COG: The Agentic Second Brain That Actually Self-Evolves
Cognition + Obsidian + Git — A self-evolving second brain powered by AI agents, markdown files, and version control. No database, no vendor lock-in — just .md files that think.
Quick Start | Skills | Features | FAQ | SETUP.md
Works with Claude Code • Antigravity • Cursor • Kiro • Gemini CLI • OpenAI Codex • any AI that reads markdown
Inspired by Garry Tan's gstack and gbrain
graph LR
A[You] -- natural language --> B[AI Agent]
B -- runs --> C[33 Skills]
C -- delegates to --> W[6 Workers]
C -- verified by --> V[4 Read-Only Verifiers]
C -- reads & writes --> D[.md Files]
W -- reads & writes --> D
V -- observes --> D
C -- syncs with --> G[GitHub / Linear / Slack / PostHog]
D --> E[Git]
D --> F[iCloud]
Quick Start
1. Clone & enter the repo:
git clone https://github.com/huytieu/COG-second-brain.git
cd COG-second-brain2. Run onboarding in your agent:
| Agent | Command | How it finds skills |
|---|---|---|
| Claude Code | code . → "Run onboarding" |
.claude/skills/ |
| Antigravity | Open folder → "Run onboarding" | .agents/skills/ + .agents/rules/cog.md |
| Cursor | Open folder → "Run onboarding" | .cursor-plugin/ + .cursorrules |
| Kiro | Open folder → "setup COG" | .kiro/powers/ |
| Gemini CLI | gemini → /onboarding |
GEMINI.md + .gemini/commands/ |
| OpenAI Codex | codex → "Run onboarding" |
AGENTS.md |
| Other agents | Point at AGENTS.md → "Run onboarding" |
AGENTS.md |
Or install via skills.sh:
npx skills add huytieu/COG-second-brain
Done — COG is personalized and ready in ~2 minutes. See SETUP.md for optional config (Git sync, iCloud, Obsidian Tasks, etc.).
Agent Support Matrix
COG ships a full Claude Code surface, a full Antigravity surface, plus core native surfaces for Kiro and Gemini CLI, with AGENTS.md as the universal fallback for Codex and other markdown-reading agents.
| Surface | Current support | Notes |
|---|---|---|
| Claude Code | 33 native skills + 10 agents (6 workers + 4 verifiers) | Full first-class surface |
| Antigravity | 33 skills + 10 agents (pointer-stub format) | Full surface — thin stubs in .agents/ delegate to the .claude/ playbooks, which stay authoritative |
| Agent Plugins standard | Root plugin.json + skills/ (33 skills) |
Spec 1.0.0 conformant; any standard-aware client loads COG as a plugin |
| Cursor | Plugin manifest + rules | .cursor-plugin/plugin.json + .cursorrules |
| Kiro | 7 native powers | Core workflows today |
| Gemini CLI | 7 native commands | Core workflows today |
AGENTS.md |
33 documented commands | Universal fallback for Codex and other agents |
Before publishing or updating framework files, run ./scripts/validate-agent-surface.sh to catch drift between manifests, docs, and shipped files. See docs/AGENT-SUPPORT.md for the detailed support matrix and contributor rules.
Skills
Core Skills (Personal Knowledge)
| Skill | What it does | Try saying... |
|---|---|---|
| onboarding | Personalize COG for your workflow (run first!) | "Run onboarding" |
| braindump | Capture raw thoughts with intelligent classification | "I need to braindump" |
| daily-brief | Verified news intelligence (7-day freshness) | "Give me my daily brief" |
| url-dump | Save URLs with auto-extracted insights | "Save this URL" |
| weekly-checkin | Cross-domain pattern analysis | "Weekly review" |
| knowledge-consolidation | Build frameworks from scattered notes | "Consolidate my knowledge" |
| update-cog | Update framework files without touching your content | "Update COG" |
| memory-hygiene | Trust sweep of persistent memory — re-verify claims against the live environment, stamp last_verified + confidence |
"Audit my memories" |
Team Intelligence Skills (for Product & Engineering Leads)
| Skill | What it does | Try saying... |
|---|---|---|
| team-brief | Cross-reference GitHub + Linear + Slack + PostHog into a daily team intelligence brief with two-way Linear sync-back | "Team brief" / "What did we ship?" |
| meeting-transcript | Process meeting recordings into structured decisions, action items, and team dynamics | "Process this meeting" |
| comprehensive-analysis | Deep 7-day analysis for weekly reviews, board prep, or strategic planning (~8-12 min) | "Weekly analysis" / "Board prep" |
PM Workflow Skills (for Product Managers)
| Skill | What it does | Try saying... |
|---|---|---|
| create-user-story | Create user stories with duplicate checking across Linear, GitHub Issues, or Jira | "Create a user story for..." |
| generate-prd | Draft PRDs with approval gate before publishing to Confluence/Notion | "Generate a PRD for..." |
| generate-release-notes | Generate release notes from GitHub milestones, Linear cycles, or manual input | "Generate release notes for v2.1" |
| export-open-issues | Audit and export open issues from any tracker into a structured vault summary | "Export open issues" |
| publish-to-confluence | Publish any vault markdown file to Confluence | "Publish this to Confluence" |
| update-knowledge-base | Maintain product knowledge base from releases, features, and project changes | "Update the knowledge base with v2.1 changes" |
PM Workflow: These skills form a complete product management lifecycle: Research (
/auto-research) → PRD (/generate-prd) → Stories (/create-user-story) → Development → Release Notes (/generate-release-notes) → Knowledge Base (/update-knowledge-base). Use/export-open-issuesfor audits and/publish-to-confluenceto share externally.
