claude-code-boss
Semantic knowledge base (Brain) + curated shell execution + lightweight learning for Claude Code. Auto-indexed KB with vector search, anti context-bloat curation, and advisory pattern/correction capture. Relies on native Agent/Workflow tools for orchestration.
By Allan SantosLicense: MIT0 GitHub starsUpdated 6 days ago
Directory evidence
- Runtimes
- Claude Code
- Parsed components
- 9 skill or MCP entries
- Source updated
- Sep 17, 2026
- Manifest status
- Canonical path parsed
The directory validates manifest shape and source location. It does not execute the plugin or provide a security endorsement. Review the indexing methodology →
Install claude-code-boss for Claude Code
claude plugin marketplace add IchenDEV/agent-plugin-mkt
claude plugin marketplace update agent-plugin-marketplace
claude plugin install claude-code-boss@agent-plugin-marketplacePaste and run these commands in a terminal with Claude Code. They add and refresh the PluginsMP catalog, then install this plugin.
The installer fetches third-party code from the source repository shown on this page. This directory validates manifest structure and source location, but does not perform a security audit; review the manifest, components, and source before installing.
Get the source manually
git clone https://github.com/AllanSantos-DV/claude-code-pluginsClone the source repository, then follow its setup instructions to add the plugin to a compatible client. The plugin root is claude-code-boss/.
Plugin files
├── .claude-plugin/plugin.json├── skills/brain-knowledge/SKILL.md├── skills/brain-research/SKILL.md├── skills/config-dashboard/SKILL.md├── skills/curation-script-pattern/SKILL.md├── skills/embedder-switch/SKILL.md├── skills/native-session-management/SKILL.md├── skills/plugin-install/SKILL.md├── skills/shell-execution/SKILL.md└── .mcp.json
Included Skills8
Knowledge Base — SQLite vector store with semantic search via Transformers.js (or Ollama/Voyage). Auto-indexed by hooks, auto-retrieved before every action. Scoped by project, searchable cross-project.
Brain Research methodology — multi-source fan-out research, quality gating, source authority ranking, and knowledge base persistence patterns.
**WORKFLOW SKILL** — Local plugin configuration dashboard. USE FOR: opening the config UI, viewing/changing brain-config, hooks, searching the knowledge base. DO NOT USE FOR: editing config files directly via Write tool when user asks for dashboard (use the dashboard instead), non-plugin configuration.
Pattern for curated shell scripts — output contract OK/FAIL, shells.json schema, refine-existing-first priority. Used when the Stop hook signals noisy commands need curation (in-loop, no subagent). Language-agnostic — scripts can be any executable (.mjs, .ps1, .sh, .py, etc.) as long as they honor the output contract.
**WORKFLOW SKILL** — Safely switch the brain KB embedder (provider/model) end-to-end: validate via Test/Install endpoint, persist config, run mandatory re-embed, advise on hook-driven restart. USE FOR: 'trocar pra voyage-4-lite', 'usar ollama', 'mudar modelo do brain', 'configurar embedder'. DO NOT USE FOR: KB retrieval thresholds, hooks toggle, or other non-embedder config changes.
Enforces use of Claude Code's native session management (--continue, --resume, /checkpoint, /rewind, /branch, /bg, /rename, /export, claude agents) instead of workarounds. Agents recommend the right native command at the right moment.
**WORKFLOW SKILL** — Install, set up, and troubleshoot the claude-code-boss plugin on any machine. USE FOR: installing the plugin, fixing 'plugin only works on the author's machine', missing node_modules, better-sqlite3 / node-gyp / 'gyp ERR' / Visual Studio Build Tools errors, 'brain not working', SQLite backend issues, embedding model not downloaded, patterns not becoming skills / recurrence stuck, semantic search not working, checking Node version, CLAUDE_PLUGIN_ROOT setup, verifying the install. DO NOT USE FOR: opening the config dashboard (use config-dashboard), switching the embedder model (use embedder-switch), authoring new skills.
Curated shell execution protocol — how curation-guard (PreToolUse), curation-detect (PostToolUse), and curation-stop (Stop) work together to keep raw command output out of the context window. Language-agnostic — curated scripts may be .mjs, .ps1, .sh, .py, etc.
MCP servers1
- url
- http://127.0.0.1:58217/mcp
Plugin manifests1
{
"name": "claude-code-boss",
"displayName": "Claude Code Boss — Brain & Curation",
"description": "Semantic knowledge base (Brain) + curated shell execution + lightweight learning for Claude Code. Auto-indexed KB with vector search, anti context-bloat curation, and advisory pattern/correction capture. Relies on native Agent/Workflow tools for orchestration.",
"author": {
"name": "Allan Santos"
},
"homepage": "https://github.com/AllanSantos-DV/claude-code-plugins",
"license": "MIT",
"keywords": [
"knowledge-base",
"semantic-search",
"curation",
"research"
],
"userConfig": {
"language": {
"type": "string",
"title": "Output Language",
"description": "Language for user-facing communication (pt-BR or en)",
"default": "pt-BR",
"required": true
}
}
}For maintainers
If you maintain this plugin, link to this source-backed listing from your README so users can review its manifest and indexed components.
[claude-code-boss on Agent Plugins Marketplace](https://pluginsmp.com/plugins/claude-code-boss)