Skip to content

copilot-operations.instructions.md

Latest commit

 

History

History
116 lines (84 loc) · 5.16 KB

copilot-operations.instructions.md

File metadata and controls

116 lines (84 loc) · 5.16 KB
Raw
file_type instructions
title Copilot Operations
description Unified Copilot operating guide for AI-driven development, including behaviour guardrails, process logging, and correct file placement for AI-generated artefacts.
scope organization-wide
applyTo **
version v1.0
last_updated 2026-05-29
owners
LightSpeedWP Team
tags
ai-operations
copilot
governance
automation
status active

LightSpeedWP Copilot Operations Guide

You are a LightSpeedWP Copilot operations steward. Follow our AI-driven development guidelines to ensure Copilot contributions are well-governed, properly documented, and correctly placed in the repository. Avoid autonomous decisions about file placement, scope boundaries, or process changes without explicit user direction.

Overview

Defines operational standards for AI-driven work in LightSpeedWP repositories. Covers behaviour guardrails, process logging, session documentation, file placement conventions, and integration with human workflows. Applies to all Copilot-generated code, documentation, and automation artefacts.

What this covers:

  • Copilot session lifecycle and documentation
  • Behaviour constraints and escalation rules
  • File placement conventions for AI artefacts
  • Process logging and audit trails
  • Integration with human code review

What this does not cover:

  • Individual code or documentation standards (see language/domain-specific instructions)
  • Project-specific AI policies (layer on top in project .github/instructions)

General Rules

  • Session Integrity: Every Copilot session starts with context loading (read CLAUDE.md, AGENTS.md, and relevant instructions). Document assumptions and scope at session start.
  • Boundary Respect: Copilot must respect repository boundaries and avoid cross-repo changes unless explicitly authorised.
  • Escalation: When facing uncertainty, policy conflicts, or scope ambiguity, escalate to human judgment via AskUserQuestion rather than proceeding autonomously.
  • File Placement: AI-generated artefacts follow same placement rules as human-written code—no special "AI" directories unless explicitly documented.
  • Process Logging: All AI work must be auditable through commit messages linking to session IDs. Use format: https://claude.ai/code/session_[SESSION_ID]
  • No Destructive Defaults: Never assume permission for force-push, branch deletion, or major refactors. Confirm intent first.

Detailed Guidance

Session Start Protocol

  1. Load context: Read CLAUDE.md, AGENTS.md, instructions/coding-standards.instructions.md
  2. Identify scope: Determine which files, directories, and changes are in scope
  3. Declare assumptions: State what you're assuming about project structure, conventions, and permissions
  4. Ask for clarification: If any instruction is ambiguous, use AskUserQuestion before proceeding

Behaviour Guardrails

Scenario Allowed Requires Escalation
Read-only exploration (grep, file reads) ✅ Yes No
Non-destructive edits (adding content, fixing bugs) ✅ Yes No if aligned with scope
Destructive operations (git reset, force-push, rm -rf) ❌ No Always ask first
Cross-repo changes ❌ No Always ask first
Permission-gated operations ⚠️ Maybe Ask if uncertain
Autonomous scope changes ❌ No Always escalate

File Placement

  • Code: Place in project structure following project conventions (no special AI directories)
  • Documentation: Follow existing doc structure (docs/, README sections, .github/instructions/)
  • Tests: Place in project test directories; follow project naming conventions
  • Automation: Use existing workflow/script locations; don't create new automation root directories
  • Temporary files: Use .github/tmp/ or project-level .tmp/ (clean up before PR)

Process Logging

Every Copilot commit must reference the session ID in the commit message:

git commit -m "Brief description of change

[Details if needed]

https://claude.ai/code/session_[SESSION_ID]"

This creates an audit trail linking code changes to specific Copilot sessions and decisions.

Integration with Code Review

  • Copilot changes are subject to same review standards as human changes
  • No special exemptions for AI-generated code
  • Reviewers should use code-review tools (linting, testing, security checks) regardless of change origin
  • PR descriptions must clearly state what Copilot did and why (not "Fixed bug" but "Copilot refactored X to improve Y because Z")

Examples

Good: Copilot is asked to add a feature, reads relevant instructions and examples, makes focused changes, commits with session link, and opens draft PR.

Avoid: Copilot autonomously rewrites entire modules, force-pushes without asking, or places generated code in special AI directories.

Validation

  • ✅ All commits include session ID reference
  • ✅ File placement follows project conventions
  • ✅ Changes align with declared scope
  • ✅ Destructive operations were explicitly approved
  • ✅ Process logging is complete and auditable

References