# Ccc Browse > Visual, filterable catalog of every CC Commander skill, agent, and command. Cascades from broad category picker into 4-at-a-time detail cards. Use when the user types… - Skill: `kevinzai/ccc-browse` (Agent Skill) - Install (CLI): `npx skillmds@latest add kevinzai/ccc-browse` - Raw SKILL.md: https://api.skillmd.com/api/skills/kevinzai/ccc-browse/raw - Safety review: pending - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: AI & ML - Author: KevinZai (https://skillmd.com/u/kevinzai) - Updated: 2026-09-17 - Page: https://skillmd.com/skills/kevinzai/ccc-browse --- # /ccc-browse — Skill + Agent Catalog A click-first browser for the entire CC Commander surface: 11 CCC domains, 9 workflow skills, 22 specialist agents, plus meta skills. User never scrolls a wall of markdown — we cascade through `AskUserQuestion` 4-at-a-time. ## Response shape (EVERY time) Output exactly these three sections in order: ### 1. Brand header (one line, markdown) ``` **CC Commander** · v{VERSION} · Browser · 60 skills · 22 agents · 11 domains ``` Read `VERSION` from `${CLAUDE_PLUGIN_ROOT}/.claude-plugin/plugin.json`. ### 2. Context strip (one paragraph, markdown) Count categories live with a single Bash call: - `ls ${CLAUDE_PLUGIN_ROOT}/skills/ | wc -l` → total skills - `ls ${CLAUDE_PLUGIN_ROOT}/agents/*.md | wc -l` → agents - `ls ${CLAUDE_PLUGIN_ROOT}/skills/ccc-* -d 2>/dev/null | wc -l` → ccc-* skills Render: > 🧭 Catalog: `` skills · `` agents · `` CCC domains · filter by category below. ### 3. The picker — `AskUserQuestion` with the Cockpit + 4 categories **Never dump a numbered list. Never ask the user to type a name.** Call `AskUserQuestion`: ``` question: "What do you want to browse?" header: "CC Commander catalog" multiSelect: false options: - label: "🎛️ Open the Commander Cockpit — the whole catalog, clickable" description: "Generate and publish the searchable, offline catalog as a living Artifact page." preview: "Every Commander skill, agent, idea, and prompt pattern in one private client-side page." - label: "🎯 Domains (11)" description: "MEGA skills — design, marketing, saas, devops, seo, testing, security, data, research, mobile, makeover." preview: "Each domain bundles 8-45 sub-skills behind one entry point. Pick one to drill in." - label: "🛠️ Workflows (9)" description: "Step-by-step skills — plan, build, review, ship, start, learn, xray, linear, fleet." preview: "These are the click-first workflows you'll use 80% of the time." - label: "🤖 Agents (15)" description: "Specialist personas — architect, builder, debugger, reviewer, qa, security, designer, devops, pm, writer, researcher, data, content, perf, fleet." preview: "Each agent has a persona voice. Delegate to one from any /ccc flow." - label: "🗺️ Show everything as grid" description: "One-shot overview — sorted table of all skills + agents with one-line descriptions." preview: "No cascading. Dump once and let the user scan. Use when they're window-shopping." ``` Recommendation (⭐): - User asks for the full, visual, searchable, or offline catalog → ⭐ "🎛️ Open the Commander Cockpit — the whole catalog, clickable" - First-time browser (no prior `/ccc-browse` use in session) → ⭐ "Workflows (9)" — most actionable - User came from `/ccc-start` → ⭐ "Domains (11)" - Argument `all` passed → skip picker, go straight to grid ## Handle the selection ### Commander Cockpit → generate and publish the living page 1. Use the same stable path on every run: `/commander-cockpit.html`. Do not add timestamps or alternate filenames; republishing the same file preserves the living page URL. **`/ccc-console publish` deliberately writes to this same path** (v7.4.0, Kevin's call): the Cockpit and the console snapshot share one living Commander page rather than minting a fifth URL, so keep the favicon `🎛️` in both