hiromaily Workflow Triage
Analyze a request and determine whether it warrants a GitHub issue only or a full Spec-Driven Development (SDD) workflow (with or without a preceding issue).
When to Use
Invoke this skill before starting any significant piece of work to select the right process and avoid over- or under-engineering the workflow.
/triage-workflow "<description of the request>"
Scoring Rubric
Evaluate the request on five dimensions. Score each 1–3.
| Dimension | 1 — Low | 2 — Medium | 3 — High |
|---|---|---|---|
| Scope | Single file or function | Multiple files, one subsystem | Cross-layer or multiple subsystems |
| Complexity | Solution path is obvious | Some design decisions required | Requires architectural design or research |
| Architectural impact | No interface/layer changes | New interface or minor restructure | Changes contracts, layer boundaries, or data models |
| Risk | Isolated, easily reversible | Moderate risk to adjacent code | Could break existing features or APIs |
| Ambiguity | Requirements are clear | Some clarification needed | Requirements are incomplete or contested |
Total score range: 5–15
Decision Table
| Total Score | Recommendation |
|---|---|
| 5–7 | GitHub Issue only — implement directly |
| 8–11 | GitHub Issue + SDD Spec — create an issue to track, then run the spec workflow |
| 12–15 | Full SDD Spec — spec-driven design first; issue is created as part of spec output |
Execution Steps
Step 1 — Read the request
If $ARGUMENTS is provided, use that as the request description. Otherwise, ask the user to describe the request in 1–3 sentences.
Step 2 — Score each dimension
Think through each of the five dimensions carefully. Write a 1-sentence rationale for each score. Sum the scores.
Step 3 — Output the recommendation
Follow the format below. Always include:
- The score breakdown table
- The recommendation label
- Concrete next steps (commands or actions)
Step 4 — Offer to proceed
Ask the user: "Shall I proceed with the recommended workflow?" If yes, execute the first step of that workflow automatically.
Output Format
## Workflow Triage
**Request**: {brief restatement of the request}
### Dimension Scores
| Dimension | Score | Rationale |
|-----------|-------|-----------|
| Scope | {1–3} | {one sentence} |
| Complexity | {1–3} | {one sentence} |
| Architectural impact | {1–3} | {one sentence} |
| Risk | {1–3} | {one sentence} |
| Ambiguity | {1–3} | {one sentence} |
| **Total** | **{5–15}** | |
### Recommendation: {GitHub Issue only | GitHub Issue + SDD Spec | Full SDD Spec}
{2–3 sentence explanation of why this workflow fits.}
### Next Steps
{Bullet list of concrete actions, e.g.:}
- Create GitHub issue: `gh issue create --title "..." --body "..."`
- Run spec init: `/kiro:spec-init "<description>"`
- Or: implement directly with a feature branch and PR
Examples
Example A — Minor fix (score 5–7)
"Fix the typo in the error message when config file is missing"
- Scope: 1 (single string in one file)
- Complexity: 1 (trivial change)
- Architectural impact: 1 (none)
- Risk: 1 (no logic change)
- Ambiguity: 1 (clear)
- Total: 5 → GitHub Issue only
Example B — Medium feature (score 8–11)
"Add a
--dry-runflag to the CLI that skips side effects but logs what would happen"
- Scope: 2 (CLI parsing + service layer)
- Complexity: 2 (needs consistent propagation through services)
- Architectural impact: 2 (new port parameter or flag threading)
- Risk: 2 (touches execution path)
- Ambiguity: 2 (behaviour in edge cases needs definition)
- Total: 10 → GitHub Issue + SDD Spec
Example C — Major redesign (score 12–15)
"Redesign the memory system to support multiple backends with a pluggable adapter interface"
- Scope: 3 (domain + application + infra + tests)
- Complexity: 3 (multiple design options, trade-offs to evaluate)
- Architectural impact: 3 (new port interface, adapter contracts)
- Risk: 3 (existing memory behaviour must be preserved)
- Ambiguity: 2 (goal is clear, implementation path is not)
- Total: 14 → Full SDD Spec
Design Principles
- Bias toward lightweight: When on the boundary, prefer the lighter-weight workflow.
- One recommendation: Output a single clear recommendation, not a range.
- Action-oriented: Always end with runnable commands or instructions.
- No over-engineering: A GitHub issue is a valid and often correct answer.