Strategic Research
| Skill | What it does | Try saying... |
|---|---|---|
| auto-research | Deep strategic research engine — decomposes questions into parallel research threads with multiple agents | "Research the future of AI testing tools" |
Content Creation
| Skill | What it does | Try saying... |
|---|---|---|
| content-factory | Autonomous content pipeline — scout announcements, triage by trend momentum + your unique angle, publish with ledger dedup, hard volume caps, and screenshot-verified posting | "Run the content factory" |
Verification Harness
Opt-in. Ask for one of these and work walks a V: decompose left into falsifiable criteria, build at the apex, verify right with evidence traced back to each criterion. Say nothing and none of it runs: ordinary work carries no checkpoints and no evidence ledger. Turn it on per request, or by default for build tasks with verification_harness: on in 00-inbox/MY-PROFILE.md. Full lifecycle in WORKFLOW.md.
| Skill | What it does | Try saying... |
|---|---|---|
| closed-loop | Plan → build → component verify → integration verify → acceptance. The worker never grades its own homework; verifiers observe the artifact, not the worker's summary | "Run this through the closed loop" |
| ultragoal | A goal too big for one session, run as a chain of phases with cross-session state and a north-star acceptance gate | "Track this as an ultragoal" |
| harvest | Capture session learnings, stage them for approval, propose skill patches. Never writes durable knowledge unapproved | "Harvest what we learned" |
| retro | Audit the run's checkpoints and evidence quality. A run can pass every gate on weak observations | "Retro this run" |
| review-cockpit | One living doc per multi-item session: progress, per-item cards, and an approval slot you edit in place | "Set up a review cockpit" |
Craft Skills (Writing & Visuals)
| Skill | What it does | Try saying... |
|---|---|---|
| no-ai-slop | Sharpen a draft without flattening its voice, or just detect AI-slop patterns | "Make this less AI-sounding" |
| slop-gate | Refuse slop tells in anything about to be written, published, or sent; runs in CI or as a hook | "Scan this before I publish" |
| voice-baseline | Measure the words and shapes you over-use so an agent stops amplifying your tics | "Why do my drafts all sound the same?" |
| release-video | Turn a release into a motion recap and explained per-feature demos, with sound tied to motion | "Make a video for this release" |
| editorial-illustrations | Claim → geometry: derive the right figure from what the text argues, render it as self-contained theme-aware HTML/SVG | "Make a diagram for this argument" |
| data-forms | 20+ chart and diagram forms with when-to-use, failure modes, and the encoding discipline that makes any of them readable | "The bar chart is burying the point" |
| museum-art | Real public-domain artwork from museum open-access APIs instead of AI-generated or stock imagery | "Find a hero image for this post" |
| daily-journal | A passive work journal the agent keeps for you, plus a guided reflection on request | "Reflect on today" |
Design Skills (Anti-Slop UI)
A paired set. taste-skill owns first-impression surfaces, product-ui-taste owns surfaces people use every day. Never run both on the same component.
| Skill | Owns | Try saying... |
|---|---|---|
| taste-skill | Landing pages, portfolios, marketing, editorial. States a "Design Read" before writing code so it stops defaulting to one house aesthetic | "Build a landing page for this" |
| product-ui-taste | Dashboards, data tables, forms, wizards, settings, admin consoles. Budgets the frame in px first, then survives real data: overflow, long labels, empty/error/permission-denied states | "Build the admin table for this" |
Agents (Specialist Sessions)
COG uses a worker agent architecture inspired by garrytan/gstack specialist sessions and garrytan/gbrain knowledge patterns. Workers handle data-heavy tasks cheaply (Sonnet) while the lead session does reasoning (Opus).
