| name | work-item-creation |
| description | Rules and checklist for creating work items on GitHub Issues or Azure DevOps. Use when you need to create issues, user stories, tasks, epics, or features on the board. Do not use for platform configuration (use board-config), managing workflow of existing tasks (use work-item-workflow), or for comments on work items. |
Required Checklist
Before creating ANY work item, verify:
- Platform: Read
.memory/board-config.md to identify GitHub or Azure DevOps
- Duplicates: Check if a similar work item already exists on the board
- Required tags/labels: all present (see section below)
- Title in correct format: see Title section
- Body filled in: per the platform format (see GitHub and Azure DevOps sections)
- Hierarchy: parent linked correctly
- Durable descriptions: body describes behavior using domain language, not implementation details (see Durability Rules below)
- Content style: title and body follow the Content Style section below
Content Style
Work items live on the board, get rendered in GitHub/ADO UIs, and are read by humans and agents long after creation. Apply these rules to every title, description, comment, and acceptance criterion.
| Rule | Why | Example to avoid |
|---|
| No emojis or decorative Unicode (e.g., →, •, ★, ✓, 🎯, ✅, ⚠️, ❌) | Renders inconsistently across boards, screen readers, and exports | 🎯 Goal: ship X |
| No em-dash (—), en-dash (–), or hyphen as separator between concepts | Ambiguous in titles, breaks search, often AI-generated tell. Rewrite the sentence. | Auth flow — refactor → Refactor auth flow |
Never use #<number> in free text | Azure DevOps auto-converts to a work item link, breaking references and creating wrong cross-links. Use "first", "third option", or rephrase. | #1 priority → top priority |
| No contrastive constructions ("not just X, it's Y", "more than just", "goes beyond") | Generic AI framing that adds words without information | This is not just a fix, it's a rethink → Replace the retry logic |
| Direct active voice, no rhetorical questions | Work items are commands or descriptions, not essays | What if we cached responses? We should cache them. → Cache responses for 60 seconds |
| No promotional or exaggerated language | Work items track work, not pitch it | seamless, revolutionary, world-class |
| Keep titles short and concrete; bodies focused on observable behavior | Titles surface in lists; bodies are read once during planning and again during review | See Durability Rules for body content rules |
These rules apply to content sent to GitHub Issues, Azure DevOps work items, board comments, PR descriptions written from a work item, and triage artifacts. They do not replace the markdown rules in the documentation-style skill, which still govern files under docs/**/*.md.
AI Model Traceability
All generated work items must include the tag/label ai-model:<model-name>.
How to identify the model in use
- Try reading Copilot CLI logs:
grep -h "Using.*model" ~/.copilot/logs/*.log 2>/dev/null | tail -1
- If not found, check config:
cat ~/.copilot/config.json 2>/dev/null | grep -i model
- If unable to identify, ask the user
- Last resort: use
ai-model:unknown
Required Tags/Labels (every work item)
Auto-creation of Labels (GitHub)
Before applying labels to a work item, check if they exist in the repository:
github/list_label(owner, repo)
If any required label does not exist, create it automatically via github/label_write:
| Label | Color (hex) |
|---|
copilot-generated | 1d76db |
ai-model:* | 5319e7 |
type:user-story | 0e8a16 |
type:task | 0e8a16 |
type:adr | 0e8a16 |
type:tech-debt | d93f0b |
feature:* | c5def5 |
priority:p1 | b60205 |
priority:p2 | fbca04 |
priority:p3 | 0e8a16 |
phase:* | bfdadc |
severity:high | b60205 |
severity:medium | fbca04 |
severity:low | 0e8a16 |
epic | 3e4b9e |
feature | c5def5 |
parallel | d4c5f9 |
blocked | b60205 |
copilot-candidate | 1d76db |
needs-human | e4e669 |
scope:cross-cutting | bfdadc |
scope:feature-scoped | bfdadc |
Cache the list of existing labels to avoid repeated calls during the same creation session.
| Tag/Label | Required | Description |
|---|
copilot-generated | Always | Identifies item created by AI |
ai-model:<name> | Always | Model used |
Tags/Labels by Type
User Stories
| Tag/Label | Required |
|---|
type:user-story (GitHub) | Yes |
feature:<name> | Yes |
priority:<p1|p2|p3> | Yes |
Tasks
| Tag/Label | Required |
|---|
type:task (GitHub) | Yes |
feature:<name> | Yes |
phase:<phase> | Yes |
Epics
| Tag/Label | Required |
|---|
epic (GitHub) | Yes |
Features
| Tag/Label | Required |
|---|
feature (GitHub) | Yes |
priority:<p1|p2|p3> | Yes |
Missing ADRs
| Tag/Label | Required |
|---|
type:adr | Yes |
scope:cross-cutting or scope:feature-scoped | Yes |
feature:<name> | Yes (if feature-scoped) |
Tech Debt
| Tag/Label | Required |
|---|
type:tech-debt | Yes |
severity:<high|medium|low> | Yes |
feature:<name> | No (optional) |
Optional Labels
| Label | Usage |
|---|
parallel | Parallelizable task |
blocked | Task with unresolved dependencies |
copilot-candidate | Task delegable for autonomous agent execution |
needs-human | Task requiring human judgment, access, or approval |
Title
- User Stories and Tasks:
[<feature>] <description>
- Epics:
[Epic] <name>
- Features:
[Feature] <name>
- ADRs:
[ADR] <domain> or [<feature>][ADR] <domain>
- Tech Debt:
[Tech Debt] <description>
Hierarchy
- Tasks are children of User Stories
- User Stories are children of Features
- Features are children of Epics
- Dependencies between tasks documented in the body or via links
Creation Order
- User Stories first
- Tasks second
- Parent-child links third
- Dependencies between tasks fourth
NEVER create duplicate work items. If a similar item already exists, skip it and record it in the report.
Durability Rules
Work items live longer than the code they reference. File paths, line numbers, and commit SHAs rot as the codebase evolves. Write descriptions that remain useful after major refactors.
| Do | Do not |
|---|
| Describe behavior and symptoms in domain language | Reference file paths or line numbers |
| "When a user submits a form with an expired session, the system silently drops the submission" | "In src/handlers/form.ts:247, the catch block swallows the SessionExpiredError" |
| Reference domain concepts and user-visible outcomes | Reference internal module names, class names, or variable names |
| Link implementation context in PRs and commits | Embed implementation details in the work item body |
Implementation-specific context (file paths, code snippets, stack traces) belongs in linked PRs, commits, and comments, not in the work item description.
Autonomy Classification
Classify each task to indicate whether it can be completed by an AI agent autonomously or requires human judgment.
Agent-autonomous (copilot-candidate)
Add copilot-candidate to tasks that meet ALL criteria:
- Low impact (no schema change, public API, or integration)
- Well-defined scope (specific files, clear behavior)
- Established pattern (follows existing code or ADR)
- No pending architectural decisions
Human-required (needs-human)
Add needs-human to tasks that meet ANY criteria:
- Requires architectural or design judgment not covered by an ADR
- Involves external system access, manual testing, or environment-specific verification
- Requires stakeholder input or approval (UX decisions, business rules, compliance)
- Touches security-sensitive code without an established pattern
- Cross-team coordination needed
Format by Platform
Read .memory/board-config.md to identify the platform, then read only the corresponding reference:
| Platform | Reference |
|---|
| GitHub Issues | Read references/github-format.md |
| Azure DevOps | Read references/azdo-format.md |
