dong-ai-company

name: dong-ai-company description: "AI company engine — any domain, any output. CEO=200K context, smart model routing gateway, difficulty-aware pipeline, cross-project graph memory. pip install dong-ai." tags: [ai-company, self-learning, governance, column-memory, metacognition, orchestration] dong-plugin: true

Dong AI Company — Hermes Skill

The universal interface to Dong AI is the dong CLI. Any agent that can run shell commands can call it. This SKILL.md is the Hermes-specific binding; see CLAUDE.md / .cursorrules / README.md for other platforms.

Install once, call from any agent:

pip install dong-ai
dong setup
cp SKILL.md ~/.hermes/skills/dong-ai-company/

============================================

TOOL INDEX (agent-agnostic)

============================================

Each tool below works identically across all agents — just shell out to dong <command>.

============================================

=== Project Execution ===

• name: dong_make description: "Self-directed making for any domain. Tell the company what you want to make — a market report, a comic series, a business plan, a technical analysis, a marketing strategy. It researches the domain, proposes directions for your approval, deep-dives, executes, self-reviews, and delivers. Add --auto for fully autonomous mode." exec: "dong make '{{request}}' 2>&1 | tail -60"

• name: dong_run description: "Execute a complete project through the AI company governance pipeline. The CEO analyzes requirements, conducts red/blue team debate, recruits specialist workers, executes with self-healing and cross-review, then board-approves the result. The project is automatically debriefed afterward with lessons extracted and stored for future recall." exec: "dong run '{{request}}' 2>&1 | tail -50"

• name: dong_quick description: "Lightweight execution mode that bypasses the full CEO pipeline. Use for simple tasks that don't need debate, multiple workers, or board review. Direct LLM call with file read/write/run/web tools available. Faster than dong_run for straightforward requests." exec: "dong quick '{{request}}' 2>&1"

=== Code & File Workflow ===

• name: dong_analyze description: "Read and analyze any source file. Ask questions about code structure, logic, or behavior. Equivalent to dragging a file into a chat UI but runs in your terminal. Automatically reads .dong_rule for project context." exec: "dong analyze {{path}} '{{question}}' 2>&1 | tail -40"

• name: dong_edit description: "Read a file, propose changes based on your requirements, show a unified diff preview, and apply changes on confirmation. Includes syntax-aware code generation and respects project-level .dong_rule conventions." exec: "dong edit {{path}} '{{instruction}}' 2>&1"

• name: dong_debug description: "CI failure root cause analysis. Fetches logs from the latest failed GitHub Actions run, parses error locations, cross-references against graph memory for relevant symbols and dependencies, reads source code around failure points, and produces an LLM-generated root cause analysis with fix recommendations." exec: "dong debug 2>&1 | tail -80"

=== Company Runtime ===

• name: dong_company_start description: "Start a persistent 7x24 AI company instance in the background. Supports autonomous domain operation — describe any domain in natural language and the agent researches it, sets up monitoring strategies, and runs continuously. Configurable duration with --duration (30m/2h/1d) or --until (17:00)." exec: "dong company start --domain '{{domain}}' --duration {{duration}} 2>&1"

• name: dong_company_stop description: "Stop the running AI company instance. All domains are shut down, current state is saved." exec: "dong company stop 2>&1"

• name: dong_company_status description: "View the current status of the AI company — uptime, active domains, tasks completed, pending governance confirmations, and prediction accuracy statistics." exec: "dong company status 2>&1"

=== Metacognition & Governance ===

• name: dong_knowledge_map description: "Generate the organizational metacognition report. Shows what the AI company knows (domain expertise with scores), what it's learning (improvement trends), what strategies work best, and knowledge gaps (domains not yet explored)." exec: "dong company knowledge 2>&1"

• name: dong_governance_review description: "Run a governance self-review. Shows total decisions, confidence distribution, risk levels, and decision audit trail. Lists any pending high-risk actions that require human confirmation." exec: "dong company review 2>&1"

• name: dong_governance_confirm description: "Confirm a pending high-risk decision. When the SafetyGovernor blocks an action due to low confidence or high risk, use this to explicitly authorize it." exec: "dong company confirm 2>&1"

=== Context & Memory ===

• name: dong_graph_query description: "Query the cross-project graph memory. Lists all projects with their symbol and dependency counts, then drills into a specific project to show functions, classes, file locations, and dependency relationships." exec: "dong graph list 2>&1 && echo '---' && dong graph view {{project_id}} 2>&1"

• name: dong_session description: "List recent working sessions from persistent memory — session IDs, titles, message counts, timestamps. Then optionally view the last N messages of a specific session." exec: "dong session list 2>&1 && echo '---' && dong session view {{session_id}} 2>&1 | tail -30"

=== Plugins ===

• name: dong_plugin_install description: "Install an MCP plugin from the Dong AI plugin registry." exec: "dong plugin install {{name}} 2>&1"

• name: dong_mcp_discover description: "Discover and test all configured MCP servers — scans .mcp.json and Hermes config, connects to each server, and lists available tools with descriptions." exec: "dong mcp 2>&1"

• name: dong_gateway description: "Smart model routing gateway. View all providers with speed/quality/context scores, see which model is selected per task type. Supports dong gateway list, dong gateway set <provider>, dong gateway tier <provider> cheap|expensive|auto." exec: "dong gateway list 2>&1"

• name: dong_demo description: "Zero-config demo of cross-project graph memory. No API key needed, no network calls. Creates fake projects in an isolated SQLite DB and shows graph list, cross-project search, dependency analysis, and session resume." exec: "dong demo 2>&1"

=== System ===

• name: dong_status description: "Full system status: available models, Column Memory budget, graph memory statistics, active plugins, running company instances, and pending governance confirmations." exec: "dong detect 2>&1"

• name: dong_update description: "Check PyPI for a newer version and upgrade if available." exec: "dong update 2>&1"

Architecture

Any Agent (Hermes / Claude Code / Cursor / Copilot / Codex / etc.)
  │
  ├── shell out: dong make "a report"
  ├── shell out: dong run "build a CLI"
  ├── shell out: dong quick "fix this"
  ├── shell out: dong analyze path question
  └── shell out: dong company start --domain "monitor API"
        │
        ▼
   Dong AI Company Engine  ← pip install dong-ai
        │
        ├── Column Memory      (cross-project context)
        ├── Experience Engine  (learns from past projects)
        ├── SafetyGovernor     (confidence + risk gates)
        ├── Metacognition      (knowledge map + strategy evolution)
        └── Domain Runtimes    (7x24 autonomous monitoring)

The dong CLI is the universal contract. The Engine is the same regardless of which agent wraps it.

What Your Agent Gains

Calling from Any Agent

All tools are dong <command> shell invocations. Template parameters use {{ }} syntax — substitute with your agent's template engine.

Claude Code — copy CLAUDE.md from this repo:

cp CLAUDE.md .claude/

Cursor — copy .cursorrules from this repo:

cp .cursorrules .cursorrules

GitHub Copilot — create a .github/copilot-instructions.md:

The @dong CLI provides project execution, code analysis, company runtime,
metacognition, and memory capabilities. Use `dong <command>` to invoke.
See README.md for full tool documentation.

Any agent — shell out directly:

pip install dong-ai
dong make "a quarterly market report"
dong analyze path="src/main.py" question="What does this do?"

Capabilities Summary

Links

• Engine: https://github.com/Dong04-123/Dong-AI-Company
• PyPI: https://pypi.org/project/dong-ai/
• Hermes: https://github.com/NousResearch/hermes-agent