| Agent | What it does | Model |
|---|---|---|
| worker-data-collector | Structured extraction from GitHub, Slack, Jira, Linear | Sonnet |
| worker-researcher | Web research with source citations | Sonnet |
| worker-file-ops | Vault file operations, metadata, profiles | Sonnet |
| worker-executor | Pre-approved mutations (Jira, Linear, APIs) | Sonnet |
| worker-publisher | Publishing to Slack, Confluence, Notion | Sonnet |
| brief-people-updater | Batch-update people profiles from meetings/briefs | Sonnet |
Read-only verifiers (they cannot edit files or mutate external state):
| Agent | What it does | Checkpoint |
|---|---|---|
| task-verifier | Checks output against acceptance criteria by observing the artifact | CP-3v |
| integration-verifier | Cross-task wiring and global acceptance | CP-4 |
| fix-agent | Targeted fixes after a FAIL:fixable verdict; max 2 attempts |
CP-3v retry |
| harvest-curator | Shapes session learnings into adoption notes; propose-only | CP-7 |
Workers write results to
/tmp/files and return only a status + path. The lead reads the file for synthesis. This eliminates slow token generation in agent output.Verifiers receive paths only, never a worker's output. Pasted context induces narrativisation: the verifier classifies the framing instead of independently reading the source.
People CRM (Knowledge-Based Team Profiles)
Track the people you work with using progressive, evidence-based profiles in 05-knowledge/people/. Profiles auto-escalate via tiered enrichment:
- Tier 3 (Stub) — 1 mention: name, role, one-line context
- Tier 2 (Moderate) — 3+ mentions: executive snapshot, working style, strengths
- Tier 1 (Full) — 8+ mentions or direct meeting: complete profile with all sections
Every observation includes a source citation with confidence level. See 05-knowledge/people/README.md for details.
Role Packs (Personalized Recommendations)
COG matches your role during onboarding to a role pack that prioritizes the most relevant skills and integrations for you. Available role packs: Product Manager, Engineering Lead, Engineer, Designer, Founder, Marketer — or create your own from the template.
New to team skills? These require GitHub CLI (
gh) and work best with Linear, Slack, and PostHog MCP integrations. They degrade gracefully — start with just GitHub and add integrations over time. See SETUP.md for configuration.
The Evolution Cycle
graph TD
A[Daily: Braindump thoughts] --> B[Daily: News intelligence]
A --> T[Daily: Team intelligence brief]
M[Meetings: Process transcripts] --> T
T -- syncs back to --> L[Linear / GitHub]
B --> C[Weekly: Pattern analysis]
T --> CA[Weekly: Comprehensive analysis]
C --> D[Monthly: Knowledge consolidation]
CA --> D
D -- COG learns your patterns --> A
R[Strategic: Auto-research] --> P[PM: Generate PRD]
P --> S[PM: Create user stories]
S -- after development --> RN[PM: Release notes]
RN --> KB[PM: Update knowledge base]
- Daily capture — braindump raw thoughts; COG classifies by domain and extracts action items
- Daily intelligence — personalized news briefings with verified, sourced news
- Daily team brief — cross-reference GitHub, Linear, Slack, PostHog, meetings into one brief with two-way sync
- Meeting processing — extract decisions, action items, and team dynamics from transcripts
- Weekly reflection — pattern analysis across all domains surfaces insights you'd miss
- Weekly deep dive — comprehensive analysis for board prep, retros, and strategic planning
- Monthly synthesis — scattered notes become consolidated frameworks and a knowledge base
- Strategic research — deep multi-agent investigation of strategic questions with real sources
- PM workflow — full product lifecycle from PRD to release notes to knowledge base updates
Features at a Glance
| Self-Evolving — Learns your patterns, auto-organizes content, builds frameworks | Self-Healing — Rename files or restructure; cross-references update automatically | Verification-First — Sources required, 7-day freshness, confidence levels on all analysis |
Privacy-First — Local .md files, strict domain separation, no external servers |
Multi-Device — iCloud sync to iPhone/iPad/Mac; Git for version history | Obsidian Tasks — 📅 YYYY-MM-DD emoji format works with Tasks plugin dashboards |
| Garry Tan Inspired — gstack specialist sessions + gbrain knowledge patterns | Multi-Platform — Listed on skills.sh, agentskill.sh, cursor.directory | Worker Agents — Sonnet handles I/O, Opus handles thinking |
