Knowledge Pack

Manage, preview, and download the knowledge files that power the FCOS agent. Both Claude Code and the GPT can add new files and merge existing ones to build the best knowledge base. ⬆ UPDATED files need re-uploading to GPT builder. Once uploaded, click Mark All Uploaded to reset the badges.

-- files
-- words
-- updated
-- downloaded
Loading knowledge files...

GPT Configuration

Live instructions, description, and starters — always generated from the current system state. ⬆ UPDATED means content changed since you last copied. ✔ COPIED means you've copied it this session.

GPT Instructions
Loaded from 09-master-prompt-pack.md — paste into GPT Builder → Instructions field.
Loading...
GPT Description
Generated live from system state — paste into GPT Builder → Description field (300 char limit).
Loading...
Conversation Starters
Generated from current FCOS capabilities — paste into GPT Builder → Conversation starters.
    Links & Actions
    GPT Builder Open GPT Builder
    Live GPT Open FCOS Agent
    After uploading
    Copied state
    Operation Breakdown
    Loading menu data...
    Sync Validation
    Extracted from File 09 and verified against the live manifest and menu. All three must be green before uploading to GPT Builder.
    Load GPT config to validate...

    System Status

    Live operational health, truth checks, and sync tools.

    --
    Total Operations
    --
    Total Menus
    --
    Knowledge Files
    OK
    Health

    Claude Code Agent Builder

    Build knowledge packs, custom agents, skills, hooks, MCP servers, and memory systems for Claude Code.

    CLAUDE.md — Project Instructions
    The primary way to customise Claude Code. Loaded at session start and followed throughout.
    Where CLAUDE.md Files Go
    Claude Code loads instructions from multiple locations. Higher priority overrides lower.
    Project./CLAUDE.md or ./.claude/CLAUDE.md — shared with team via git
    User~/.claude/CLAUDE.md — personal preferences, all projects
    Rules.claude/rules/*.md — modular rules with path-scoped frontmatter
    Nestedsubdir/.claude/CLAUDE.md — loaded on demand when Claude reads files in that dir
    Import@path/to/file syntax inside CLAUDE.md to import other files

    Claude Projects

    Build knowledge bases and project instructions for Anthropic's Claude Projects on claude.ai.

    Project Instructions Builder
    Each Claude Project has a system prompt that guides all conversations within it. Up to 200K tokens of context.
    Knowledge Base Specs
    FilesUnlimited per project. Each up to 30 MB. PDF, DOCX, CSV, TXT, HTML, EPUB, RTF, ODT.
    Context200K token window. Auto-switches to RAG for larger knowledge bases (up to ~10x expansion).
    SyncGoogle Workspace integration — sync Docs, Sheets, and Gmail directly into projects.
    SharingProjects can be shared with team members. Each member sees the same instructions and knowledge.
    ScopeInstructions apply to ALL conversations within the project. Global custom instructions apply to all projects.
    Best Practices
    TipFront-load your most important rules. Claude reads instructions top-to-bottom — put non-negotiable rules first.
    TipUse structured headings (## Rules, ## Context, ## Style) so Claude can quickly reference sections.
    TipUpload reference docs (brand guides, API specs, code standards) as knowledge files rather than pasting into instructions.
    TipKeep instructions under 4,000 words. Longer instructions dilute focus. Put detail in knowledge files instead.

    OpenAI Assistants / Responses API

    Build system prompts, tool configurations, and function calling schemas for production AI agents.

    System Prompt Builder
    The system prompt defines your agent's identity, capabilities, and rules. Used in both Assistants API and Responses API.
    Function Calling Schema Builder
    Define tools your agent can call. JSON Schema format, used for structured outputs and API integrations.
    Platform Reference
    Responses APIThe current standard (replaces Assistants API, deprecated Aug 2026). Supports web search, file search, code interpreter, computer use, image generation, remote MCP.
    File SearchUp to 10,000 files per vector store. 100 GB per project. Automatic chunking and embedding.
    ToolsFunction calling, web search, file search, code interpreter, computer use, image generation, MCP connectors.
    ModelsGPT-4o, GPT-4o-mini, GPT-4.1, o3, o4-mini. Temperature, top_p, max_tokens configurable.
    ConnectorsGoogle Drive, Dropbox, SharePoint via built-in connectors. Or custom via MCP.

    Gemini Gems

    Build custom AI experts for Google Gemini. Prompt-based — no file uploads, pure instruction design.

    Gem Instructions Builder
    Gems are saved instruction sets that turn Gemini into a specialised expert. Instructions only — no knowledge file uploads.
    Gem Specs
    FormatPrompt-only. No file uploads, no tools, no API connections. Pure instruction design.
    LimitNo official word limit documented, but keep under 2,000 words for best adherence.
    AccessGemini Advanced (paid) for creation. Free tier can use pre-made Gems.
    SharingShareable via link. Can be published to the Gem gallery.
    CodeGEMINI.md config files for Gemini-aware code tools (similar to CLAUDE.md).
    Design Tips
    KeySince you can't upload files, encode all domain knowledge directly in the instructions. Be specific and structured.
    KeyUse persona framing: "You are a [role] with expertise in [domain]. You always [behaviour]."
    KeyInclude example outputs in your instructions. Show Gemini exactly what good output looks like.

    Microsoft Copilot Studio

    Build AI agents for the Microsoft 365 ecosystem. Constrained knowledge sources make optimisation critical.

    Agent Instructions Builder
    Natural language description that defines your Copilot agent's behaviour, personality, and scope.
    Knowledge Source Limits
    Copilot Studio has strict limits on knowledge sources — structuring them well is essential.
    4 URLsMaximum 4 public website URLs as knowledge sources.
    SharePointFiles, folders, and entire SharePoint sites. The primary enterprise knowledge source.
    5 TeamsMaximum 5 Teams chat URLs. Pulls conversation context.
    FilesUploaded files (Word, PDF, PowerPoint). Size limits per tenant.
    M365Microsoft Graph connectors — email, calendar, OneDrive, Dynamics 365.
    Integration Points
    TeamsDeploy agents directly into Microsoft Teams as chatbots or meeting assistants.
    Power PlatformConnect to Power Automate flows, Power Apps, Dataverse.
    AzureAzure AI services, Cognitive Search, custom APIs via Power Platform connectors.
    ChannelsDeploy to web, Teams, Slack, Facebook, custom channels.

    CrewAI

    Build multi-agent teams with role-based YAML definitions. The fastest-growing open-source agent framework.

    Agent Definition Builder
    CrewAI agents are defined in YAML with role, goal, and backstory. Each agent is a specialist in your team.
    Agent YAML Reference
    roleRequired. The agent's job title. e.g. "Senior Market Research Analyst"
    goalRequired. What the agent is trying to achieve. Be specific and measurable.
    backstoryRequired. Context and expertise. Shapes the agent's personality and approach.
    toolsArray of tool names: [search_tool, scrape_tool, calculator]
    llmModel to use: gpt-4o, claude-sonnet-4-6, etc.
    max_iterMaximum reasoning iterations before stopping. Default: 25.
    allow_delegationtrue/false — can this agent delegate work to other agents?
    verbosetrue/false — show detailed reasoning output.
    Crew Orchestration Patterns
    SequentialAgents execute tasks in order. Output of one feeds into the next. Best for pipelines.
    HierarchicalA manager agent delegates and coordinates. Best for complex projects with dependencies.
    ParallelAgents work simultaneously on independent tasks. Best for research and data gathering.
    Install: pip install crewai | Docs: docs.crewai.com | 5.2M+ downloads

    Preview

    FCOS Assistant