UAI-1 is the current public message standard for structured AI-to-AI communication.
UAIX shield mark Public interoperability charter UAIX.org Universal Artificial Intelligence Exchange Public standards publication
Current UAI-1
  • Current version UAI-1
  • Plain-English definition Open message format for auditable AI-to-AI exchange
  • Current attribution Michael Joseph Kappel, MCP
  • Start here Get Started, Project Handoff, UAI-1, validator, roadmap, and changelog

AI Memory

agent-instructions.uai

Gives role-specific agent instructions that do not belong in global identity, totem, taboo, or coding rules. Required for project/developer packages that declare agent-specific operating instructions.

  • Record UAIX-MEMR-2985
  • Path /en-us/ai-memory/uai-files/agent-instructions-uai/
  • Use Canonical public record

Document status

Public standards page Published on UAIX as part of the current public standards record
Code
UAIX-MEMR-2985
Surface
AI Memory
Access
Public and linkable

How to use this page

Use this file detail page to confirm the exact required scope, review requirement, no-op trigger, and related .uai records before relying on the file in a package.

Package memory path

AI Memory Package WizardProject HandoffAgent File HandoffAGENTS.md Specification

Required File Sets

Every .uai file belongs to a declared package scope

Use this guide to keep package files in one of the closed UAIX states: universal required, profile-required, content-triggered required, configuration-specific required, or out of scope.

Universal

Five launch-baseline anchors travel first

identity.uai, world-context.uai, totem.uai, taboo.uai, and short-term-memory.uai are required for every launch-baseline UAIX memory or handoff package.

Profile

Declared package type controls the rest

Project Handoff, Agent File Handoff, Long-Term Memory, LLM Wiki, and release-bound packages each carry their own required file set.

No-code assistants

Coding standards stay out of scope until code-like work exists

.uai/coding-standards.uai is required for project/developer and code-bearing packages; it is out of scope for no-code assistant packages until code, automation, executable workflow, or code-like artifacts are declared or detected.

Package memory path

AI Memory Package WizardGenerate supported file sets from the canonical source rules.Project HandoffProject-startup package profile and read order.Agent File HandoffLocal file intake, disposition, archive, and proof-of-use path.AGENTS.md SpecificationFront-door project guidance and local memory discovery.
Core ruleClosed file states
universal required | profile-required | content-triggered required | configuration-specific required | out of scope

Validators and agents should rely only on files required for the declared package scope and ignore out-of-scope files as absent.

Gives role-specific agent instructions that do not belong in global identity, totem, taboo, or coding rules.

File .uai/agent-instructions.uai
Category Active project memory files
Required status Required for project/developer packages that declare agent-specific operating instructions
Main reader Receiving AI and instruction reviewer
Update frequency When role-specific instructions, tool order, collaboration style, or escalation behavior changes.
Human review requirement Required for instructions that change authority, safety, public claims, deployment, or user-facing behavior.
No-op trigger type Instruction conflict, hidden authority change, stale tool order, or contradiction with taboo.uai.

What this file is

agent-instructions.uai is a bounded UAIX memory record. It is meant to be read before action, compared against related package files, and updated only through a reviewed package-memory change.

Why necessary

The file keeps this class of knowledge explicit instead of burying it in chat history, private assumptions, or stale generated summaries.

When required

Required for project/developer packages that declare agent-specific operating instructions.

What belongs

  • Accepted current facts for the declared package scope.
  • Source, owner, reviewer, or evidence pointers needed to verify the record.
  • No-op or escalation triggers that protect the package boundary.
  • Compact instructions the next reader can act on without guessing.

What does not belong

  • Raw chat dumps, unresolved brainstorming, or unreviewed generated summaries.
  • Secrets, private credentials, personal data beyond the declared minimum, or unsupported public claims.
  • Rules that belong in a more specific file such as taboo.uai, constraints.uai, test-plan.uai, or deployment-memory-and-test-report.uai.
  • Historical notes that should live in archives until reviewed and promoted.

How receiving AI should use it

  • Read it before acting in the scope it controls.
  • Compare it with identity.uai, world-context.uai, totem.uai, taboo.uai, and short-term-memory.uai before trusting it as current.
  • Use the no-op trigger as a stop point when the file is missing, stale, contradictory, or outside the declared package profile.
  • Record any update in short-term-memory.uai and the package changelog when the change affects future agents.

How human reviewer should use it

  • Confirm the file states a concrete required scope rather than a softer invitation to read it.
  • Check that current facts have an owner, evidence path, or review source.
  • Confirm outdated material was moved to archives or replaced with a dated current record.
  • Reject vague language that lets agents ignore package-critical material.

Scenario examples

  • A receiving agent opens a Project Handoff package and uses this record to decide whether work may proceed.
  • A reviewer sees a contradiction between files and uses the no-op trigger to require clarification before release.
  • A maintenance pass promotes a reviewed fact into this file and retires the stale note from active memory.

No-op and human-review triggers

  • Missing file for its declared required scope.
  • Conflict with universal launch-baseline files.
  • Current package work depends on an unreviewed fact.
  • The file claims authority beyond its declared profile, mode, package type, or capability.

Update and maintenance rules

  • Keep entries short enough to be read at startup.
  • Date meaningful updates and record why the change happened.
  • Move stale detail to archives after accepted current facts are captured.
  • Align generated exports and human-facing memory after changes.

Relationship to other .uai files

  • identity.uai names the package and owner.
  • world-context.uai locates the operating environment.
  • totem.uai and taboo.uai anchor mission and boundaries.
  • short-term-memory.uai points to the newest active state.

Minimal example

Code example
# agent-instructions.uai
file: .uai/agent-instructions.uai
required_status: Required for project/developer packages that declare agent-specific operating instructions
owner: human-reviewed owner or steward
current_state: accepted current record for this package scope
no_op_trigger: Instruction conflict, hidden authority change, stale tool order, or contradiction with taboo.uai.
evidence: link or local pointer to reviewed source
last_reviewed: YYYY-MM-DD

Completeness checklist

  • Required status names the exact scope.
  • Owner or reviewer is clear.
  • Current facts match related .uai files.
  • No-op triggers are concrete.
  • Stale material has an archive or replacement path.

Common mistakes

  • Using softer language that agents treat as skippable.
  • Mixing current truth with raw source intake.
  • Letting the file contradict another required record.
  • Leaving no review path for high-impact changes.

Machine-readable digest

Agents should treat this digest as page-orientation evidence, not runtime authority.