feat: 8 themed skill packs + /toolkit command + session hook + split roadmap

.claude-plugin/marketplace.json

@@ -8,9 +8,65 @@
     {
       "name": "agent-toolkit",
       "source": "./plugins/agent-toolkit",
-      "description": "One bundled plugin shipping Basti's vetted skills (second-brain, yuno-cleaner, mmx-cli, drucker-experte), subagents (coder, perf-tuner, security-auditor), and MCP server declarations (github). Enable once to sync the whole toolkit.",
+      "description": "One bundled plugin shipping Basti's vetted skills (129 skills in 8 themed packs), 17 subagents (coder, perf-tuner, security-auditor, zc-* team), 9 MCP server declarations (github, gmail, calendar, …), a /toolkit browse command, and a session-start status hook. Enable once to sync the whole toolkit.",
       "category": "productivity",
       "homepage": "https://github.com/Toqsick/my-agent-tools/tree/main/plugins/agent-toolkit"
+    },
+    {
+      "name": "agent-toolkit-core",
+      "source": "./plugins/agent-toolkit",
+      "description": "[PACK STUB — themed split pending; see PACKS.md] Core & Personal: the daily-driver layer — Obsidian second-brain, Yuno cleaner, 3D printing, Yuno team orchestration/routing/preferences, model selection, folder tidy, daily briefing, skill↔MCP router. Installing this handle currently installs the full agent-toolkit plugin.",
+      "category": "productivity",
+      "homepage": "https://github.com/Toqsick/my-agent-tools/tree/main/plugins/agent-toolkit/packs/core"
+    },
+    {
+      "name": "agent-toolkit-hermes-dev",
+      "source": "./plugins/agent-toolkit",
+      "description": "[PACK STUB — themed split pending; see PACKS.md] Hermes / Yuno Platform: CLI internals, gateway protocol/clients, mobile clients, Ariadne memory, gateway adapters, desktop plugins, messaging gateways, themes, ModelHub. Installing this handle currently installs the full agent-toolkit plugin.",
+      "category": "development",
+      "homepage": "https://github.com/Toqsick/my-agent-tools/tree/main/plugins/agent-toolkit/packs/hermes-dev"
+    },
+    {
+      "name": "agent-toolkit-cybersecurity",
+      "source": "./plugins/agent-toolkit",
+      "description": "[PACK STUB — themed split pending; see PACKS.md] Cybersecurity: 50 defensive-security and DFIR skills — CIS hardening, Docker/K8s, network hunting (Zeek/Suricata/Wireshark), forensics (Volatility/IR), compliance/supply-chain (SBOM/SLSA/gitleaks). Installing this handle currently installs the full agent-toolkit plugin.",
+      "category": "security",
+      "homepage": "https://github.com/Toqsick/my-agent-tools/tree/main/plugins/agent-toolkit/packs/cybersecurity"
+    },
+    {
+      "name": "agent-toolkit-methodology",
+      "source": "./plugins/agent-toolkit",
+      "description": "[PACK STUB — themed split pending; see PACKS.md] Engineering Methodology: the Superpowers workflow set (brainstorm→plan→TDD→debug→verify→finish), subagent-driven development, ZCode SubAgent Team, Queen-Bee swarm dispatch, multi-agent master workflow. Installing this handle currently installs the full agent-toolkit plugin.",
+      "category": "development",
+      "homepage": "https://github.com/Toqsick/my-agent-tools/tree/main/plugins/agent-toolkit/packs/methodology"
+    },
+    {
+      "name": "agent-toolkit-media",
+      "source": "./plugins/agent-toolkit",
+      "description": "[PACK STUB — themed split pending; see PACKS.md] Media & Generation: MiniMax mmx CLI, agent builder, crypto trading, DOCX/PDF, Nano Banana Pro image gen. Installing this handle currently installs the full agent-toolkit plugin.",
+      "category": "creative",
+      "homepage": "https://github.com/Toqsick/my-agent-tools/tree/main/plugins/agent-toolkit/packs/media"
+    },
+    {
+      "name": "agent-toolkit-docs-web-research",
+      "source": "./plugins/agent-toolkit",
+      "description": "[PACK STUB — themed split pending; see PACKS.md] Docs, Web & Research: PPT/McKinsey decks, research papers, frontend design, web scraping, Excel, n8n, SEO/GEO, job hunting, sales Power Maps, SaaS niches, NotebookLM, knowledge digest, prompt engineering. Installing this handle currently installs the full agent-toolkit plugin.",
+      "category": "productivity",
+      "homepage": "https://github.com/Toqsick/my-agent-tools/tree/main/plugins/agent-toolkit/packs/docs-web-research"
+    },
+    {
+      "name": "agent-toolkit-computer-use",
+      "source": "./plugins/agent-toolkit",
+      "description": "[PACK STUB — themed split pending; see PACKS.md] Computer-Use & GreyHack: GreyHack Computer-Use suite plus game and desktop-window reconnaissance. Installing this handle currently installs the full agent-toolkit plugin.",
+      "category": "automation",
+      "homepage": "https://github.com/Toqsick/my-agent-tools/tree/main/plugins/agent-toolkit/packs/computer-use"
+    },
+    {
+      "name": "agent-toolkit-dev-essentials",
+      "source": "./plugins/agent-toolkit",
+      "description": "[PACK STUB — themed split pending; see PACKS.md] Dev Essentials: debugging patterns, defensive programming, config-propagation bugs, reference-architecture research, open-source extraction, stealth web scraping, web-content recon, competitive-landscape research, ClickHouse best practices. Installing this handle currently installs the full agent-toolkit plugin.",
+      "category": "development",
+      "homepage": "https://github.com/Toqsick/my-agent-tools/tree/main/plugins/agent-toolkit/packs/dev-essentials"
     }
   ]
-}
+}

