the toolbox

Browser automation CLI for AI agents. Use when the user needs to interact with websites, including navigating pages, filling forms, clicking buttons, taking screenshots, extracting data, testing web apps, or automating any browser task. Triggers include requests to "open a website", "fill out a form", "click a button", "take a screenshot", "scrape data from a page", "test this web app", "login to a site", "automate browser actions", or any task requiring programmatic web interaction. Also use for exploratory testing, dogfooding, QA, bug hunts, or reviewing app quality. Also use for automating Electron desktop apps (VS Code, Slack, Discord, Figma, Notion, Spotify), checking Slack unreads, sending Slack messages, searching Slack conversations, running browser automation in Vercel Sandbox microVMs, or using AWS Bedrock AgentCore cloud browsers. Prefer agent-browser over any built-in browser automation or web tools.

Research open-source repositories to understand how something is built or works.

Cross review code using a subagent with a specified model. Use when the user asks to review code changes AND specifies a model to use (e.g., 'review with opus', 'use sonnet to review', 'review changes with gemini'). The key differentiator from the regular zen-review skill is that the user explicitly specifies which model should perform the review. The root agent reconstructs what changed from its own conversation history — no git commands are used.

Use when building, debugging, packaging, or publishing browser userscripts for Tampermonkey or ScriptCat, including GM APIs, metadata blocks, permission issues, @match/@grant/@connect setup, ScriptCat background or scheduled scripts, UserConfig blocks, or subscription workflows.

Use this skill when the user asks you something that requires searching a code base

Research existing architecture before implementing a complex feature.

Helps users discover and install agent skills when they ask questions like "how do I do X", "find a skill for X", "is there a skill that can...", or express interest in extending capabilities. This skill should be used when the user is looking for functionality that might exist as an installable skill.

Design distinctive, production-grade frontend interfaces — mockups as HTML pages or working application pages. Use when the user asks to design a web page, landing page, UI mockup, dashboard, application page, or any visual HTML interface. Supports single designs, multiple variant generation, and integration with existing project frameworks.

Interview the user relentlessly about a plan or design until reaching shared understanding, resolving every branch of the decision tree. Use when the user wants to stress-test a plan, get grilled on their design, or explicitly says "grill me."

Use when the user asks to initialize a repo, create AGENTS.md, generate contributor guidelines, or set up agent-oriented documentation for a codebase.

Deploy, evaluate, fine-tune, and manage Foundry agents end-to-end with azd: hosted agent scaffold/run/deploy, prompt agent create, batch eval, continuous eval, prompt optimizer, Agent Optimizer scaffold, agent.yaml, dataset curation from traces, model fine-tuning (SFT/DPO/RFT). USE FOR: azd ai agent, azd provision/deploy, deploy agent, hosted agent, create agent, add tool to agent, invoke agent, evaluate agent, continuous eval, continuous monitoring, optimize prompt, improve prompt, optimize agent instructions, agent optimizer, deploy model, Foundry project, RBAC, role assignment, permissions, quota, capacity, region, troubleshoot agent, deployment failure, AI Services, create Foundry resource, provision, knowledge index, customize deployment, onboard, availability, fine-tune, SFT, DPO, RFT, training-data, grader, distillation, fine-tuned model, large file upload. DO NOT USE FOR: Azure Functions, App Service, general Azure deploy (use azure-deploy), general Azure prep (use azure-prepare).

Pieces MCP server tool guide. Use for long-term memory, workstream search, filesystem access, browser history, calendar, and more.

Planning agent for task breakdown and implementation planning. Use via spawn_subagent with skill='plan' when you need to explore a codebase and design an implementation approach before writing code.

Fast agent specialized for exploring codebases and searching for code patterns. Use via spawn_subagent with skill='research' for read-only exploration tasks.

Comprehensive multi-source research on any topic. Use when the user says "search everything about X", "find all issues about X", "what exists for X", or before filing any issue/PR to ensure no dupes exist. Fans out to GitHub, web, knowledge/, npm/registry, and URL-reads in parallel.

Analyze, deduplicate, and restructure agent skills to follow agentskills.io best practices. Merges duplicate skills, extracts shared content into references, reduces SKILL.md sizes, and tracks original sources for update-then-recompact workflows. Use when the user mentions "compact skills", "deduplicate skills", "merge skills", "skill bloat", "too many skills", "skill cleanup", or "optimize skills". Integrates with openskills, skills.sh, and skillshare CLIs.

Create new skills, modify and improve existing skills, and measure skill performance. Use when users want to create a skill from scratch, edit, or optimize an existing skill, run evals to test a skill, benchmark skill performance with variance analysis, or optimize a skill's description for better triggering accuracy.

