Kcc Office Ai Agents

AI Agents collaboration system for KCC Office v2 - enables autonomous operation and collaboration of office agents (Komi, CEO, CFO, CTO, COO, EDN) with persi...

Security Scan

The skill's files and scripts mostly match its stated purpose (an on-disk multi‑agent workspace), but there are notable inconsistencies and sloppy assumptions (undeclared API keys/endpoints in docs, scripts with minor bugs and path assumptions, guidance that encourages committing/pushing data) that warrant caution before use.

Purpose & Capability

The name/description (office AI agents with persistent memory and autonomous behavior) aligns with the shipped files and scripts: markdown agent definitions, WAL/working-buffer patterns, and small shell scripts to create/manage per-agent workspaces. However, the tools documentation lists external services and environment variables (GEMINI_API_KEY, PIXELLAB_API_KEY, and a Gemini endpoint) even though the skill's registry metadata declares no required env vars or credentials. It's plausible the skill legitimately needs those keys for asset generation, but the omission from requires.env is an inconsistency.

Instruction Scope

Runtime instructions and scripts operate on files inside the provided workspace (creating ONBOARDING.md, SHUTDOWN.md, writing memory files, heartbeats). They instruct agents to read and write local memory files (SESSION-STATE.md, MEMORY.md, working-buffer.md) which is expected. A few policy-style lines encourage autonomous actions (e.g., 'Don't ask permission. Just do it.' and 'Commit and push your own changes'), which could lead to data leaving the environment if a git remote or other outbound connection is configured. The SKILL.md does warn 'Ask first' for things that leave the machine, but the mix of directives is inconsistent and grants broad discretion to an agent without clear guardrails.

Install Mechanism

There is no install spec; this is instruction-and-scripts only. That is low-risk from an installer perspective — nothing is fetched or executed automatically from remote hosts by an install step.

Credentials

The repo/docs reference external APIs and environment variables (GEMINI_API_KEY, PIXELLAB_API_KEY, NODE_ENV, PORT, LOG_LEVEL) while the registry metadata lists no required env vars. Requiring external API keys would be proportional for image/asset generation, but the skill fails to declare them formally. This mismatch makes it unclear what credentials the skill expects and whether secrets might later be used or requested by agent behavior or auxiliary tooling.

Persistence & Privilege

The skill does not request permanent/always-on inclusion and does not declare modifications to other skills or system-wide agent config. Its persistence is limited to creating and modifying files inside its own workspace. Autonomous invocation is allowed (platform default), which combined with file-write capability can have impact, but alone is not unusual here.

What to consider before installing

This package is largely what it claims: a filesystem-based multi-agent workspace with onboarding/setup/status scripts. Before installing or running anything, do the following: 1) Inspect the files locally and run scripts in a sandboxed container or throwaway VM — the scripts operate on relative paths and may behave unexpectedly depending on your current working directory. 2) Check for undeclared external dependencies: TOOLS.md references GEMINI_API_KEY and PIXELLAB_API_KEY and a Gemini endpoint; decide whether you will provide those keys and confirm how/when the agents will call external APIs. 3) Review and harden any git configuration or remotes in the workspace — the docs encourage committing/pushing changes, which could exfiltrate sensitive workspace content if a remote exists. 4) Note small code-quality issues (e.g., a probable missing fi in scripts/heartbeat-agent.sh and some fragile relative-path copies in onboard-agent.sh) — test scripts manually before relying on them. 5) If you intend to allow autonomous behavior, enforce policy controls (e.g., block outbound network access or require manual approval for actions that send data off-host). If you need further help, provide the exact env where you plan to run this (OS, shell cwd, presence of git remotes, network policy) and I can point out specific risky lines and a safe run procedure.

KCC Office v2 AI Agents 🏢

AI Agents collaboration system for the KCC Office v2 pixel art office environment.

Enables autonomous operation of office agents with persistent memory, proactive behaviors, and continuous improvement.