and never move this path. 2. Run the generator — **prefer the in-plugin copy first** (works for every install, marketplace or dev checkout), falling back to the repo-root shim only if that path doesn't exist: ```bash if [ -f "${CLAUDE_PLUGIN_ROOT}/scripts/build-cockpit.mjs" ]; then node "${CLAUDE_PLUGIN_ROOT}/scripts/build-cockpit.mjs" --out /commander-cockpit.html else node scripts/build-cockpit.mjs --out /commander-cockpit.html fi ``` (v7.3.0: the real builder ships inside the plugin at `${CLAUDE_PLUGIN_ROOT}/scripts/build-cockpit.mjs` — see W2+/codex 7. A marketplace-only install still produces a full Cockpit; it just won't have the 467-skill ecosystem catalog to merge in — the generator degrades that gracefully, never crashing.) 3. Publish that exact file with the Artifact tool using: - title: `Commander Cockpit` - favicon: `🎛️` - description: `Every Commander skill, agent, idea and prompt pattern — clickable, searchable, offline.` - capabilities: `{downloads: true}` — enables the ⬇ Save button for edited agent profiles; the page feature-detects this capability and hides the button elsewhere. 4. **First publish this session** → ask before publishing (skill/agent names and any local telemetry the page bakes in leave the machine for the artifact URL). **Every later run of `/ccc-browse`** → invoking the skill again IS the refresh consent: republish the same file path to the same URL without re-asking. Say that the page is fully client-side and nothing leaves the page beyond that one publish. To refresh later: run `/ccc-browse` again — same URL updates in place; viewers reload. If neither generator path exists (very old cached plugin version), say so and fall back to the existing category browse flow below; do not fabricate a Cockpit artifact. > **v7.4.0 — the Cockpit is the console's artifact-side catalog.** The Cockpit's heavy client-side tabs (Enhance, Prompts, Ideas, Agent manager) stay here, on a full-width page where they work; `/ccc-console`'s Launch tab is the compact, one-click version for the 680px inline widget. Same product, two surfaces — mention `/ccc-console` when the user wants the live agent/usage/safety picture rather than the catalog. > **v7.3.0:** the Cockpit's 8th tab, **Prompts**, is a searchable browser across 4 sources — Anthropic's Claude Code docs prompt library, the CCC field-tested prompt library (+ patterns + course modules), the repo's `prompts/` templates, and a ReadyIQ teaser (agent names + descriptions only). Each entry offers Copy and "✨ Enhance" (hands it straight to the Enhance tab). > **On Codex:** some panels may be empty on Codex today — Safety's failure hotspots, Mission Control's agent roster, and Usage's savings hero are fed from Claude-only hooks right now. A follow-up workstream wires the matching Codex telemetry. ### Domains (11) → cascade Read `${CLAUDE_PLUGIN_ROOT}/skills/` glob for `ccc-` directories excluding meta (`ccc`, `ccc-start`, `ccc-browse`, `ccc-plan`, etc.). First `AskUserQuestion` (4 of 11): ``` - 🎨 ccc-design · 41 sub-skills · UI/UX, animation, a11y - 📢 ccc-marketing · 46 sub-skills · CRO, content, SEO, growth - 💼 ccc-saas · 21 sub-skills · auth, billing, multi-tenant - ⋯ Next 4 domains ``` Second `AskUserQuestion` (4 of 11): ``` - 🚀 ccc-devops · 21 sub-skills · CI/CD, deploys, infra - 🔎 ccc-seo · 20 sub-skills · technical + AI search - 🧪 ccc-testing · 15 sub-skills · TDD, E2E, coverage - ⋯ Next 3 domains ``` Third `AskUserQuestion` (3 of 11 + back): ``` - 🔐 ccc-security · 8 sub-skills · OWASP, secrets, CVE - 📊 ccc-data · 8 sub-skills · SQL, ETL, viz - 🔬 ccc-research · 8 sub-skills · competitive, lit review - ↩️ Back to categories ``` (One more pass for ccc-mobile + ccc-makeover.) On pick → invoke that domain's skill (e.g. `ccc-design`) which runs its own sub-picker. ### Workflows (9) → cascade