INDEX.json

@@ -1,6 +1,6 @@
 {
   "schemaVersion": "2.0",
-  "generated_at": "2026-07-24T14:16:22+02:00",
+  "generated_at": "2026-08-05T03:41:38+00:00",
   "repo": "Toqsick/my-agent-tools",
   "how_to_use": "See ROUTING.md. Fetch this file, match a task against each skill's triggers/tags/category, rank, then fetch the chosen skill's `path`. Installed skills also load in-session as their `namespace`. Multi-step work: see `workflows`.",
   "counts": {

NAVIGATION.md

@@ -1,7 +1,7 @@
 # NAVIGATION — my-agent-tools skill index
 
 > **Generated by `scripts/build_index.py` — do not hand-edit.** The machine-readable source of truth is [`INDEX.json`](INDEX.json); routing rules are in [`ROUTING.md`](ROUTING.md).
-> Snapshot: HEAD `2026-07-24T14:16:22+02:00` · **1373 skills** (129 installed + 1244 library) · 17 agents · 6 workflows · 43 categories
+> Snapshot: HEAD `2026-08-05T03:41:38+00:00` · **1373 skills** (129 installed + 1244 library) · 17 agents · 6 workflows · 43 categories
 
 Two tiers: **installed** skills load into every Claude Code session as `agent-toolkit:<name>`; **library** skills are browsable reference, fetched on demand by `path` via the GitHub MCP (never auto-loaded).
 
@@ -171,6 +171,21 @@ Two tiers: **installed** skills load into every Claude Code session as `agent-to
 | `yuno-user-preferences` | `agent-toolkit:yuno-user-preferences` | Use when user asks how to apply Basti’s working-style preferences, honest testing standard, recommendation-first decisio |
 | `zcode-subagent-team` | `agent-toolkit:zcode-subagent-team` | ZCode SubAgent Team — a Hermes-Kanban multi-lane agent swarm where a Queen (GLM ZCode orchestrator/verifier) delegates b |
 
+## Skill Packs (installed-skill grouping)
+
+The 129 installed skills are grouped into 8 themed packs. Browse with the `/toolkit` command; see [PACKS.md](PACKS.md) for the pack roadmap.
+
+| Pack | Title | Category | Skills |
+|---|---|---|---|
+| [`core`](plugins/agent-toolkit/packs/core/README.md) | Core & Personal | productivity | 10 |
+| [`hermes-dev`](plugins/agent-toolkit/packs/hermes-dev/README.md) | Hermes / Yuno Platform | development | 18 |
+| [`cybersecurity`](plugins/agent-toolkit/packs/cybersecurity/README.md) | Cybersecurity | security | 50 |
+| [`methodology`](plugins/agent-toolkit/packs/methodology/README.md) | Engineering Methodology | development | 18 |
+| [`media`](plugins/agent-toolkit/packs/media/README.md) | Media & Generation | creative | 6 |
+| [`docs-web-research`](plugins/agent-toolkit/packs/docs-web-research/README.md) | Docs, Web & Research | productivity | 15 |
+| [`computer-use`](plugins/agent-toolkit/packs/computer-use/README.md) | Computer-Use & GreyHack | automation | 3 |
+| [`dev-essentials`](plugins/agent-toolkit/packs/dev-essentials/README.md) | Dev Essentials | development | 9 |
+
 ## Library (browsable via MCP, fetch by path)
 
