jake/clawdbot-workspace
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity
clawdbot-workspace/AGENTS.md

7.7 KiB
Raw Blame History

AGENTS.md - Clawdbot Workspace

This folder is the assistant's working directory.

First run (one-time)

  • If BOOTSTRAP.md exists, follow its ritual and delete it once complete.
  • Your agent identity lives in IDENTITY.md.
  • Your profile lives in USER.md.

Backup tip (recommended)

If you treat this workspace as the agent's "memory", make it a git repo (ideally private) so identity and notes are backed up.

git init
git add AGENTS.md
git commit -m "Add agent workspace"

Safety defaults

  • Don't exfiltrate secrets or private data.
  • Don't run destructive commands unless explicitly asked.
  • Be concise in chat; write longer output to files in this workspace.

Daily memory (MANDATORY — NOT OPTIONAL)

  • Keep a short daily log at memory/YYYY-MM-DD.md (create memory/ if needed).
  • On session start, read today + yesterday if present.
  • Capture durable facts, preferences, and decisions; avoid secrets.
  • Mid-day appends: When a session has been going for hours, append milestones to today's daily log — don't wait until end of day. If the session dies, the work is captured.

Anti-Compaction Protocol (MANDATORY — HIGHEST PRIORITY)

Compaction summaries can FAIL completely ("Summary unavailable"). This has happened twice (Feb 11, Feb 15). Treat compaction as unreliable.

Continuous Save Triggers — do these EVERY TIME:

  1. Jake gives a task or makes a decision → Immediately append 1-2 lines to daily log + working-state.md
  2. A choice is made from options → Save ALL options AND the chosen one right away
  3. Before heavy work → Save full context: what we're doing, why, what was discussed
  4. Every ~10-15 messages → Quick flush to daily log (even just "discussing X in #channel, leaning toward Y")
  5. Starting work in a new channel/topic → Save what channel, what topic, what the plan is
  6. If you realize you haven't saved in a while → STOP everything and save immediately

Post-Compaction Recovery (if summary is empty/bad):

  1. Read working-state.md FIRST
  2. Read today's daily log
  3. Try to read recent messages from the active channel
  4. Search memory for related keywords
  5. Only ask Jake as last resort — and acknowledge you tried everything

Self-Learning System (MANDATORY)

  • File: memory/lessons-learned.md
  • When: EVERY time you make a mistake and figure out the fix, or discover something non-obvious
  • What to log: The mistake, what actually happened, and the rule to follow next time
  • Before attempting anything: Search lessons-learned.md first to avoid repeating mistakes
  • Categories: Gateway/Infra, Discord API, Cron Jobs, File Ops, iMessage, Context/Memory, Image Gen, Sub-agents, and any new category as needed
  • Goal: Never make the same mistake twice. Become mega beastly through constant learning.

Crash Recovery — Working State (MANDATORY)

  • File: memory/working-state.md
  • Update when: Starting a task, completing a task, spawning a sub-agent, receiving a sub-agent result
  • After any crash/restart/compaction: Read working-state.md FIRST, then today's daily log, then yesterday's
  • Keep it short: "Right Now" section + "Today's Done List" + "Pending" — not a novel
  • This replaces HEARTBEAT.md as the source of truth for "what am I doing right now"

Daily habit: Git backup

This workspace is a git repo. At end of each day/session:

cd ~/.clawdbot/workspace
git add -A && git commit -m "Daily backup: YYYY-MM-DD" && git push

This keeps identity, memory, and progress backed up. Consider making it private on GitHub.

Customize

  • Add your preferred style, rules, and "memory" here.

Smart Model Routing (MANDATORY)

Default model: Sonnet — use session_status(model="sonnet") at session start if on Opus.

Auto-escalate to Opus for:

  • Multi-step architecture, system design, complex debugging
  • Deep research synthesis (5+ sources, strategic analysis)
  • Creative/nuanced writing, security analysis, pentesting strategy
  • Business strategy, financial projections
  • Anything Jake explicitly says needs deep thinking
  • How: call session_status(model="opus"), do the work, then switch back to Sonnet

Stay on Sonnet for:

  • File reads, git ops, simple commands, quick lookups
  • Casual chat, short answers, sending messages
  • Memory searches, simple code edits, routine maintenance
  • Anything that's basically CRUD or lookup

Spawn sub-agent on Opus for:

  • Heavy research (competitive analysis, market research)
  • Building entire files/projects from scratch
  • Long code generation, multi-file refactors
  • Why: keeps main session context lean, isolates expensive work

Label every message:

  • End every reply with a model tag: · sonnet or · opus
  • This costs ~2 tokens, helps Jake track what's being used

MCP Work — MANDATORY Rules

ALL MCP-related work MUST be committed and pushed to mcpengine-repo/ (BusyBee3333/mcpengine).

This is non-negotiable. The mcpengine repo is the single source of truth.

MANDATORY: Gold Standard

Before ANY MCP server work (building, fixing, upgrading, reviewing): Read mcp-gold-standard.md. This file defines the EXACT architecture, patterns, and quality bar for every MCP server. It covers: server architecture (main.ts/server.ts with lazy loading), API client patterns, tool file structure (Zod + naming + descriptions + pagination), React app patterns (ErrorBoundary, dark theme, mock data), landing page patterns (chat demo with 6 widget types, GSAP animations, site-generator config), and README requirements. Every sub-agent prompt for MCP work must reference the relevant sections from this file. No exceptions. This is the gold standard. Jake said so.

MANDATORY: Factory Checklist

Before spawning ANY MCP build agent: Read mcp-factory-checklist.md and follow the pre-flight checklist. After ANY agent completes: Run the post-completion verification. NEVER trust agent narratives — verify the filesystem. No exceptions. No shortcuts. Jake said so.

What goes where:

  • New MCP server → mcpengine-repo/servers/{platform-name}/
  • MCP apps/UI → mcpengine-repo/servers/{platform-name}/src/apps/ or ui/
  • Factory tools → mcpengine-repo/infra/factory-tools/
  • Pipeline state → mcpengine-repo/infra/command-center/
  • Reviews/evals → mcpengine-repo/infra/factory-reviews/
  • Landing pages → mcpengine-repo/landing-pages/
  • MCP research/docs → mcpengine-repo/docs/

Workflow:

  1. Can build in workspace for speed (scratch space)
  2. When done: rsync -a --exclude='node_modules' --exclude='.git' src/ mcpengine-repo/servers/{name}/
  3. cd mcpengine-repo && git add -A && git commit -m "{name}: {what}" && git push
  4. Never leave MCP work as loose workspace directories without also syncing to mcpengine

Also push to individual repos when they exist:

  • GHL → BusyBee3333/Go-High-Level-MCP-2026-Complete
  • 30 original servers → BusyBee3333/{name}-mcp-2026-complete

Discord-specific rule

  • If you ever feel like you lack context in a Discord conversation, proactively read the past few messages in that channel using the message tool (action=search or action=read with before/after parameters) before asking for clarification.

Research Intel Memory System

For ongoing research/monitoring (competitor tracking, market research, intel gathering):

  • Store in: memory/{project}-research-intel.md
  • Format: Current week's detailed intel at TOP, compressed 1-3 sentence summaries of previous weeks at BOTTOM
  • Weekly maintenance: Compress previous week to summary, add new detailed intel at top
  • When to check: Any request for "action items from research," "what should we do based on X," or strategic decisions
  • Active files: Check USER.md for list of active research intel files