Apply R&D framework to optimize prompts and context. Use when optimizing context window usage, reducing prompt size, delegating to specialized agents, or applying systematic context management.
View on GitHubJanuary 21, 2026
Select agents to install to:
npx add-skill https://github.com/melodic-software/claude-code-plugins/blob/main/plugins/tac/skills/reduce-delegate-framework/SKILL.md -a claude-code --skill reduce-delegate-frameworkInstallation paths:
.claude/skills/reduce-delegate-framework/# Reduce & Delegate Framework Skill Apply the R&D framework to optimize prompts, workflows, and context management. ## Purpose There are only two ways to manage context: **Reduce** and **Delegate**. This skill helps you systematically apply both strategies to any context optimization challenge. ## When to Use - Context window approaching limits - Agent performance degrading over conversation - Prompts growing unwieldy - Workflows consuming too many tokens - Need to scale agent work ## The R&D Analysis Process ### Step 1: Identify the Context Problem Categorize the issue: | Problem Type | Indicator | Primary Strategy | | --- | --- | --- | | Context Rot | Old info guiding decisions | Reduce (fresh instance) | | Context Pollution | Unfocused, tangential | Reduce (remove irrelevant) | | Toxic Context | Contradictory behavior | Reduce (clear conflicts) | | Context Overflow | Approaching limits | Delegate (offload work) | ### Step 2: Apply Reduce Strategies For each context element, ask: 1. Is this necessary for the current task? 2. Can this be loaded on-demand instead? 3. Is this information stale or outdated? 4. Does this contradict other context? Reduction techniques: | Technique | Application | | --- | --- | | Fresh instance | New task type, reset history | | Output styles | Control verbosity, reduce tokens | | Focused reads | Specific files vs directories | | Priming commands | Replace static memory | | MCP cleanup | Remove unused servers | ### Step 3: Apply Delegate Strategies For complex or parallel work, ask: 1. Does this subtask need different context? 2. Can this run independently? 3. Would a specialized agent perform better? 4. Is there parallel work opportunity? Delegation techniques: | Technique | Application | | --- | --- | | Sub-agents | Focused tasks with isolated context | | Background agents | Parallel work, async execution | | Agent experts | Domain-specific knowledge | | Spec files | Handoff between agents | ## Optimization Workflow