Fair Source Framework

cognitive-core

AI does the work. Humans stay in control.

The portable framework that turns any AI coding assistant into an intelligent development platform. Hooks, agents, skills, and security — installed in one command.

View on GitHub Quick Start

~/my-project

$ bash install.sh

◻◻◻• cognitive-core v1.0.0

AI-native development framework

✓ 10 hooks installed (security guard active)

✓ 10 agents registered

✓ 47 skill packs active

✓ 12 rules loaded

✓ Integrity checksums verified

$ Ready. Start a session to begin.

Why cognitive-core

Framework vs. SaaS -- what makes this different

Not an IDE, not a SaaS

A portable framework that installs into any existing project. No vendor lock-in, no subscriptions, no cloud dependency.

Multi-adapter

Claude Code, Aider + Ollama (fully local), IntelliJ. One framework, multiple runtimes -- choose what fits your workflow.

Multilingual

Works in any language the model supports. 11 first-class language packs included, from Perl to Rust to Angular.

Self-evolving

Grows with every prompt, every issue, every session. Skills capture team knowledge. Updates propagate across all connected projects.

Fair Source

Source code publicly accessible, auditable, and transparent. Not open source, not proprietary -- the best of both worlds.

Your rules

Create custom skills, agents, and hooks for any specialized technology. The framework adapts to your team, not the other way around.

Everything you need for intelligent development

One framework. Every Claude Code project. Automatic security, quality, and consistency.

Security Guard

Multi-tool security enforcement with PreToolUse hooks. Blocks exfiltration, pipe-to-shell attacks, secret leaks, and unauthorized file access. Three security levels: minimal, standard, strict.

Agent Teams

Hub-and-spoke agent architecture with a coordinator, 9 specialist agents, and smart delegation. Each agent has least-privilege tool restrictions.

Skills System

47 composable skill packs for code review, testing, smoke tests, lint debt tracking, session management, project boards, and CTF/pentesting. Language and database-specific skills included.

Hooks Framework

SessionStart, PreToolUse, and PostToolUse hooks for environment setup, safety validation, integrity checking, and auto-linting. Fully configurable per project.

Language Packs

First-class support for Perl, Python, Node.js, Java, Go, Rust, C#, React, Angular, and Spring Boot. Each pack includes language-specific lint rules, test patterns, and coding skills.

CI/CD Pipeline

5-gate evolutionary CI/CD with fitness scoring, Playwright smoke tests, Docker, Kubernetes, and monitoring (Prometheus + Grafana). Self-hosted or GitHub-hosted runners.

Hub-and-Spoke Architecture

The coordinator delegates to specialist agents. Every project inherits the same secure foundation.

project-coordinator

Hub / Orchestrator

solution-architect

Architecture & Design

code-standards-reviewer

Code Quality

test-specialist

Testing & QA

research-analyst

Research & Analysis

database-specialist

Database & SQL

security-analyst

Security & Pentesting

skill-updater

Framework Sync

angular-specialist

Angular & Frontend

spring-boot-specialist

Spring Boot & Java

Security Guard (all tool calls)

validate-bash

Blocks dangerous commands

validate-read

Protects sensitive files

validate-write

Catches hardcoded secrets

validate-fetch

Audits external URLs

setup-env

Integrity verification

compact-reminder

Survives compaction

post-edit-lint

Lint after every edit

angular-version-guard

Angular version safety

spring-boot-version-guard

Spring Boot version safety

Specialist Agents

Composable Skills

Framework Health

Live test results from the latest cognitive-core build

839/839 tests passed

| 16/16 suites | Mar 23, 2026 | 8d2e0e5

ShellCheck

42/42 passed

Skill Frontmatter

64/64 passed

Hook Protocol

25/25 passed

Install Dry-Run

44/44 passed

Update Flow

6/6 passed

Security Hooks

49/49 passed

Agent Permissions

24/24 passed

Workspace Monitor

57/57 passed

Adapter Interface

29/29 passed

Aider Adapter

38/38 passed

IntelliJ Adapter

49/49 passed

MCP Server

18/18 passed

Plugin Structure

94/94 passed

Project Board Providers

140/140 passed

Gitignore Policy

138/138 passed

Recursive Epic Structure

22/22 passed

Component Catalog

Browse all agents and skills — click any card to view the source on GitHub

Agents (10)

angular-specialist

Angular migration, patterns, and architecture specialist (v18-21).

