Context Optimize Skill (oh-my-product)

Optimize prompt and context files for maximum signal density.

Use when

• GEMINI.md or context files have grown large and unfocused
• Session context feels bloated with stale information
• Token budget is tight and every token must count
• After a major feature addition that expanded context

Workflow

1. Audit: Measure current context size and composition
2. Score: Rate each section for signal density (useful / total tokens)
3. Identify issues: Flag duplicates, stale entries, verbose sections
4. Propose changes: Show diff of recommended optimizations
5. Apply: With user approval, apply the optimizations

Optimization Strategies

Strategy	Description	Risk
Deduplicate	Remove repeated instructions across layers	Low
Compress	Shorten verbose sections preserving meaning	Low
Prioritize	Move high-signal content earlier in context	Low
Prune	Remove stale or irrelevant entries	Medium
Restructure	Improve formatting for model parsing	Low

Metrics

Before: 4,200 tokens | Signal: 62% | Redundancy: 18% | Stale: 12%
After:  2,800 tokens | Signal: 89% | Redundancy: 3%  | Stale: 0%
Savings: 33% token reduction

5-Layer Context Model

oh-my-product contexts follow this priority hierarchy:

1. System/Runtime: Gemini CLI constraints (immutable)
2. Project Standards: GEMINI.md + context/omp-core.md
3. Session Memory: .omp/state/, memory entries
4. Active Task: Current plan, taskboard, PRD
5. Execution Traces: Recent iteration results

Optimization targets layers 2-5 (layer 1 is immutable).

Safety Rules

• Never remove user-authored content without confirmation
• Always show diff before applying changes
• Preserve semantic meaning — compress, don't delete meaning
• Back up original before modifying

Related commands

• /omp:optimize — quick context optimization command
• /omp:memory compact — compact memory entries specifically