Skip to main content

LLM Knowledge Update Template

Use this template to brief language models on a new feature or policy change. Store the completed file as docs/llm/<feature-name>.md and align it with the corresponding user article and shared release note. Write in short, declarative sentences and prefer structured formats (tables, key/value lists) for easy parsing.

Feature Snapshot​

FieldValue
Feature name
Release date
Primary goal
AvailabilityPlans / roles / regions
Rollout statusPilot / Partial / GA

Key Facts the Model Must Know​

  • List facts as atomic bullets (one fact per line).
  • Use modal verbs (Must, Never, Only when) to communicate rules.
  • Call out irreversible actions, limits, or safety requirements.

Supported Scenarios​

  • Bullet the core use cases the model should recommend.
  • Provide explicit guardrails that start with verbs (Escalate when…, Do not suggest when…).

Required Inputs & Outputs​

  • Prefer table or key/value blocks for parameters, command syntax, or URLs.
  • Document accepted values, defaults, and validation rules.
  • Use the table below and extend as needed:
ParameterTypeRequiredDefaultAccepted values / validationNotes

Process Walkthrough​

  1. Outline how the feature works from the model’s perspective using imperative verbs.
  2. Encode decision points as explicit conditionals (IF user_role = admin THEN allow bulk upload ELSE escalate).
  3. Mention dependencies on other features or configurations.
  4. If workflow branches are complex, add a small state table or numbered sub-lists.

Tools & Functions​

  • Identify callable tools, APIs, or database lookups.
  • Include function signatures, required parameters, and authentication requirements.

Example Conversations​

Successful assistance​

User: <goal or problem>
Assistant: <ideal answer, with reasoning or clarifying questions>

Escalation or refusal​

User: <scenario that triggers escalation>
Assistant: <brief escalation or refusal message with rationale>
  • Provide at least two contrasting examples (success vs. escalation) and keep assistant replies concise and policy-aligned.

Escalation & Safety​

  • Define when the model must refuse, defer, or escalate using rule-style statements.
  • Reference compliance triggers or audit logging requirements.
  • Link to the matching user help article.
  • Add internal dashboards, runbooks, or spec documents for cross-reference.

Update Log​

  • Append timestamped notes (2024-06-01 — Added new parameter X) whenever guidance changes so downstream systems can track revisions.