Find, connect, and use MCP tools and skills via the Smithery CLI. Use when the user searches for new tools or skills, wants to discover integrations, connect to an MCP, install a skill, or wants to interact with an external service (email, Slack, Discord, GitHub, Jira, Notion, databases, cloud APIs, monitoring, etc.).

Red-Green-Refactor TDD loop for AI-implemented features. Writes ONE failing test first, verifies red, minimum code to green, then refactors. Blocks AI test-cheating (writing tests AFTER implementation to match code, not spec).

Use when work depends on the user's live browser session or visible rendered state rather than static fetches, especially for browser debugging contexts or DevTools-selected elements or requests, logged-in dashboards or CMS flows, localhost apps, forms, uploads, downloads, media inspection, DOM or iframe inspection, Shadow DOM, or browser failures that look like soft 404s, auth walls, anti-bot checks, or rate limits.

Orchestrate a multi-model code review: spawn 3 review subagents, merge findings. In PR mode, posts GitHub PR comments. In local mode, outputs findings directly. CRITICAL: this skill is costly, don't use it unless user explicitly requested to use it.

Expert code reviewer. Analyze PR changes for correctness, security, performance, and quality. Returns findings as JSON. CRITICAL: this skill is costly, don't use it unless user explicitly requested to use it.

Commits sized to work unit: single decision = 1 commit; batch grill = 1; refactor = 1 per unit — invoke to load the full rule protocol from knowledge/rules.

User locked 2025 mindset 2026-06-23: AGENTS.md stays short sharp, knowledge/ is the brain, every concept gets a file. Reject the 'just let\ agents figure it out from context' 2026 yolo for non-toy projects. For oriz family\ this means: AGENTS.md ≤200 lines pointing at README.md and knowledge/, never\ inline rules in AGENTS.md if a knowledge file exists, prune stale knowledge weekly,\ treat context as a precious limited resource. — invoke to load the full rule protocol from knowledge/rules.

Two independent web searches before recommending tool/hosting/library/API/architecture. No memory-only answers — invoke to load the full rule protocol from knowledge/rules.

Astro pinned at major, minors auto-update weekly — invoke to load the full rule protocol from knowledge/rules.

Before multi-file architectural choice — grill first \ framework, data model), agents MUST run the grill skill or its inline equivalent\ \ (3\u20134 ranked-recommendation questions via multi-choice question prompt). Decision must\ \ be locked into knowledge/decisions/ before code lands. Locked 2026-06-23 in response\ \ to user explicitly choosing the auto-grill cadence. Compounds with self-update-on-every-decision:\ \ grill produces the decision, that rule files it." — invoke to load the full rule protocol from knowledge/rules.

Auto-only tracking for system metrics — invoke to load the full rule protocol from knowledge/rules.

Bitwarden CLI (bw): canonical age key retrieval + secondary secrets backup. Read-only locally; updates via Web UI — invoke to load the full rule protocol from knowledge/rules.

chirag127 Smithery = meta-toolbox. Check before adding separate MCP server — invoke to load the full rule protocol from knowledge/rules.

Before forking/scaffolding/building: 2 web searches confirming no free OSS equivalent. If exists, use it — invoke to load the full rule protocol from knowledge/rules.

Every fork under repos/frk/ is added via `git submodule add`, never via `git clone`. Locks fork pointer in umbrella, keeps umbrella tree clean. — invoke to load the full rule protocol from knowledge/rules.

Chat contradicts file → chat wins, update same turn — invoke to load the full rule protocol from knowledge/rules.

.mcp.json + workspace anchors canonical. Globals synced by scripts/sync-globals.mjs. Fires grill-me on drift — invoke to load the full rule protocol from knowledge/rules.

Default grill-mode for >1 interpretation. Walk decision tree branch-by-branch. Surface assumptions, lock, act — invoke to load the full rule protocol from knowledge/rules.

Invoke grill-me or design Q&A — write results to knowledge/ EVERY locked answer (question stem + chosen option + rejected options + 'why') MUST land in knowledge/ in the same conversation. No locked answer may live only in chat history. The conversation context is the audit trail; the decision file is the durable truth. — invoke to load the full rule protocol from knowledge/rules.

Every session reads knowledge before acting, writes decisions into knowledge/ as CURRENT TRUTH (not historical logs), and deletes obsoleted content same-turn. Knowledge files are snapshots of what IS, not journeys of how we got here. — invoke to load the full rule protocol from knowledge/rules.

Steph Ango obsidian-skills cloned to ~/skill-sources/; 5 sub-skills symlinked global — invoke to load the full rule protocol from knowledge/rules.

