Corpas Core Starter

AI-augmented cognitive decision support. Think clearer, ship faster, protect what matters.

The Problem

Knowledge workers are drowning. Not in tasks, but in decisions. Every email, meeting request, collaboration offer, and "quick favour" competes for the same finite cognitive budget. By the end of a typical week, you have been busy but cannot point to a single outcome that moved your mission forward.

AI tools promise "10x productivity." But speed without direction just produces more noise, faster.

The Approach

Corpas Core is a different kind of tool. It does not automate your work. It enforces strategic discipline on your attention.

The system is built on three rules (from the AI Fluency framework):

1. Does this accelerate my core mission? If no, decline or defer. "Interesting but tangential" is the enemy of focused impact.
2. Would accepting this delay my core deliverables? Time is zero-sum. Protect your deep work hours.
3. If uncertain, build a prototype and test reality. Ship something small. Reality provides feedback that speculation cannot.

Every task, decision, and approval passes through this filter.

What It Does

Corpas Core Starter is a command-line tool that provides:

• Inbox triage: Classify emails by urgency and get draft replies in your voice. Pipe in emails or read from file.
• Email drafting: Generate replies with configurable tone (direct, warm, formal, decline, defer) and word limits.
• Strategic reasoning: Ask the system to evaluate decisions against your mission. "Should I accept this invitation?"
• Three-Rule Filter: Run any incoming request through the framework. Get a PROCEED/DECLINE/DEFER/PROTOTYPE verdict.
• Document digest: Summarise meeting transcripts, papers, or reports into actionable points with suggested tasks.
• Session handoff: Capture full project state as a structured document. Resume seamlessly across tools and sessions.
• Approval gates: Nothing gets published or sent without explicit human sign-off.
• Decision logging: Every significant decision recorded with reasoning, so future-you can review and learn.
• Weekly shipping rhythm: Track artifacts shipped per week. Did you ship something that matters, or did the week evaporate?
• Strategy-as-config: Define your mission, priorities, and voice in a YAML file. The system holds you to them.

Quick Start

# Install
pip install -e .

# Initialise (creates database and strategy config)
corpas-core init

# See your mission and priorities
corpas-core brief

# Triage your inbox (pipe in emails or use a file)
cat emails.txt | corpas-core triage
corpas-core triage --file inbox_export.txt

# Draft a reply to a specific email
corpas-core draft --file email.txt --tone direct --sign-off "Dr Jane Smith"

# Reason about a decision
corpas-core think "Should I accept this conference invitation in April?"

# Filter an incoming request using the Three-Rule Framework
corpas-core filter "A colleague wants me to co-author a review paper on topic X"

# Summarise a meeting transcript into actionable points
corpas-core digest --file meeting_notes.txt

# Add a task
corpas-core add "Write first blog post" --priority 1 --public

# Claim next task and get an execution brief
corpas-core autopilot --once

# Complete it
corpas-core complete --task-id 1 --outcome "Published blog post" --artifact-url "https://..."

# Weekly review
corpas-core review --local

# Save session state for continuity
corpas-core handoff --notes "Finished draft, needs review"

Commands

Daily workflows:

Command	Purpose
triage	Classify emails and recommend actions with draft replies
draft	Draft a reply in your voice (--tone direct/warm/formal/decline/defer)
digest	Summarise a document into actionable points with suggested tasks
think	Reason about a question using your strategy as context
filter	Apply the Three-Rule Framework to an incoming request

Task management:

Command	Purpose
add	Add a task (--priority 1-5, --public, --description)
tasks	List tasks (filter by --status pending/in_progress/done)
autopilot --once	Claim next high-priority task with execution brief
complete	Mark task done with outcome and optional artifact URL
log	Record a decision with context and reasoning

Governance and continuity:

Command	Purpose
gate	Queue something for human approval before publishing
approvals	View approval queue
resolve	Approve or reject a queued item
review	Weekly strategic review (--local for offline, or LLM-powered)
brief	Show mission, priorities, and snapshot (--json for machine output)
handoff	Save full project state as a markdown document (--print for stdout)

LLM Reasoning

Set ANTHROPIC_API_KEY and install the optional dependency:

pip install -e ".[llm]"
export ANTHROPIC_API_KEY=sk-ant-...

The think, filter, review, triage, draft, and digest commands will use Claude to reason in the context of your strategy. Without the API key, these commands output a structured prompt you can paste into any LLM.

Configure For Your Purpose

Edit config/strategy.template.yaml:

mission:
  statement: "Your mission here."

priorities:
  - name: "Your first priority"
    definition_of_done: "How you know it is done."
  - name: "Your second priority"
    definition_of_done: "Measurable outcome."

Customise your voice for email drafts:

voice:
  tone: "clear, concise, confident"
  style: "Claim-Justification-Implication"
  avoid:
    - "I hope this email finds you well"
    - "Sorry for the delay"
    - "hedging language"
  prefer:
    - "Active voice"
    - "Concrete next steps"
    - "Specific timelines"

The system does not judge your mission. It holds you to it.

The Full System Behind This

This starter is extracted from a working system that runs daily. The full Corpas Core manages:

• 2 AI agents across Telegram and WhatsApp, each with distinct personas
• 8 scheduled job groups (research tracking, inbox triage, podcast intelligence, marking automation)
• 14,000+ embedded documents in a personal knowledge base (notes, emails, papers)
• Automated paper tracking, grant pipeline management, and weekly artifact shipping
• Human-in-the-loop approval for every outbound action

One academic. Two agents. 500 lines of reusable scaffolding. The starter gives you the decision-support skeleton so you can build your own version.

Design Principles

From AI Fluency: A Practical Guide to Leveraging AI Chatbots for Academic and Professional Work by Dr Manuel Corpas:

• You are the principal; AI is the agent. Your expertise, values, and voice remain central. AI deploys them more efficiently.
• A 10x professional does not think 10x harder. They think once and reuse that insight 10x. That is what systematic AI collaboration enables.
• Human-in-the-loop is not a limitation. It is the design. Anthropic's research shows a 17% comprehension drop when people fully delegate to AI. The approval gate exists for a reason.
• Local-first, privacy-first. All state in SQLite. No cloud dependency. No telemetry. Your decisions stay yours.
• Ship weekly or it did not happen. Strategy without execution is delusion. The weekly artifact count is the accountability mechanism.

What This Is Not

• Not a task manager. There are better ones. This is a strategic filter.
• Not a chatbot. It reasons when asked, then gets out of the way.
• Not a replacement for judgment. It is a system for recording, reviewing, and improving judgment over time.

Requirements

• Python 3.10+
• PyYAML (installed automatically)
• Optional: anthropic SDK for LLM reasoning

License

MIT. See LICENSE.

Author

Dr Manuel Corpas is a Senior Lecturer at the University of Westminster, Turing Fellow, and builder of AI-augmented research workflows. This tool is the operational companion to AI Fluency.