code-standards-reviewer

Reviews code against project conventions and CLAUDE.md guidelines.

database-specialist

Oracle/PostgreSQL optimization, query tuning, and bulk operations.

project-coordinator

Hub orchestrator that delegates to specialist agents and manages cross-project workflows.

research-analyst

External research — libraries, best practices, and API documentation.

security-analyst

Offensive security specialist for pen-testing, vulnerability analysis, and CTF challenges.

Framework sync via /skill-sync — keeps components up to date.

solution-architect

Strategic architect for feature design, technical feasibility, and governance analysis.

spring-boot-specialist

Spring Boot migration, patterns, and architecture specialist (v2-4).

test-specialist

Unit and integration testing, coverage analysis, and QA.

Skills (20)

acceptance-verification

Verify GitHub issue acceptance criteria against codebase evidence.

Batch processing — agent swarms, API batches, and sequential chains.

Language-agnostic code review with project-aware convention checks.

Structured pen-testing and CTF methodology for security learning.

e2e-visual-regression

"Visual regression — baselines, tolerance, debugging, cross-platform."

Quality fitness checks with configurable thresholds and gates.

Track, audit, and create GitHub issues for lint suppression comments across your codebase. Ensures every lint bypass has

Lint and syntax checks on staged files before committing.

Project board — issues, sprints, releases, and triage. Supports GitHub, Jira, YouTrack.

Project status reports from git history, sessions, and WIP state.

1Password-based secrets management — zero plaintext secrets across all machines and CI/CD.

security-baseline

OWASP-aware security rules that adapt to your project's language and framework.

Auto-loads session context — git activity, WIP state, and continuity.

Cross-machine sync — verifies agents, skills, and config match remote.

Checksum-based framework updater that preserves your customizations.

Smoke test all endpoints and auto-create GitHub issues for failures.

Dependency updates, security advisories, and ecosystem intelligence.

Generate test file scaffolds — language-agnostic with config support.

workflow-analysis

Business workflow analysis — stakeholder maps, flows, and risk assessments.

workspace-monitor

Proactive log and build monitoring that detects issues before you report them.

Industry First

AI-Governed Development Lifecycle

The only framework that governs the entire AI-augmented workflow — from research quality to production deployment. Four features no other tool has.

Source Authority Model

5-tier classification (T1-T5) for research sources. AI agents weight findings by authority — official docs (T1) outweigh blog posts (T4). AI-generated slop (T5) is automatically discarded. Decisions require T1-T2 backing.

No equivalent in any AI coding framework

Team-Aware Estimation

Every task tagged with its executor: human, AI, or human+AI. Parallel AI work is free — the human review bottleneck is the real critical path. No more fictional "developer-days" estimates.

Scrum.org identifies the problem — no tool solves it

Graduated Fitness Gates

Quality thresholds that increase as code moves toward production: lint 60% → commit 80% → test 85% → merge 90% → deploy 95%. One unified scoring model across the entire pipeline.

SonarQube has 1 gate — we have 5 graduated

Recursive Epic Verification

Verify an epic and it automatically verifies every sub-issue. Evidence gathered from code, tests, and git history. PASS / PARTIAL / FAIL per criterion. Epic blocked until all children pass.

Jira checks open/closed — we verify acceptance criteria

Board Workflow with Transition Matrix

Strict forward flow. Controlled backward transitions. No skipping columns.

Full transition matrix, compliance mapping, and competitive analysis

Roadmap

Ideas

Backlog

Accepted

Todo

Sprint

In Progress

Active

To Be Tested

Review

Done

Verified

Human Approval Gate — no ticket reaches Done without explicit sign-off. CI automation stops at To Be Tested.