What's New in v1.0.0

• Initial implementation of AI agents for KCC Office v2
• Based on Proactive Agent framework (WAL Protocol, Working Buffer)
• Autonomous agent execution capabilities
• Inter-agent communication and coordination

Core Features

🤖 Autonomous Agents

Each office role operates as an independent AI agent:

• Komi: Task coordination and overall management
• CEO: Strategic direction and decision-making
• CFO: Financial analysis and investment strategies
• CTO: Technical implementation and architecture
• COO: Operational execution and process optimization
• EDN: User support and independent assistance

🧠 Persistent Memory

• WAL Protocol: Critical details written before responding
• Working Buffer: Captures exchanges in danger zone (>60% context)
• Compaction Recovery: Recovers context from working buffer after session restart
• Semantic Search: Searches past interactions for relevant context

⚡ Proactive Behaviors

• Anticipates office needs before being asked
• Reverse prompting: Suggests improvements humans haven't considered
• Proactive check-ins: Monitors office health and reaches out when needed
• Growth loops: Tracks patterns and identifies automation opportunities

🔄 Self-Improvement

• Learns from every interaction and updates operating procedures
• Safe evolution with Anti-Drift Limits and Value-First Modification
• Relentless resourcefulness: tries 10 approaches before asking for help
• Verification before reporting: tests outcomes, not just outputs

Agent Roles & Responsibilities

Komi (Task Coordinator)

• Central back position (big desk, pink hoodie + devil horns)
• Receives tasks from human (Gasol) and delegates to appropriate agents
• Tracks overall office status and progress
• Coordinates cross-agent initiatives
• Maintains office-wide context and priorities

CEO (Chief Executive Officer)

• Central front position
• Sets overall vision and strategy for the office
• Makes high-level decisions affecting all agents
• Represents the office in external interactions
• Ensures alignment with company goals

CFO (Chief Financial Officer)

• Upper right position
• Manages office budget and financial resources
• Analyzes investment opportunities and risks
• Tracks financial performance and forecasts
• Ensures fiscal responsibility and sustainability

CTO (Chief Technology Officer)

• Lower right position (this agent)
• Oversees technical implementation and architecture
• Evaluates and adopts new technologies
• Ensures system reliability, security, and performance
• Coordinates technical work across agents

COO (Chief Operations Officer)

• Upper left position
• Optimizes office operations and workflows
• Implements processes and procedures
• Tracks operational metrics and efficiency
• Ensures smooth day-to-day operations

EDN (Independent User Assistant)

• Central or flexible position
• Provides direct assistance to office users
• Answers questions and resolves user issues
• Gathers user feedback for improvement
• Acts as liaison between users and other agents

Technical Implementation

Memory Architecture

kcc-office-ai-agents/
├── agents/
│   ├── komi/
│   │   ├── SOUL.md
│   │   ├── USER.md
│   │   ├── AGENTS.md
│   │   ├── MEMORY.md
│   │   └── memory/
│   ├── ceo/
│   ├── cfo/
│   ├── cto/
│   ├── coo/
│   └── edn/
├── SESSION-STATE.md     # Active working memory (WAL target)
├── HEARTBEAT.md         # Periodic self-improvement checklist
└── memory/
    ├── YYYY-MM-DD.md    # Daily raw capture
    └── working-buffer.md  # Danger zone log

Communication Protocol

• Agents communicate through shared memory files
• Direct messaging for immediate coordination
• Public announcements for office-wide notifications
• Structured updates for status reporting

Installation

Since this is a custom implementation for KCC Office v2, copy the files to your workspace:

mkdir -p ~/your-workspace/kcc-office-ai-agents
cp -r ./kcc-office-ai-agents/* ~/your-workspace/kcc-office-ai-agents/

Then configure each agent's workspace with their specific SOUL.md, USER.md, and AGENTS.md files.

Usage

1. Start with ./scripts/onboard-agent.sh for each agent role
2. Agents will detect ONBOARDING.md and offer to get to know their responsibilities
3. Answer questions to populate USER.md and SOUL.md
4. Run initial setup: ./scripts/setup-agent.sh
5. Agents become operational and begin autonomous operation

Best Practices

• WAL before responding: Write critical details to SESSION-STATE.md first
• Search before acting: Use semantic search for past context
• Verify before reporting: Test outcomes, not just outputs
• Promote learnings: Move valuable insights to AGENTS.md or SOUL.md
• Review regularly: Use heartbeat system for self-improvement

License & Credits

License: MIT — use freely, modify, distribute.

Based on: Proactive Agent v3.1.0 and Self-Improvement Skill frameworks.

Created by: Komi (CTO agent) for KCC Office v2 project.

---