and workflow docs. Use when the user asks how to make repeated steering second nature or asks what Codex setup is missing.

Use when testing web apps with Perplexity/Comet browser automation, debugging with real browsers, verifying authenticated flows, or when client-side visual verification is needed. Load for dynamic content testing, form automation, console error checking, or when server-side logs alone are insufficient. This skill provides a 3-phase token-efficient methodology: (1) Server-side checks via process_logs MCP and curl, (2) Client-side verification via Comet MCP with specific prompts, (3) Learning loop via context graph storage.

Analyze and report what is consuming the 200K context window. Use when the user says /context-budget, "what's eating my context", "how much context am I using", or wants to optimize token usage.

Use when storing decision traces, querying past precedents, or implementing learning loops. Load in COMPLETE state or when needing to learn from history. Covers semantic search with Voyage AI embeddings, ChromaDB for cross-platform vector storage, and pattern extraction from history.

Single entry for authoring, auditing, and optimizing Codex skills under .codex/skills/NAME. Three lanes: Create scaffolds a folder and SKILL.md template; Audit runs deterministic checks over frontmatter, naming, body size, description quality, and reference integrity; Optimize fixes audit findings such as description rewrites, body to references split, frontmatter repair, and name normalization. Use when the operator says create a skill, scaffold a skill, add SKILL.md, audit a skill, check skills, validate skill conformance, optimize a skill, fix a skill description, shrink a skill, or why is this skill not activating.

Use when verifying outcomes with code instead of LLM judgment, versioning prompts with hashes, or ensuring reproducible agent behavior. Load for any critical verification. Scripts return boolean exit codes, not subjective assessments. Prompts use semantic versioning with SHA256 validation.

Use when designing UI components, building layouts, creating page structures, or developing design systems following Dieter Rams' principles. Load for any task involving buttons, sliders, cards, forms, inputs, menus, headers, navigation, hero sections, grids, responsive layouts, page templates, tabs, toasts, modals, alerts, loading states, or UI/layout elements that need to be innovative, useful, aesthetic, understandable, unobtrusive, honest, long-lasting, thorough, environmentally friendly, or embody 'less but better' design philosophy.

Use when implementing hooks that BLOCK invalid actions, creating quality gates for state transitions, or enforcing tested:true verification. Load when designing enforcement mechanisms. Uses exit code 2 to block, JSON permissionDecision:deny, or updatedInput modification. Rules are instructions; hooks are enforcement.

Scaffold PoC projects, test workflows against eval criteria. Use for /eval, "test this workflow". NOT for scoring (/eval-score) or improving (/autoimprove).

Create living execution documents for complex multi-step tasks. Use for /execplan, "create a plan". NOT for quick tasks or code reviews — only structured phased planning.

Use when a user asks to publish, debug, or fix a GitHub change until the PR is ready to merge. Create or reuse the PR, diagnose GitHub Actions failures with `gh`, address active merge-blocking review comments autonomously when safe, and stay in a PR-to-mergeable loop until required checks are green and no fixable blocker remains. After the PR is ready to merge, run `session-introspection` to capture anti-recurrence controls.

Check system setup health across 7 categories. Use for /harness-audit, "verify my setup", "check harness". NOT for project quality (/audit) or eval scoring (/eval-score).

Produce a single self-contained HTML file as the deliverable for information-dense tasks that benefit from tables, SVG diagrams, inline code annotations, side-by-side comparisons, tabs, interactive controls, or copy-back affordances. Lanes: plan, review, report, prototype, editor. Use when the operator asks for an HTML artifact, HTML file, single-page HTML, HTML explainer, HTML spec, PR review in HTML, incident or status report as HTML, implementation plan in HTML, feature explainer, concept explainer, flowchart, SVG diagram, slide deck, design-system docs, component variant gallery, drag-drop triage board, feature-flag admin UI, prompt tuner, shareable explainer, or throwaway UI/editor. Do not use for high-design marketing pages or production apps.

Implement the next pending feature from feature-list.json. Lean loop: pick feature, implement with tests, verify all checks pass, commit, mark done. Use when features are pending and you're ready to build. Also use when user says 'build next feature', 'implement', 'start coding', 'what's next', or after /init-project has created the feature list.

Save a reusable pattern from this session to the knowledge base. Use for /learn, "remember this", "save this pattern". NOT for session state (/save-session) or full session review (/introspect).

Use when presenting learning recommendations from background analysis, reviewing pending pattern suggestions, or approving automation scripts from traces. Load when learning-agent has findings to present.

Run high-stakes questions, ideas, and decisions through a council of 5 AI advisors who analyze independently, peer-review each other anonymously, and synthesize a final verdict with artifacts. Mandatory triggers: 'council this', 'run the council', 'war room this', 'pressure-test this', 'stress-test this', 'debate this'. Strong triggers when paired with a real decision or tradeoff: 'should I X or Y', 'which option', 'what would you do', 'is this the right move', 'validate this', 'get multiple perspectives', 'I can't decide', 'I'm torn between'. Do not use for factual lookups, simple yes/no questions, summarization, or low-stakes choices.

Use when creating, updating, or verifying a local macOS WidgetKit widget, especially desktop widgets with a native black dashboard visual system, reliable local install/run workflow, app-group snapshot/cache internals, timeline reloads, and real desktop screenshot proof.

Use when analyzing macOS apps for optimization opportunities, improving memory/RAM usage, reducing memory leaks, optimizing SwiftUI performance, or analyzing Swift code for efficiency. Load for any macOS app performance analysis or memory optimization task.

Guide for creating high-quality MCP (Model Context Protocol) servers that enable LLMs to interact with external services through well-designed tools. Use when building MCP servers to integrate external APIs or services, whether in Python (FastMCP) or Node/TypeScript (MCP SDK).

Use when setting up MCP servers for the first time or verifying MCP configuration. Runtime-aware: writes Claude .mcp.json or Codex .codex/config.toml, referencing central mcp-servers only (no per-project cloning).

Configure or modify OneCLI integrations in agentic apps. Use when adding `@onecli-sh/sdk`, wiring OneCLI into container or process launchers, managing OneCLI agent identifiers, or adding manual approval flows.

Run the autonomous control plane loop — orient, identify, act, verify, update, introspect. Use for /operate, "run the loop", "what needs doing". NOT for single-task work — use specific skills instead.

Decide where instructions, memory, plans, and artifacts should live between global Codex baseline and project-specific files. Use when creating/updating AGENTS/ROADMAP/memory docs, introducing new workflow rules, or deciding whether content belongs in ~/.codex/rules versus a repository.

Bootstrap organized project or workspace governance using Linear for execution, Notion for durable memory, AGENTS-triggered docs, MCP/auth setup, and no-drift workflow controls. Use when starting a new repo/workspace or retrofitting ad hoc project management into a governed operating model.

Run a bounded quality audit across a repo or control plane, split the scan into 3-5 evidence lanes, consolidate overlapping findings, and remediate the highest-leverage issues with one main-agent owner. Use when a user wants a broad hygiene pass, drift audit, regression triage, control-plane review, or a parallel analysis followed by direct remediation.

Bootstrap or assess a Codex-first day-one readiness baseline for an unfamiliar repo. Use when entering a repo for the first time, making a repo agent-runnable, or needing one canonical readiness report before implementation.

Process a research article or blog post into actionable insights for the knowledge base. Use for /research, "process this article", "extract insights from". NOT for web search or code review.