Skip to main content

Documentation Index

Fetch the complete documentation index at: https://mintlify.com/xxyoudeadpunkxx/canon-boundary-guard-codex/llms.txt

Use this file to discover all available pages before exploring further.

Canon Boundary Guard organizes every piece of information Codex works with into one of six provenance layers. These layers do not judge whether something is true — they describe where material came from and whether it can safely become persistent project content. Understanding them is the foundation for working with the skill: once you know which layer a piece of content belongs to, the persistence and tagging rules follow directly.

The Six Layers

The layer definitions below come directly from SKILL.md and the compact reference frame in references/frame.md. They are reproduced here without modification.
TagNameOne-line description
L0EVIDENCEProject files, git state, tests, schemas, lockfiles, diagnostics, command output, verified tool output from the current task.
L1SHAPINGConversation material not explicitly approved as a durable project decision. Present in context, not in the project.
L1AAUTHORIZED DELTAConversation material the operator has explicitly approved for persistence in this turn. Becomes evidence only after written to a persistent artifact.
L2AGENT CONTROLInstructions, steering, reminders, or constraints given to the agent to shape its behavior. Not project content.
L2ACODEX INSTRUCTION CHAINAGENTS guidance loaded by Codex from the instruction chain. Agent-control authority, not project content.
L3MODEL PRIORUnverified model memory, generic best practice, assumed framework behavior, version claim, or unstated convention not grounded in local evidence.
Layers do not decide whether something is true — they describe where material came from and whether it can safely become persistent project content. A claim from model memory is not made more reliable by tagging it L3; the tag simply makes its origin visible so the operator can decide what to do with it.

The Persistence Boundary

The most consequential distinction in the taxonomy is between L0 and everything else. L0 is the only layer that can become persistent project content without explicit operator approval. It represents verified evidence already grounded in the repository: files on disk, confirmed git state, test results, command output. Preserve or reorganize L0 freely; it is the ground truth of the session. L1, L1A, L2, L2A, and L3 can all shape how Codex reasons and what it produces, but none of them should silently enter the repository:
  • L1 does not persist. Conversation material that has not been explicitly approved remains in context only.
  • L1A can be written, but only within the scope the operator has explicitly approved for this turn. Once written to a persistent artifact, it graduates to L0.
  • L2 does not persist unless the operator explicitly requests agent-facing operating instructions as a project artifact.
  • L2A governs Codex behavior at its scope and does not persist into project content unless explicitly requested.
  • L3 does not persist unless it has been verified against L0 evidence or the operator has explicitly approved it.
When provenance is unclear before a write operation, surface it rather than resolving it silently. If evidence conflicts, stop and report — do not resolve by recency, confidence, or intuition.

What Comes Next

L0 Evidence

What qualifies as ground-truth evidence, how conflicts are handled, and why L0 is the only layer that persists without explicit approval.

Non-L0 Layers

Detailed definitions, example uses, and persistence rules for L1, L1A, L2, L2A, and L3.

Build docs developers (and LLMs) love