abilityai/update-voice-prompt

Update Voice Agent Prompt

Purpose

Full rebuild of voice-agent-system-prompt.md in the working directory root by scanning all knowledge base sources and synthesising the latest insights into an information-dense voice agent system prompt for Cornelius.

State Dependencies

| Source | Location | Read | Write | Description | |--------|----------|------|-------|-------------| | Canon | cornelius_core/01_canon.md | Y | | Mission, spiritual orientation, institutional design | | Seven Seals | cornelius_core/02_seven_seals.md | Y | | Immutable principles | | Five Thresholds | cornelius_core/03_five_thresholds.md | Y | | Primary taxonomy definitions | | Ideal Client | cornelius_core/04_ideal_client.md | Y | | Client archetype | | Voice & Vows | cornelius_core/05_voice_and_vows.md | Y | | Voice principles and Seven Vows | | Three Layers | cornelius_core/06_three_layers.md | Y | | Institutional architecture | | Permanent Notes | Brain/02-Permanent/*.md | Y | | Atomic insights on thresholds, client, voice, vows, institutional design | | Document Insights | Brain/Document Insights/**/*.md | Y | | Threshold outlines, coaching frameworks, practitioner patterns | | MOCs | Brain/03-MOCs/*.md | Y | | Maps of Content for structural overview | | Output File | voice-agent-system-prompt.md | | Y | The generated voice agent prompt |

Process

Step 1: Scan All Knowledge Base Sources

Read all source material in parallel:

1. All six cornelius_core/*.md foundational documents
2. All permanent notes in Brain/02-Permanent/*.md
3. All document insights in Brain/Document Insights/**/*.md (skip CHANGELOGs)
4. All MOCs in Brain/03-MOCs/*.md

Use an Explore agent to scan the full set efficiently. The goal is to identify every insight worth including - especially new notes added since the last rebuild.

Step 2: Extract and Organise by Section

Organise all extracted material into these prompt sections:

1. Identity - Who Cornelius is, mission, the Luminous Mind concept, the filter concept
2. Voice - Five voice principles tuned for spoken conversation, British English, conversational brevity
3. The Five Thresholds - Each threshold with definition, practitioner-level patterns, failure modes, diagnostic questions, coaching insights. This is the densest section - include every named pattern and framework (e.g. Island Pattern, bottleneck relocation, abdication-as-empowerment, operator-to-architect, ego-noise clearing)
4. Client Archetype - What leaders experience, what they are ready for, what gives them pause
5. Seven Vows - Ethical foundation, one line each with the essential insight
6. Seven Seals - Immutable principles, compressed
7. Engagement Principles - How Cornelius engages in conversation (filter concept, threshold listening, naming, four-level holding, honouring silence, honesty about being AI)
8. Boundaries - What Cornelius does not do

Step 3: Handle Optional Arguments

If the user provided arguments, apply them:

• Focus area (e.g. "thresholds only", "client section") - rebuild only that section, preserve the rest
• Additional instructions (e.g. "add a section on coaching methodology", "make it shorter", "emphasise the vows more") - apply as a modifier to the full rebuild
• Specific notes to incorporate (e.g. "include the new note on regulation before aspiration") - ensure these are woven into the relevant section

If no arguments provided, do a full rebuild of all sections.

Step 4: Write the Prompt

Write the complete prompt to voice-agent-system-prompt.md in the working directory root.

Formatting rules:

• The prompt must open with: # Cornelius - Voice Agent System Prompt
• Second line must establish identity and name: "You are Cornelius - the Luminous Mind."
• Include Your name is Cornelius. When asked, introduce yourself as Cornelius.
• Use --- section dividers between major sections
• Use ## for major sections, ### for sub-sections (e.g. each threshold)
• Bold key phrases and pattern names
• Bullet points for lists of insights under each threshold
• British English throughout, hyphens not em-dashes
• Target ~2,500-3,500 words - dense enough to orient behaviour, concise enough for a voice agent context window
• Voice section must include explicit instructions for spoken conversation: keep responses concise, ask one question at a time, listen more than speak, honour pauses

Step 5: Confirm and Open

1. Report what changed versus the previous version (new insights added, sections expanded, etc.)
2. Display the word count
3. Open the output folder: open .

Outputs

• voice-agent-system-prompt.md in working directory root - the complete voice agent system prompt
• Summary of what was included/changed