Cognitive Core Skills
Cognitive core skills are the mental operating capabilities an LLM or AI Agent needs to move from chat response to useful digital co-worker. An LLM predicts and generates language. An AI Agent wraps that LLM with memory, planning, tools, integrations, verification, and action control so it can pursue goals inside real workflows.
Cognitive core skills are the agent operating brain: perception to know the situation, memory to preserve continuity, reasoning to judge, planning to sequence work, action selection to execute, verification to confirm progress, learning to improve, and governance to stay safe.
Repository overview
This repository contains a universal, industry-neutral taxonomy package for:
- LLM capabilities
- SLM capabilities
- AI agent capabilities
- World model capabilities
- Text-based cognitive skills
- Human-action cognitive skills
- Persistent LLM Wiki knowledge operations
- OKF-style knowledge cataloging operations
Why star or fork this project
- Practical: ready-to-use taxonomy data, schemas, and validation tests.
- Operational: includes human-action skills and governance-oriented constructs.
- Reusable: JSON, YAML, and markdown skill cards for easy integration.
- Collaborative: contribution templates, roadmap, and CI in place.
Quickstart
- Read the taxonomy package docs in
docs/cognitive-core-skills/. - Explore machine-readable data in
data/. - Browse generated skill cards in
Skills/index.md. - Run validation locally.
Current status
- Taxonomy version:
1.0.0 - Skill count:
159 - Domains covered:
13 - Generated skill files in
Skills/:159skill cards plusSkills/index.md - CI enabled for taxonomy and benchmark fixture tests
Capability model differences
| System type | Primary strength | Typical limitation without augmentation |
|---|---|---|
| LLM | Broad language understanding, synthesis, reasoning, drafting | Weak continuity and no autonomous bounded action by default |
| SLM | Low-latency, low-cost, private, constrained inference | Narrower transfer and lower open-domain depth |
| AI Agent | Goal pursuit across memory, planning, tools, and actions | Requires robust verification, control gates, and oversight |
| World Model | State and dynamics modeling, simulation, forecasting | Requires grounded observations and updated environment state |
Key package paths
Documentation:
docs/cognitive-core-skills/README.mddocs/cognitive-core-skills/taxonomy.mddocs/cognitive-core-skills/text-based-skills.mddocs/cognitive-core-skills/human-action-skills.mddocs/cognitive-core-skills/world-model-skills.mddocs/cognitive-core-skills/llm-wiki-skills.mddocs/cognitive-core-skills/okf-skills.mddocs/cognitive-core-skills/evaluation-rubric.mddocs/cognitive-core-skills/cognitive-debt-index.mddocs/cognitive-core-skills/glossary.mddocs/cognitive-core-skills/CHANGELOG.md
Machine-readable artifacts:
data/cognitive-core-skills.jsondata/cognitive-core-skills.yamlschemas/cognitive-core-skill.schema.jsonschemas/cognitive-core-taxonomy.schema.json
Benchmarks and examples:
benchmarks/rubric-task-traces.jsonexamples/cognitive-core-skill-card.mdexamples/agent-skill-assessment.md
Generated skills directory:
Skills/index.mdSkills/<skill_id>.md(one file per taxonomy skill)
Tests and CI:
tests/test_cognitive_core_taxonomy.pytests/test_rubric_benchmark_fixtures.py.github/workflows/taxonomy-ci.yml
Community and project health:
CONTRIBUTING.mdGOOD_FIRST_ISSUE_GUIDE.mdRELEASE_CHECKLIST.mdROADMAP.md.github/ISSUE_TEMPLATE/bug_report.yml.github/ISSUE_TEMPLATE/feature_request.yml.github/pull_request_template.md
Recent taxonomy update
Added human-action skill:
ha_consequence_understanding- Human-Action Consequence Understanding
Purpose: understand likely downstream operational, compliance, and workflow consequences before executing human-action tokens.
Added planning skill:
plan_long_horizon_skill- Long-Horizon Skill
Purpose: maintain coherent long-horizon execution across milestones, dependencies, and replanning triggers.
Run validation locally
py -m pytest tests/test_cognitive_core_taxonomy.py tests/test_rubric_benchmark_fixtures.py -q
How to add a new skill
- Add the skill card to
data/cognitive-core-skills.json. - Mirror the same skill in
data/cognitive-core-skills.yaml. - Ensure unique
idand uniquename. - Add valid
related_skillsreferences. - Regenerate the
Skills/folder markdown cards from taxonomy data. - Update
docs/cognitive-core-skills/CHANGELOG.mdunderUnreleased. - Run the local pytest command above.
Regenerate skill markdown files
When taxonomy data changes, regenerate skill card markdown files in Skills/ so they remain synchronized with data/cognitive-core-skills.json.
Expected result:
- one markdown file per skill id
- one generated index file at
Skills/index.md
Run:
py scripts/generate_skills.py
Attribution and license boundaries
This taxonomy package adapts high-level architecture/documentation patterns from:
third-party/autoresearch-masterthird-party/knowledge-catalog-mainthird-party/knowledge-catalog-main/okf
No third-party source code is copied into taxonomy artifacts.
Star history
If this taxonomy is useful to you, a star helps others discover it and tells us which direction to invest in next.
Support this project
- Star the repo to bookmark it and help with discoverability.
- Fork the repo if you want to adapt the taxonomy, schemas, or skill cards for your own agent stack.
- Open an issue for a missing skill, a schema question, or a benchmark idea. See
GOOD_FIRST_ISSUE_GUIDE.mdfor beginner-friendly entry points. - Share a "Show and tell" write-up if you build something on top of this taxonomy once Discussions are enabled.