Workflows = the click-first skills. Cascade 4-at-a-time: First picker: ``` - 🚀 /ccc-start · first-run onboarding - 📋 /ccc-plan · feature planning flow - 🔨 /ccc-build · scaffold web/API/CLI/mobile - 🔍 /ccc-review · audit diff/security/perf ``` Second picker: ``` - 🚢 /ccc-ship · deploy + PR workflow - 🎓 /ccc-learn · extract patterns from session - 🩻 /ccc-xray · full project health scorecard - 📌 /ccc-linear · Linear issue board ``` Third picker: ``` - ⚙️ /ccc-fleet · parallel agent batch work - ↩️ Back to categories ``` On pick → invoke the skill inline. ### Agents (15) → cascade Read `${CLAUDE_PLUGIN_ROOT}/agents/*.md` frontmatter. Cascade 4-at-a-time across 4 pickers: Pass 1: architect, builder, debugger, reviewer Pass 2: qa-engineer, security-auditor, performance-engineer, designer Pass 3: devops-engineer, product-manager, technical-writer, researcher Pass 4: data-analyst, content-strategist, fleet-worker, ↩️ Back Each option: - `label: " "` - `description: ""` - `preview: "Best for: . Model: . Trigger: /ccc-."` On pick → offer two actions via second `AskUserQuestion`: ``` - "Delegate a task to now" → prompts for task, calls Agent tool - "Show me the full persona definition" → Read the .md file, render it - "Which flows use this agent?" → Grep for the agent name in skills/ - "↩️ Back" ``` ### Show everything as grid → one-shot dump Render a single markdown table, sorted alphabetically: ```markdown | Type | Name | One-liner | |------|------|-----------| | 🤖 agent | architect | System design, trade-offs, tech selection | | 🎯 domain | ccc-design | 41 skills — UI/UX, animation, a11y | | 🛠️ workflow | ccc-build | Scaffold web/API/CLI/mobile from spec | | ... | ... | ... | ``` After the grid, one `AskUserQuestion` with 4 follow-ups: "Launch one" / "Filter by keyword" / "Browse categories instead" / "Done". ## Anti-patterns — DO NOT do these - ❌ Render a single markdown list with 30+ rows and ask the user to type a name - ❌ Dump raw HTML in code blocks expecting artifact rendering - ❌ Output ASCII banners or box-drawing tables for the catalog - ❌ Put more than 4 options in a cascade `AskUserQuestion` — the five-option root picker is the sole exception; always include a "Back" option on cascades - ❌ Read every agent .md eagerly up front — read on demand only - ❌ Reference legacy CLI browse commands — this is Desktop-plugin only ## Brand rules - **Always read `VERSION` from plugin.json** — never hardcode. - **One-line descriptions only** in picker options — detail goes in the `preview` field. - **Back option on every cascade past level 1** — lets users un-dive without restarting. - **Counts always live** — read from filesystem, never hardcode "39 sub-skills" if the count can shift. ## Dispatch after selection After a user picks any skill or agent, the handler: 1. Reads the SKILL.md (or agent .md) frontmatter 2. Echoes one paragraph of what's about to happen 3. Invokes the skill/agent inline Never require the user to leave `/ccc-browse` and type another slash command. We stay in-flow. ## Tips for the agent executing this skill 1. Parallel the three count Bash calls into one. Saves a turn. 2. When cascading, maintain a simple state in your responses: "Page 1 of 3 · 4 of 11 shown". 3. If `argument-hint` is `domains`, `workflows`, `agents`, or `all`, skip the root picker and jump to that path. 4. Agent descriptions are in `${CLAUDE_PLUGIN_ROOT}/agents/*.md` frontmatter — read the whole file only on user drill-in. --- **Bottom line:** one-click Cockpit + four top-level categories → cascaded 4-at-a-time pickers → user drills in → we dispatch to the target skill. No scrolling, no typing, no numbered lists. --- > ⚙️ **Fable contract:** plan before build · verifier ≠ worker · prove before alarm · loops need gates · leave durable state — `rules/fable-method.md`