This repository contains personal governance prompts and execution artifacts used by the author to structure AI-assisted work.
These prompts are designed for contexts where incorrect inference is worse than refusal. They explicitly prioritize artifact-based authority, execution gating, and epistemic hygiene over speed, flexibility, or conversational convenience.
They are:
- opinionated
- experimental
- tailored to a specific workflow
- subject to breaking change without notice
This repository is public for transparency and versioning convenience. It is not intended as a general-purpose framework.
-
Prompts are law These documents are binding constraints, not suggestions.
-
Conversation-local authority All prompts are pasted explicitly into each conversation. There is no reliance on hidden or persistent state.
-
Separation of concerns
- Unified behavior prompt = constitution
- Overlays = conversation-specific constraints
- Execution cards = procedural law
- Examples = illustrative only
-
Versioned, immutable by default Prompts are updated deliberately, versioned explicitly, and never edited casually. Git tags—not branch head—define what is considered stable or reusable.
Global behavior prompts that apply to every conversation.
These define:
- mode mechanics
- ambiguity gates
- output minimization
- capability verification
- override semantics
Only one unified prompt should be active at a time.
Conversation-specific behavior prompts that narrow scope.
Examples:
- checklist development
- implementation
- audit / comparison
- phase readiness evaluation
These assume the unified prompt is already active and do not restate global rules.
Procedural artifacts that define how a class of conversation proceeds.
Examples:
- project-wide checklist generation
- phase-level checklist generation
- checklist consumption / stepwise execution
- project-to-phase decomposition
Execution cards are referenced or pasted verbatim when governing a conversation.
Non-authoritative examples showing:
- canonical paste order
- safe conversation openings
- correct layering of prompts
These are illustrative only and must not be treated as law.
Prompts that are no longer in use.
Deprecated prompts are retained for audit and historical reference only. They must not be reused.
This repository is for people who treat epistemic risk as a design problem.
It is most useful when incorrect inference is more damaging than delay, and when future verification, provenance, and authority boundaries matter more than conversational fluency.
This repository is for people who want AI systems that are safer than default under fatigue, interruption, and distrust.
The prompts prioritize:
- refusal over plausible invention
- explicit authority over conversational momentum
- artifact handoff over memory or continuity
- processes that remain intelligible to a skeptical future reader
The goal is not correctness guarantees, but risk reduction relative to unguarded, inference-heavy use.
This repository uses Git tags as the source of truth.
Each tag represents a known-good prompt set that has been used in real work.
Example tags:
v2025.01.0v2025.02.0v2025.02.1-audit-fix
Prompts in this repository are treated as governance artifacts, not drafts or notes. Changes should be made deliberately and with awareness of behavioral impact. A modification counts as breaking if it alters scope boundaries, mode availability, execution gating, stopping behavior, or authority hierarchy.
Behaviorally meaningful changes must be recorded in CHANGELOG.md, including:
- what changed
- why it changed
- whether the change is breaking
Formatting-only or non-semantic edits do not require CHANGELOG entries.
Every prompt file must begin with a frozen metadata header:
<!--
PROMPT: <Name>
VERSION: <Tag>
STATUS: canonical | experimental | deprecated
INTENDED USE: <short description>
INCOMPATIBLE WITH: <if any>
-->