| Runtime Trust — memory hygiene sweeps re-verify stored facts against the live environment | Post-Condition Checks — mutating skills observe the artifact before reporting success | Skill Distillation — explore once, execute cheap with small models |
| V-Model Closed Loop — every criterion has a task on the left and an evidence row on the right | Risk Lanes — ceremony scales with blast radius | Cross-Model Gate — a different model family catches a different error class |
Your Vault
COG-second-brain/
├── .claude/skills/ # Claude Code skills (33)
├── .claude/agents/ # Worker agent definitions (6)
├── .claude/roles/ # Role packs (7) — personalized recommendations
├── .agents/ # Antigravity (agy CLI + IDE) — pointer stubs to .claude/
├── .kiro/powers/ # Kiro powers
├── .gemini/commands/ # Gemini CLI commands
├── AGENTS.md # Universal agent docs
├── CLAUDE.md # Framework instructions
├── 00-inbox/ # Profiles, interests, integrations
├── 01-daily/ # Briefs & check-ins
├── 02-personal/ # Personal braindumps (private)
├── 03-professional/ # Professional braindumps & strategy
├── 04-projects/ # Per-project tracking
├── 05-knowledge/ # Consolidated insights & patterns
│ └── people/ # People CRM profiles
└── 06-templates/ # Document templates
Real-world results: 120+ braindumps processed, daily briefs with 95%+ source accuracy, 5 major strategic insights discovered — zero maintenance required.
Keeping COG Updated
COG separates framework files (skills, docs, scripts) from your content (braindumps, profiles, notes). Updates never touch your personal data.
| Method | Command |
|---|---|
| AI Agent (any) | "Update COG" or /update-cog |
| Shell script | ./cog-update.sh (interactive) • --check • --dry-run • --force |
| Manual Git | git fetch cog-upstream main then checkout specific files |
Check your version: cat COG-VERSION
Validate packaged surfaces: ./scripts/validate-agent-surface.sh
FAQ
Why not just use Notion / Roam / Obsidian alone?
COG adds self-evolving intelligence on top. It doesn't just store — it learns, analyzes, and synthesizes insights automatically.
How much does it cost?
COG is free and open-source (MIT). You only pay for your AI agent's API usage.
Is my data private?
Yes. Everything is local markdown files. The AI agent's API is only called when you invoke a skill. No data stored on external servers.
Can I customize or add skills?
Yes — edit any SKILL.md / POWER.md / AGENTS.md file. See SETUP.md for details on creating new skills.
Will updating overwrite my customizations?
No. The update process detects customized files and lets you choose per-file: keep yours, use upstream, or backup + update. Nothing is overwritten without approval.
What if I don't use Git?
Git is optional but recommended for version history. COG works fine with just iCloud sync.
Roadmap
-
Gemini CLI + OpenAI Codex support(shipped in v3.1) -
Upstream update system(shipped in v3.2) -
Role packs & integration discovery(shipped in v3.3) -
PM workflow skills & auto-research(shipped in v3.4) -
Worker agents, people CRM & specialist sessions(shipped in v3.5) -
Loop engineering for iterative skills(shipped in v3.6) -
Runtime trust: memory hygiene, post-condition checks, verifier pass, content factory(shipped in v3.7) - Web interface for knowledge graph visualization
- Mobile-first commands (optimized for Obsidian mobile)
- Team collaboration features (with privacy preservation)
- Integration with calendar/task management tools
Contributing & Support
| Contribute | Report bugs | Discussions |
| Sponsor on GitHub | Buy me a coffee | MIT License |
Acknowledgments & Inspiration
Built with Claude Code, Cursor, Kiro, Gemini CLI, OpenAI Codex, and Obsidian.
Key inspirations:
- Garry Tan's gstack — specialist sessions, clear operating gears, repo-local skill distribution. COG's worker agent architecture and model routing borrow directly from gstack's explicit mode separation.
- Garry Tan's gbrain — Compiled Truth + Timeline pattern, tiered enrichment for people profiles, brain-first lookup protocol. COG's people CRM and knowledge-first approach are adapted from gbrain's design.
- Zettelkasten — atomic, interlinked notes as the foundation of knowledge
- Building a Second Brain (Tiago Forte) — PARA organization, progressive summarization
- GTD (David Allen) — capture everything, process systematically
TL;DR: Clone, run onboarding, braindump daily. COG evolves with you — just .md files, any AI agent, zero maintenance.