Frequently Asked Questions
What is Principles Disciple?
Principles Disciple is an AI Agent Governance System that helps owners turn repeated agent corrections into reusable behavioral principles.
Every principle is owner-approved, reviewable, and reversible, and its effect on future agent behavior stays observable.
What is a principle in Principles Disciple?
A principle is not a simple rule.
It is a reusable behavioral insight extracted from experience — abstract enough to guide decisions across situations, concrete enough to change the agent's next action. See What is a Principle?.
Is PD an AI memory system?
No.
Memory stores information. PD focuses on transforming validated experience into behavioral principles.
Memory answers "what happened?". PD answers "what should change because of what happened?". See PD vs AI Memory.
Does PD automatically modify my AI agent?
No.
PD proposes principles, but activation requires owner review.
The system brings evidence and proposals; the Owner keeps judgment. Every activation is reversible, and its effect stays observable.
Who should use PD?
Developers and AI-native builders who frequently use AI agents and want consistent behavior across sessions.
If you find yourself correcting the same agent behavior again and again, PD turns that repeated correction into a lasting, governed principle.
Is PD a prompt management tool?
No.
Prompt engineering tells agents what to do before execution. PD learns from actual interactions and validated experience after execution — then converts them into owner-approved principles. See PD vs Prompt Engineering.
Does PD automatically generate rules?
Be precise about two different steps:
Drafting is automated. From real behavior evidence, PD's internalization pipeline drafts candidate principles and candidate rule implementations unattended, and replays them against historical cases.
Taking effect is not. A candidate rule stays inert until it passes owner review; an approved rule starts in observation-only (shadow) mode; and it only ever enforces after the Owner explicitly promotes it.
So nothing blocks or corrects on its own: generation produces proposals; authority stays with the Owner.
Is PD another agent runtime or harness?
No.
Claude Code, Codex, OpenClaw and their peers answer "how does the agent run?" — they execute tasks, call tools, and orchestrate the loop. PD answers a different question: how does the agent improve under owner authority?
PD does not execute tasks or manage tools. It installs as a governance layer on top of the host you already use (OpenClaw and Codex today), applying principles through the host's own extension points — prompt hooks and tool-call hooks. You don't switch agents to use PD; you keep your agent, and its behavior becomes governable.
Does PD cost extra tokens?
Some — honesty first: reflection runs LLM calls beyond your normal tasks.
Three things keep that cost grounded:
- Reflection is priced by experience, not by task. One deep reflection settles into reusable principles and rules that every later session reuses.
- Injection is hard-capped. Active principles enter the context under a strict size budget, and runtime rules enforce bottom lines entirely outside the model — zero context-window usage.
- The model is your choice. PD uses the models your host already has configured, including locally hosted ones (e.g., LM Studio), and each internal role can be bound to a different profile.
And the most expensive thing in agentic work is rarely the token bill — it is a wrong direction discovered late. Avoiding even one such detour can offset more tokens than PD's governance ever adds.
Which environments does PD support?
PD currently integrates with OpenClaw and Codex hosts. See the installation guide for details.
Does PD train the AI model?
No.
PD does not modify model weights. It governs agent behavior through owner-approved principles and runtime mechanisms — a layer around the agent, never inside the model.
Does PD only inject prompts?
No.
Prompt guidance is only one governance channel. The core idea is principle internalization: experience becomes owner-approved principles, and those principles are applied through governance mechanisms — including, for a few vital bottom lines, executable runtime rules that the Owner has explicitly promoted.
How is PD different from AGENTS.md or CLAUDE.md?
Those files provide static instructions.
PD creates an evolving governance layer based on actual experience: real behavior produces evidence, evidence becomes owner-approved principles, and principles are applied through the governance runtime — reviewable, reversible, observable at every step.
Where can I read the full product definition?
See the canonical definition in the repository, or the comparisons page for how PD differs from adjacent categories.