 Not session-loaded. Discover via [`INDEX.json`](INDEX.json) → match → fetch the skill's `path`. Category counts:

PACKS.md

@@ -0,0 +1,86 @@
+# Skill Packs
+
+The `agent-toolkit` plugin ships **129 installed skills** grouped into **8 themed packs**. The
+packs are a navigation/grouping layer over the installed skills — every skill still loads in
+every session (it's one plugin), and is still invokable as `agent-toolkit:<name>`. Packs just
+make the wall of 129 skills scannable: by domain, by count, by "when to use this pack".
+
+- **Canonical manifest:** [`plugins/agent-toolkit/packs/manifest.json`](plugins/agent-toolkit/packs/manifest.json)
+  (generated-validated by `scripts/build_packs.py`; never hand-edit the bundles).
+- **Per-pack detail:** `plugins/agent-toolkit/packs/<pack>/README.md` (one per pack).
+- **Routing bundles:** `routing/bundles/<pack>.yaml` (one per pack, emitted from the manifest).
+- **Browse live:** run `/toolkit` (overview), `/toolkit <pack>` (a pack's skills), or
+  `/toolkit doctor` (health check).
+
+## The 8 packs
+
+| Pack | Title | Category | Skills | When to use |
+|---|---|---|---:|---|
+| `core` | Core & Personal | productivity | 10 | Daily-driver layer: Obsidian second-brain, system cleanup, 3D printing, Yuno team orchestration/routing/preferences, model selection, folder tidy, daily briefing, skill↔MCP router. |
+| `hermes-dev` | Hermes / Yuno Platform | development | 18 | Building on the Hermes/Yuno platform: CLI internals, gateway protocol/clients, mobile clients, Ariadne memory, gateway adapters, desktop plugins, messaging gateways, themes, ModelHub. |
+| `cybersecurity` | Cybersecurity | security | 50 | Defensive security & DFIR: CIS hardening, Docker/K8s, network hunting (Zeek/Suricata/Wireshark), forensics (Volatility/IR), compliance & supply-chain (SBOM/SLSA/gitleaks). |
+| `methodology` | Engineering Methodology | development | 18 | The Superpowers workflow set (brainstorm→plan→TDD→debug→verify→finish), subagent-driven development, ZCode SubAgent Team, Queen-Bee swarm dispatch, multi-agent master workflow. |
+| `media` | Media & Generation | creative | 6 | MiniMax `mmx` CLI, MiniMax agent builder, crypto trading, DOCX/PDF, Nano Banana Pro image gen. |
+| `docs-web-research` | Docs, Web & Research | productivity | 15 | Document gen (PPT/McKinsey/research papers), frontend design, web scraping, Excel, n8n, SEO/GEO, business research (job hunter, sales Power Maps, SaaS niches), NotebookLM, knowledge digest, prompt engineering. |
+| `computer-use` | Computer-Use & GreyHack | automation | 3 | Desktop/game GUI automation: GreyHack Computer-Use suite + game and desktop-window reconnaissance. |
+| `dev-essentials` | Dev Essentials | development | 9 | Engineering originals: debugging patterns, defensive programming, config-propagation bugs, reference-architecture research, open-source extraction, stealth web scraping, web-content recon, competitive-landscape research, ClickHouse best practices. |
+
+**Total: 129 skills across 8 packs** — a clean partition (every installed skill in exactly one
+pack, validated by `scripts/build_packs.py`, which exits non-zero on a partition error).
+
+## How packs relate to the plugin today
+
+Right now there is **one plugin** (`agent-toolkit`) and **one marketplace entry** that installs
