Triage

Invoked as /agiflow:triage. In hosts without slash-prompts, this skill is triggered by matching intent and drives AgiFlow via its MCP tools.

Usage:

• /agiflow:triage - Diagnose project health and surface actionable problems
• /agiflow:triage <project-slug-or-id> - Triage a specific project

Examples:

• /agiflow:triage (triage current project or org)
• /agiflow:triage my-project (triage specific project)

Purpose Diagnose stalled, blocked, or unhealthy project states and recommend concrete actions. Also handles task lifecycle management: cancelling/archiving obsolete tasks, re-prioritizing (moving Todo tasks back to Planning), and fast-tracking incoming bugs. Real projects get stuck — this skill helps users identify WHY and decide WHAT TO DO, not just see status.

Guardrails

• Diagnose before prescribing — understand the problem before suggesting solutions.
• Surface problems by severity: blockers first, then warnings, then notes.
• Recommend specific actions (reassign, split, refine, cancel, re-prioritize) — not vague advice.
• Do not make changes automatically — present findings and let the user decide.
• When cancelling tasks, document the reason in a comment before changing status.

AgiFlow Project Management Guidelines

Follow the shared AgiFlow project-management guidelines in references/agiflow-agents.md — agent assignment, the task status workflow and transitions, work-unit best practices, and the tags strategy apply to this workflow.

Steps

1. Load Project State

1. Use list_active_tasks_by_org MCP tool to get all active (non-completed) tasks across the organization.
2. Use list_work_units MCP tool to get work unit status overview.
3. Use list_members MCP tool to understand team capacity.
4. If a specific project is provided, use get_project_detail to focus on that project.

2. Identify Problems

5. Analyze the data and flag issues in these categories:

Stalled Work

• Tasks in "In Progress" with no recent devInfo updates (stalled execution)
• Work units in "in_progress" where no tasks have moved in a long time
• Tasks assigned to an agent but showing no activity

For each stalled item:

• When was it last updated?
• Is there a blocker documented in comments?
• Is the task too large (needs decomposition)?
• Is the assignee available?

Blocked Items

• Work units with status "blocked"
• Tasks with blocker comments but no follow-up
• Tasks that depend on other incomplete tasks

For each blocked item:

• What's the blocker? (check comments with list_task_comments)
• Can the blocker be resolved, worked around, or descoped?
• Should the blocked task be reassigned or split?

Quality Gaps

• Tasks with no acceptance criteria (unrefined)
• Tasks with vague descriptions ("implement feature X")
• High-priority tasks that are unassigned
• Work units with more than 8 tasks (too complex for one session)

Imbalance

• One agent assigned to too many in-progress tasks (context switching)
• All tasks "In Progress", nothing "Done" (no focus)
• High-priority items in backlog while low-priority items are being worked on

3. Severity Classification

6. Classify each finding:

4. Recommend Actions

7. For each problem, recommend ONE specific action:

5. Present Triage Report

8. Present findings to the user:

## Triage Report

**Scope:** [Organization / Project name]
**Date:** [current date]

### Summary
- Active tasks: [N]
- In progress: [N]
- Blocked: [N]
- Issues found: [N blockers, N warnings, N notes]

### Blockers (must resolve)
1. **[Task/WU slug]** — [problem description]
   - Recommendation: [specific action]
   - Impact: [what's affected if not resolved]

### Warnings (should resolve)
1. **[Task/WU slug]** — [problem description]
   - Recommendation: [specific action]

### Notes (nice to fix)
1. **[observation]** — [suggestion]

### Health Score
[Quick assessment: Healthy / Needs Attention / At Risk]

### Recommended Next Steps
1. [Most important action]
2. [Second most important action]
3. [Third action]

6. Execute (with user approval)

9. After presenting findings, ask the user which actions they want to take.
10. Execute approved actions using the appropriate MCP tools.
11. After changes, re-check state with list_active_tasks_by_org to confirm improvements.

Diagnostic Patterns

Common Mistakes to Avoid

• Only reporting status without diagnosing root causes
• Suggesting vague actions ("improve task descriptions") instead of specific ones
• Making changes without user approval
• Ignoring blocked items (they don't unblock themselves)
• Not checking task comments for context on why things stalled
• Treating symptoms (reassigning a stalled task) without addressing causes (task was too vague)