Quality increases with each gate

Lint

60%

Commit

80%

Test

85%

Merge

90%

Deploy

95%

Research-First

Adopt > Adapt > Build. Mandatory research before implementation. Authority-weighted decisions.

SOX Compliance

Different-approver enforcement, dual approval, attributed closures. CI respects the gate.

WIP Limits

Configurable per-column limits prevent context-switching overload. Blocked items tracked.

Agile Metrics

Cycle time, lead time, throughput, flow efficiency. Sprint trend analysis built in.

In Pipeline

EU AI Act Ready

No AI coding framework — commercial or open source — addresses EU AI Act compliance. We are building it: 13 issues, 3 pillars, one epic.

Art. 50

Transparency

Planned

AI interaction disclosure at session startup
AI identity markers in agent YAML frontmatter
Git trailers for AI provenance tracking

Art. 9-15

Risk Management

Planned

Structured JSON Lines audit log format
Per-file SHA-256 integrity checksums
Emergency stop mechanism via marker file
Minimum security level enforcement

Art. 95

Governance

Planned

System card — capabilities, limitations, failure modes
Code of Conduct for voluntary AI Act adoption
Audit log schema + SIEM integration guide
Adversarial hook bypass testing

Zero competitors address EU AI Act in their AI coding frameworks. Cursor, Windsurf, Copilot, ruflo — none have compliance on their roadmap.

Follow the 13-issue compliance epic on GitHub

Certification Alignment

Validated against Anthropic's Claude Certified Architect — Foundations exam standard. Grade A across all 5 domains.

Score: ~959 / 1000 | Passing: 720

Last verified: 2026-03-23 13:30:00

Agentic Architecture & Orchestration

27% of exam

A

Hub-and-spoke with 10 specialist agents
Deterministic hooks for enforcement
Smart delegation with escalation

Tool Design & MCP Integration

18% of exam

A

Structured error responses (errorCategory, isRetryable)
Scoped tool access per agent
Context7 MCP integration

Claude Code Configuration & Workflows

20% of exam

A

.claude/rules/ with path-scoped globs
@import modular CLAUDE.md
context:fork + argument-hint skills

Prompt Engineering & Structured Output

20% of exam

A

Few-shot examples in all 10 agents
Multi-pass review for large PRs
Explicit criteria with structured findings

Context Management & Reliability

15% of exam

A

compact-reminder survives 100K+ compaction
Graduated escalation (allow / ask / deny)
Structured error propagation for recovery

5 / 5 Domains at Grade A Full Report →

Ref: Claude Certified Architect — Foundations Exam Guide v0.1 (Anthropic, 2026) · 5 domains, 30+ task statements

Reference Implementation

Proven in Production

TIMS (Task Information Management System) is the flagship project built on cognitive-core. The v2 workflow maturity audit scored 4.79/5 — 63% above industry average for comparable teams.

4.8 / 5

Maturity Rating

Advanced+ (Level 4-5)

10 Agents

AI Team Members

1 human + 10 AI

95

Project Board Items

7-column Kanban

5 stages

CI/CD Fitness Gates

Lint > Build > Test > Deploy > Smoke

Workflow Maturity Audit v2

Assessment vs. industry average (sources: DORA 2025, OWASP DSOMM, ThoughtWorks, Puppet)

Full audit methodology, evidence, and sources (T1-T2)

Issue Lifecycle Management 5 / 5

Sprint/Iteration Management 4.5 / 5

Branching Strategy 4.5 / 5

CI/CD Pipeline 5 / 5

Code Quality & Standards 5 / 5

AI-Assisted Development 5 / 5

Security & Governance 4.5 / 5

TIMS (cognitive-core)

Industry average (5-10 dev team)