+all 129 skills. The packs are metadata — `manifest.json` + per-pack READMEs + routing bundles —
+not separate install targets. The 8 themed marketplace entries in
+`.claude-plugin/marketplace.json` (`agent-toolkit-core`, `-hermes-dev`, …) are **stubs**: each
+points at the same `./plugins/agent-toolkit` source, so installing any themed handle today
+installs the full toolkit. They exist so the split below is a one-line source repoint, not a
+restructure.
+
+> ⚠️ **Fallback note:** if Claude Code's `/plugin marketplace` rejects multiple entries sharing
+> one `source` (the plugin's internal `name` is `agent-toolkit` regardless of the marketplace
+> entry name), keep only the canonical `agent-toolkit` entry and treat this file as the durable
+> record of the 8 future targets. The stubs are a convenience, not a requirement.
+
+## Future split — "install only what you need"
+
+When the toolkit grows or a user wants a lighter install, split each pack into its own plugin
+directory and repoint its marketplace entry. The mechanical steps:
+
+1. For each pack `P`, create `plugins/agent-toolkit-P/` with a `.claude-plugin/plugin.json` whose
+   `skills` array lists only that pack's skills (copy them from
+   `plugins/agent-toolkit/skills/<name>/` or symlink — keep copies self-contained per the repo's
+   no-symlink-outside-repo rule).
+2. Repoint the matching marketplace stub:
+   ```jsonc
+   // .claude-plugin/marketplace.json
+   {
+     "name": "agent-toolkit-cybersecurity",          // already a stub entry
+     "source": "./plugins/agent-toolkit-cybersecurity", // was: "./plugins/agent-toolkit"
+     "description": "Cybersecurity pack: 50 DFIR/defensive skills …",
+     "category": "security",
+     "homepage": "https://github.com/Toqsick/my-agent-tools/tree/main/plugins/agent-toolkit-cybersecurity"
+   }
+   ```
+3. Keep the monolithic `agent-toolkit` entry as an "install everything" convenience, or retire it
+   once the 8 themed plugins are stable.
+4. Re-run `python3 scripts/build_index.py` (the manifest is the source of truth for pack
+   membership; `build_packs.py` will re-emit bundles and validate the partition).
+
+**Migration order (recommended):** split the largest/most-self-contained pack first
+(`cybersecurity`, 50 skills — least overlap with daily-driver work), then `hermes-dev` (18) and
+`methodology` (18). The small packs (`media`, `computer-use`, `dev-essentials`) can stay bundled
+until there's a reason to split them.
+
+## Maintenance
+
+- **Adding a skill:** drop it under `plugins/agent-toolkit/skills/<name>/`, add it to the `skills`
+  array in `plugin.json`, **and** add it to exactly one pack's `skills` list in
+  `packs/manifest.json`. Then `python3 scripts/build_index.py` (which calls `build_packs`) —
+  it validates the partition (union == installed set, no dupes) and re-emits the routing
+  bundles. A non-zero exit means the manifest is out of sync with the installed skills; fix the
+  manifest, don't silence the check.
+- **Repacking:** to re-theme the packs, edit only `packs/manifest.json` and re-run
+  `build_index.py`. Per-pack READMEs and routing bundles are derived; regenerate, don't hand-edit.
+- **Never** hand-edit `routing/bundles/*.yaml`, `INDEX.json`, or `NAVIGATION.md`.

