| name | copilot-authoring |
| description | Reference for creating and modifying Copilot agent configurations: copilot-instructions.md and SKILL.md files. Essential for the retro agent's job of improving agent capabilities based on session log analysis. |
Copilot Authoring Guide — For Retro Agent
You modify agent configurations to improve their performance. This guide defines the exact formats and rules you must follow.
Agent Configuration Structure
Each agent role lives in .aiscrum/roles/<role>/:
.aiscrum/roles/<role>/
├── copilot-instructions.md # Agent identity, workflow, rules
├── skills/
│ ├── <skill-name>/SKILL.md # Specialized capabilities
│ └── <skill-name>/SKILL.md
└── log/ # Session logs (DO NOT modify, DO NOT load)
└── sprint-N/
The system loads ALL .md files from the role folder (except log/) into the agent's context window. Keep files concise.
copilot-instructions.md Format
# {Role Name} Agent — Copilot Instructions
You are a **{Role Name} Agent** for the AiScrum Pro project.
## Role
One-sentence description.
## Workflow
1. **Step**: Exact action with CLI commands
2. ...
## Rules
- Hard constraints
- Safety boundaries
- What NOT to do
When Modifying Instructions
- Preserve the structure: Keep the Role → Workflow → Rules sections
- Be surgical: Change only what the logs show needs improvement
- Add, don't rewrite: Prefer adding a new rule or clarifying a step over rewriting the whole file
- Keep CLI commands: Agents work via terminal — always include exact commands
- Stakeholder Authority rule: Never remove this — every role must include it
SKILL.md Format
---
name: skill-name
description: "What the skill does and when to use it (max 1024 chars)"
---
# Skill: {Title}
## When to Use
- Trigger conditions
## Instructions
Step-by-step with exact commands.
## Examples
Concrete input/output.
## Rules
- Constraints for this skill
Frontmatter Fields
| Field | Required | Max Length |
|---|
name | Yes | 64 chars, lowercase-with-hyphens |
description | Yes | 1024 chars, specific and descriptive |
When Modifying Skills
- Fix commands: If logs show an agent using wrong CLI syntax, fix it in the skill
- Add examples: If logs show repeated confusion, add concrete examples
- Add rules: If logs show the agent overstepping, add a constraint
- Create new skills: If logs show a recurring capability gap, create a new skill directory with SKILL.md
- Never delete skills: Deprecate by adding a note, but don't remove
How to Apply Improvements
Pattern: Fix a Recurring Error
Log shows: Refiner keeps running `gh issue edit --label` instead of `--add-label`
Action: Update .aiscrum/roles/refiner/skills/issue-editing/SKILL.md
Change: Fix the command example to use `--add-label`
Pattern: Add Missing Guidance
Log shows: Planner doesn't check for milestone conflicts
Action: Add a new rule to .aiscrum/roles/planner/copilot-instructions.md
Change: Add "Check for existing milestone before creating new one" to Rules section
Pattern: New Capability Needed
Log shows: Reviewer needs to check for breaking API changes but has no skill for it
Action: Create .aiscrum/roles/reviewer/skills/api-compatibility/SKILL.md
Change: New skill with instructions for detecting breaking changes
Safety Rules for Retro Agent
- Never modify
log/ — logs are read-only input for analysis
- Propose changes, don't auto-apply — show the user what you plan to change and get confirmation
- One improvement per change — small, focused edits, not bulk rewrites
- Preserve existing behavior — improvements must be additive, not destructive
- Context budget — every byte added to a role folder costs context tokens. Be concise
- Test the change — after modifying, verify the role folder still loads correctly
Reference
- Agent Skills standard
- Agent loader:
src/dashboard/chat-manager.ts → loadRoleContext() recursively loads .md files (excludes log/)
- Available roles: refiner, planner, reviewer, researcher, general