AI-Augmented Team Model

10 specialized AI agents operate as full team members — not passive tools

O

Project Coordinator

Opus-class

O

Solution Architect

Opus-class

O

Web Research Analyst

Opus-class

O

Database Specialist

Opus-class

S

Code Standards Reviewer

Sonnet-class

S

Test Specialist

Sonnet-class

S

Security Analyst

Sonnet-class

S

Skill Updater

Sonnet-class

S

Angular Specialist

Sonnet-class

S

Spring Boot Specialist

Sonnet-class

H

Project Owner / Lead Developer

Human

Key Audit Findings

State Machine Issue Tracking

7-column Kanban with enforced transitions, auto-sprint assignment, and acceptance verification

5-Gate Evolutionary CI/CD

Progressive fitness gates (lint 60%, commit 80%, test 85%, merge 90%, deploy 95%) with Docker-containerized runners

Code Quality 5.0 / 5

Multi-layer enforcement: editor config, pre-commit hooks, CI lint gates, and agent-assisted review

Playwright Smoke Tests

45 endpoint smoke tests run automatically after every deployment to K3s

"The cognitive-core framework with 47 skills, 10 agents, and automated hooks represents the most sophisticated AI-assisted development workflow I have encountered for a project of this size."

— Workflow Maturity Audit v2, March 2026

Roadmap

Built in Europe. For European needs. With global ambition.

Current

v1.0.0

Shipped

10 specialist agents with hub-and-spoke coordination
Source Authority Model (T1-T5) — AI slop filtering
Team-aware estimation — human+AI critical path
Graduated fitness gates (60% → 95%) across pipeline
SOX-compliant board: approval gate, WIP limits, blocked state
Recursive epic verification with evidence gathering
Batch-review skill with 3-tier processing strategy
Information provenance (W3C PROV vocabulary)
Session lifecycle state machine (A2A-inspired)
Shared MCP server for all adapters
Multi-adapter: Claude Code, Aider+Ollama, IntelliJ
11 language packs, 3 database packs, 47 skills
Certification alignment: 959/1000 (Grade A, all 5 domains)
Automated releases via release-please

Next

v1.x

In Progress

VS Code adapter (Copilot + Continue.dev + Cline)
Eclipse adapter (Copilot + EclipseLlama)
Cursor and Windsurf adapters
Enterprise Pack Manifest — installable, updateable custom bundles
Integration test suite — end-to-end verification
EU AI Act compliance — 13-issue epic (Art. 50 + voluntary Art. 9-15)
Step-by-step tutorials and video course

Enterprise

Available

Now

Custom agents and skills for proprietary technology stacks
Custom language packs (SAP, Liquid UI, legacy frameworks)
Custom hooks for corporate security policies
Stack assessment + configuration service
Team training workshops (half-day / full-day)
Free for academic and individual use — contact us for commercial licensing

Build Your Own

cognitive-core is fully pluggable — extend it with custom components

Agents

Create specialist agents with custom tool permissions and model selection. Define expertise, allowed tools, and delegation rules.

Skills

Add composable skill packs for code review, testing, deployment, or any workflow. Skills are invocable commands with scoped tool access.

Packs

Language packs, database packs, hooks — adapt the framework to your stack. Each pack brings patterns, skills, and conventions for a technology.

Community

Contribute agents, skills, and packs to cognitive-core. Help shape the future of AI-augmented development.

Join Discussions Star on GitHub Contribute

Get started in 3 steps

Clone, install, code. Your Claude Code sessions get security, agents, and skills automatically.

1

Clone the framework

$ git clone https://github.com/mindcockpit-ai/cognitive-core.git

2

Install into your project

$ ./cognitive-core/install.sh /path/to/your-project

Interactive setup: choose language, agents, skills, and security level.

3

Start Claude Code in your project

$ cd /path/to/your-project && claude

Hooks load automatically. Agents are ready. Security guard is active.

Keep up to date

Connected projects check for framework updates at session start. When updates are available:

$ ./cognitive-core/update.sh /path/to/your-project

Checksum-based updater: your customizations are preserved, only unmodified framework files are updated.

Step-by-step recipes

New to cognitive-core? Our recipes walk you through common workflows — from your first install to multi-agent coordination.