Skip to content

Self-evolving second brain with 35 AI skills, 10 agents, and people CRM. Closed-loop harness: a V-model verification lifecycle where the worker never grades its own homework. Plus paired anti-slop design skills for marketing and product UI. Works with Claude Code, Cursor, Kiro, Gemini CLI, Codex.

3.3 KB
Raw Blame

COG Agent Support Matrix

COG intentionally ships multiple agent surfaces. They are not all identical.

This document is the packaging contract: it tells contributors and maintainers which surfaces are first-class today, which ones are partial, and what must stay in sync before publishing a release.

Current Support Matrix

Surface Shipped format Coverage Status
Claude Code .claude/skills/*/SKILL.md 43 skills Full native surface
Antigravity .agents/skills/*/SKILL.md + .agents/agents/*.md 43 skills + 10 agents Full surface (pointer-stub format — delegates to .claude/, which stays authoritative)
Agent Plugins standard Root plugin.json + skills/ (generated mirror) 43 skills Spec 1.0.0 conformant; rebuild with scripts/build-agent-plugin.sh
Universal agent docs AGENTS.md 17 commands Full documented fallback
Kiro .kiro/powers/*/POWER.md 7 powers Core workflows only
Gemini CLI .gemini/commands/*.toml + .gemini/skills/*.md 7 commands Core workflows only

What “Full” vs “Core” Means

Full surfaces

These surfaces should expose the complete public COG command set:

  • onboarding
  • braindump
  • daily-brief
  • weekly-checkin
  • knowledge-consolidation
  • url-dump
  • team-brief
  • meeting-transcript
  • comprehensive-analysis
  • update-cog
  • auto-research
  • create-user-story
  • generate-prd
  • generate-release-notes
  • export-open-issues
  • publish-to-confluence
  • update-knowledge-base

Today, Claude Code, Antigravity, and AGENTS.md are the full surfaces.

Core surfaces

These surfaces intentionally cover the most common personal workflows first:

  • onboarding
  • braindump
  • daily-brief
  • weekly-checkin
  • knowledge-consolidation
  • url-dump
  • update-cog

Today, Kiro and Gemini CLI are core surfaces.

Packaging Rules

If you add, remove, rename, or materially change a public COG skill:

  1. Update the Claude skill in .claude/skills/
  2. Update the matching Antigravity pointer stub in .agents/skills/<name>/SKILL.md (same template as the existing stubs — name + description only change, the body always points back at .claude/skills/<name>/SKILL.md as authoritative)
  3. Update AGENTS.md for the universal surface
  4. Update .claude-plugin/plugin.json so marketplace metadata stays truthful
  5. Update README / SETUP / this file if counts or support levels change
  6. If the skill is supported natively in Kiro or Gemini, update those files too
  7. Add new framework files to FRAMEWORK_FILES in cog-update.sh
  8. Run ./scripts/validate-agent-surface.sh

Validation

Run this before tagging a release, opening a packaging PR, or after using ./cog-update.sh:

./scripts/validate-agent-surface.sh

The validator checks:

  • manifest JSON validity
  • declared skill paths in .claude-plugin/plugin.json
  • plugin skill count vs actual shipped Claude skills
  • AGENTS.md coverage for all manifest skills
  • common packaging drift like agents.md vs AGENTS.md

Safe Update Workflow

Recommended maintainer flow:

./cog-update.sh --check
./cog-update.sh
./scripts/validate-agent-surface.sh

If you have local framework customizations, prefer interactive update mode so you can backup per-file before replacing anything.