Home / Skills / COG-second-brain

huytieu/COG-second-brain

huytieu / COG-second-brain

Agent SkillsPythonMIT

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 …

1.2kstars
150forks
9open issues
2026-10-02last pushed
MITlicense

What it is, and how people use it

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 … It sits on the Agent Skills shelf of the catalog, maintained by @huytieu.

✓Drop-in for Claude Code, Codex CLI and Cursor - install and it just runs.
✓Read the SKILL.md / README first; most setups are one command from the docs.
✓Teams use it to standardise how agents behave across every repo in the org.
agent-loopagent-pluginsagent-skillsagenticai-agents

Why builders keep it installedAgent Skills benefits

Your agent stops guessing the workflow

A skill packages the exact steps, file conventions and checklists an expert would follow. The model spends its context on your feature instead of figuring out how the tool works.

One install, every repo behaves the same

Commit the skill to your project and every teammate's Claude Code, Codex or Cursor run inherits the same quality bar - including the new teammate on day one.

Cheaper runs, fewer retries

Tuned instructions cut failed attempts and half-baked output. Fewer retries means fewer tokens burned and noticeably shorter sessions on paid model plans.

Survives the next model release

Skills are plain markdown and scripts the community versions together. A model upgrade changes behaviour, not your workflow - unlike a bespoke prompt that breaks silently.

Contributorsloading from GitHub…

Languagesestimated from snapshot

  • Python 85.2%
  • Shell 11.1%
  • JavaScript 3.7%

Community pulsesnapshot ratios, normalized

starsforksissues ×3watchers