README.md

@@ -51,9 +51,12 @@ Installed = curated fast-path; library = the comprehensive Hermes arsenal, pulle
 │   └── agent-toolkit/
 │       ├── .claude-plugin/
 │       │   └── plugin.json        # plugin manifest (registers the installed skills)
-│       ├── skills/                # 128 installed skills (SKILL.md bundles) — session-loaded
+│       ├── skills/                # 129 installed skills (SKILL.md bundles) — session-loaded
 │       ├── agents/                # 17 subagents (coder, perf-tuner, security-auditor + 14 zc-*)
-│       └── .mcp.json              # MCP server declarations (8 servers)
+│       ├── commands/              # /toolkit — browse packs + health check (auto-discovered)
+│       ├── hooks/                 # SessionStart banner: skills · packs · agents · MCP counts
+│       ├── packs/                 # 8 themed skill packs (manifest.json + per-pack READMEs)
+│       └── .mcp.json              # MCP server declarations (9 servers)
 ├── library/                      # 1,244 browsable Hermes-arsenal skills (NOT session-loaded)
 │   └── <category>/<skill>/SKILL.md
 ├── routing/                       # zcode-skills MCP-aware routing metadata
@@ -63,10 +66,34 @@ Installed = curated fast-path; library = the comprehensive Hermes arsenal, pulle
 │   └── manifests/                  # imported provenance lockfiles
 ├── workflows/                    # machine-readable multi-skill workflow patterns
 └── scripts/
-    ├── build_index.py             # regenerates catalogs + routing metadata
+    ├── build_index.py             # regenerates catalogs + routing metadata (calls build_packs)
+    ├── build_packs.py             # validates packs/manifest.json + emits routing/bundles/*.yaml
     └── check-mcp.sh               # local MCP server health check
 ```
 
+## Skill packs (installed-skill grouping)
+
+The 129 installed skills are grouped into **8 themed packs** — a navigation layer over the one
+plugin, so the wall of skills becomes scannable by domain. Every skill still loads in every
+session and is invokable as `agent-toolkit:<name>`; packs just make browsing and routing
+easier. See [`PACKS.md`](PACKS.md) for the full pack index, the per-pack skill rosters, and the
+future "install only what you need" split roadmap.
+
+| Pack | Skills | Theme |
+|---|---:|---|
+| `core` | 10 | Daily-driver layer — Obsidian second-brain, Yuno cleaner, 3D printing, Yuno team orchestration, model selection, daily briefing, skill↔MCP router |
+| `hermes-dev` | 18 | Hermes/Yuno platform — CLI internals, gateway protocol/clients, mobile, Ariadne memory, adapters, themes |
+| `cybersecurity` | 50 | Defensive security & DFIR — CIS hardening, Docker/K8s, network hunting, forensics, compliance/supply-chain |
+| `methodology` | 18 | Superpowers + ZCode team orchestration — brainstorm→plan→TDD→debug→verify→finish, swarm dispatch |
+| `media` | 6 | MiniMax + generation — `mmx` CLI, agent builder, crypto trading, DOCX/PDF, Nano Banana Pro |
+| `docs-web-research` | 15 | Docs, web & research — PPT/decks, research papers, frontend design, scraping, Excel, n8n, SEO/GEO, business research |
+| `computer-use` | 3 | Computer-use & GreyHack — GUI automation + game/desktop-window reconnaissance |
+| `dev-essentials` | 9 | Engineering originals — debugging patterns, defensive programming, recon, extraction, ClickHouse |
+
+Browse live in any session with the `/toolkit` slash command: `/toolkit` (overview),
+`/toolkit <pack>` (a pack's skills), or `/toolkit doctor` (health check). A `SessionStart` hook
+prints a compact `skills · packs · agents · MCP servers` banner at session open.
+
 ### Skills
 
 **Seed set:**
@@ -232,7 +259,7 @@ with owner-agent + skills + exit criteria per phase): `superpower-10x-pipeline`,
 
 ### MCP servers
 
-All 8 servers are declared in [`plugins/agent-toolkit/.mcp.json`](plugins/agent-toolkit/.mcp.json).
+All 9 servers are declared in [`plugins/agent-toolkit/.mcp.json`](plugins/agent-toolkit/.mcp.json).
 Credentials are **always** referenced as `${ENV_VAR}` — never stored here. Copy [`.env.example`](.env.example) to `.env` and fill in your values.
 
 | Server | Transport | Purpose | Required env vars |
@@ -245,6 +272,7 @@ Credentials are **always** referenced as `${ENV_VAR}` — never stored here. Cop
 | `memory` | npx (`@modelcontextprotocol/server-memory`) | Persistent knowledge graph across sessions | `MEMORY_FILE_PATH` (default `~/.agent-memory/memory.json`) |
 | `puppeteer` | npx (`@modelcontextprotocol/server-puppeteer`) | Headless browser / web automation | — |
 | `sequential-thinking` | npx (`@modelcontextprotocol/server-sequential-thinking`) | Structured multi-step reasoning | — |
+| `basti-tools` | `uv run mcp-server-basti` (FastMCP/stdio, local) | Local system-status tools (`get_system_status`, `echo_tool`, `get_repo_info`) | — |
 
 #### Gmail + Google Calendar — Quick OAuth2 Setup