Skip to content

COL: CO for Learners

Application COL Status In Development License CC BY 4.0 Base CO v1.1

COL gives you a structured six-phase workflow for producing academic work with AI. You do the thinking. AI helps you think better. Every phase requires your explicit approval before the next begins.

COL works in any academic discipline. The base workflow stays the same; your discipline adds specialized agents and knowledge. Finance students get COL-F (24 finance agents). History, biology, and other subjects can be added the same way.

PhaseCommandWhat happensEvidence required
01 Study/studyResearch the topic, gather sourcesSources cited and verified
02 Outline/outlineCreate a prioritized task planStudent explicitly approves plan
03 Draft/draftProduce the deliverable, one task at a timeEvidence-based completion
04 Challenge/challengeReview agents critique the workReview assessment documented
05 Learn/learnCapture reusable knowledge (study patterns, source lists, writing conventions) for future sessionsHuman approval required
06 Submit/submitPackage and submit the final outputSubmission checklist complete

Between each major phase, the student makes an explicit judgment call:

  1. Study to Outline: Is the research scope correct and complete?
  2. Outline to Draft: Is the task plan appropriate for the assignment?
  3. Draft to Challenge: Is the deliverable substantive enough for review?
  4. Challenge to Submit: Have review findings been addressed?

AI cannot bypass these gates. The student’s judgment at each gate is the assessment signal.

COL provides the universal workflow. Your discipline adds specialized agents, knowledge, and rules:

COL (base methodology)
├── Academic Writing agents
├── Research agents
├── Study Support agents
├── Review & Quality agents
└── Management agents
├── COL-F (Finance) ← 24 agents, 20 skills, 13 rules (production)
├── COL-H (History) ← planned
├── COL-B (Biology) ← planned
└── COL-[X] (your discipline)
CategoryBase COL providesSubject layers add
Academic Writingacademic-writer, citation-specialist, presentation-designer
Researchresearch-assistant, peer-reviewerSubject-specific research methods
Study Supportconcept-explainer, exam-coachSubject tutors (course-specific)
Review & Qualitydeep-analyst, assignment-analystSubject-specific quality criteria
Managementtodo-manager

Amnesia: AI forgets assignment constraints, rubric criteria, and course conventions between sessions. COL reloads your assignment parameters, rubric dimensions, and integrity rules at every session start.

Convention drift: AI reverts to generic academic style instead of course-specific conventions. Wrong citation format. Wrong source quality threshold. Wrong level of analysis. COL’s guardrail rules enforce course-specific conventions.

Safety blindness: AI presents its analysis as the student’s own work. Fabricates citations. Skips pedagogical scaffolding. COL’s integrity rules and enforcement hooks detect and block these patterns.

COL enforces integrity as an absolute directive:

  • No fabricated citations. Every reference must be verifiable.
  • Proper AI disclosure per venue requirements.
  • Student judgment must be visible at every approval gate.
  • The deliberation log (what the student decided and why) is the primary evidence of learning.

You need Claude Desktop (free, any operating system) or Claude Code (command line). No coding required.

  1. Download the COL workspace from GitHub and unzip it
  2. Open Claude Desktop, switch to the Cowork tab, and open the unzipped folder
  3. Type /start and tell COL what you are working on

The full setup guide walks through each step with screenshots.

COL-F (Finance) is the first production implementation. 24 specialized agents, 20 skill directories, 13 enforcement rules, 5 hooks. Active use in undergraduate and graduate finance education.