triage

Auto-Adaptive Priority Memory

This skill auto-evolves. Observe prioritization signals, detect patterns, confirm before internalizing.

Core Loop: 1. Assess — When tasks arrive, evaluate urgency + importance 2. Classify — Assign priority level (P0-P3) 3. Route — P0 immediately, P1-P3 into queue 4. Learn — Notice when user overrides priority → propose pattern 5. Confirm — After 2+ corrections, ask: "Should X always be P[n]?"

Check signals.md for urgency indicators. Check patterns.md for learned priority rules.

Priority Levels

Level	Response	Examples
P0	Interrupt immediately	Server down, security breach, deadline today
P1	Next available slot	Blocking work, waiting users, same-day tasks
P2	Scheduled queue	Important but not urgent, planning, reviews
P3	Backlog	Ideas, "someday", low-impact optimizations

Default: When uncertain, ask. Start conservative, learn boundaries.

Urgency Signals

Automatic P0 triggers: - Words: "urgent", "ASAP", "down", "broken", "emergency" - Context: External deadlines, blocked team members - Pattern: User previously escalated similar tasks

Automatic P3 triggers: - Words: "when you have time", "no rush", "idea for later" - Context: No deadline mentioned, exploratory

Entry Format

One line: pattern: priority (level) [context]

Examples: - deploy-issues: P0 (confirmed) [always urgent] - refactoring: P2 (pattern) [user deprioritized 3x] - docs-updates: P3 (confirmed) [explicit "low priority"]

Work Categories

Time Patterns

Source Routing

Overrides

Queue Management

When multiple tasks compete: 1. Group by priority — P0 first, always 2. Within same priority — Order by arrival or explicit sequence 3. Report queue — "3 P1 tasks queued, handling X first" 4. Re-triage on change — New P0 interrupts P1 work

Learning Triggers

Phrases that signal priority pattern: - "This should be higher priority" - "Drop everything and..." - "This can wait" - "Handle [X] before [Y]" - "Not urgent" / "No rush"

After hearing these: Update entry, wait for 2nd occurrence, then confirm permanent rule.

Empty sections = still learning. Start conservative, observe corrections, propose only after patterns emerge.