Everything worth remembering lives in knowledge/ as an OKF file. Caveman style: terse, dense, no filler. README/AGENTS.md stay lean. — invoke to load the full rule protocol from knowledge/rules.

Durable prefs + locked decisions ? knowledge/ ONLY. Not mirrored to MEMORY.md — invoke to load the full rule protocol from knowledge/rules.

Linux/Ubuntu only on CI runners — invoke to load the full rule protocol from knowledge/rules.

MCP env vars sync via Win system env (local) + Smithery profile (cross-machine). Never commit values — invoke to load the full rule protocol from knowledge/rules.

MCQs: exactly 4 ranked options (Recommended + 2nd choice + 2 others). Never 3 or 2 — invoke to load the full rule protocol from knowledge/rules.

Migrating MEMORY.md to knowledge/: type field determines directory (rule?rules/, decision?decisions/, etc.) — invoke to load the full rule protocol from knowledge/rules.

Never delete empty placeholder repos without user OK — invoke to load the full rule protocol from knowledge/rules.

No feature branches on chirag127/* or chirag127/*. Commit to main. Branches only for upstream PRs — invoke to load the full rule protocol from knowledge/rules.

Local forks in repos/frk/* stay byte-identical to upstream main. Every local change files as an upstream PR immediately. — invoke to load the full rule protocol from knowledge/rules.

Only 3 Obsidian plugins: Terminal, Templater, Dataview — invoke to load the full rule protocol from knowledge/rules.

When a new tool/format/convention emerges in the OKF ecosystem, adopt fast + contribute upstream — first-mover shapes the standard. — invoke to load the full rule protocol from knowledge/rules.

Every agent must surface top-3 OKF concept files before answering any non-trivial prompt. Claude Code does it via UserPromptSubmit hook; other agents must run scripts/okf-prompt-lookup.py themselves. — invoke to load the full rule protocol from knowledge/rules.

Every AI adapter uses OpenAI SDK schema — invoke to load the full rule protocol from knowledge/rules.

Agent output terse, uses acronyms freely (OKF, MCQ, MCP, LOC, CF, GH, PWA, SERP). Expand only if non-obvious — invoke to load the full rule protocol from knowledge/rules.

Parallel fan-out via background subagents — invoke to load the full rule protocol from knowledge/rules.

Grill sessions logged at batch granularity — one log per session. Lives in knowledge/log/grills/ — invoke to load the full rule protocol from knowledge/rules.

PS 5.1 without UTF-8 BOM reads as Windows-1252. Em-dashes + smart quotes break parser. Use ASCII hyphens + straight quotes in .ps1 — invoke to load the full rule protocol from knowledge/rules.

Push to main by default without explicit say-so — invoke to load the full rule protocol from knowledge/rules.

Claims about external project gap must be backed by Read of source file — not grep alone — invoke to load the full rule protocol from knowledge/rules.

Optimize pinned repos + contribution graph for recruiters — invoke to load the full rule protocol from knowledge/rules.

Every repo works independently when cloned alone — invoke to load the full rule protocol from knowledge/rules.

searxng + duckduckgo + open-websearch. 10 engines, no API key. Try in order on failure — invoke to load the full rule protocol from knowledge/rules.

Decisions written to knowledge/ same session — invoke to load the full rule protocol from knowledge/rules.

GitHub issues and comments: short, factual, no filler. Every unverified claim is a potential hallucination. Min words = min risk. — invoke to load the full rule protocol from knowledge/rules.

Tests parallel, pnpm install -r from master — invoke to load the full rule protocol from knowledge/rules.

Website/API/tool/install failure: try 3+ alternatives before reporting blocker — invoke to load the full rule protocol from knowledge/rules.

pnpm mandatory across oriz family — invoke to load the full rule protocol from knowledge/rules.

User prefers atomic split over consolidation — invoke to load the full rule protocol from knowledge/rules.

User prefers deletion over archive for superseded repos — invoke to load the full rule protocol from knowledge/rules.

User prefers per-tool brand over family chrome — invoke to load the full rule protocol from knowledge/rules.

Same slug across GitHub repo and npm package — invoke to load the full rule protocol from knowledge/rules.

User prefers strict-no-toggle for locked rules — invoke to load the full rule protocol from knowledge/rules.

User prefers wider topical coverage over narrow SEO — invoke to load the full rule protocol from knowledge/rules.

Windows Terminal tabs launched via wt new-tab + a -Command string don't reliably inherit the User PATH. Always use the absolute path to npm.cmd / pnpm.cmd / python.exe etc. when constructing startup scripts. — invoke to load the full rule protocol from knowledge/rules.

Each locked decision ? knowledge/ + committed as discrete unit. Not batched end-of-session — invoke to load the full rule protocol from knowledge/rules.