A Claude Code plugin that helps you learn context engineering patterns from Claude Code and apply them to your own agent projects.
Claude Code is a production-grade AI coding agent with sophisticated context engineering — how it structures system prompts, manages tools, handles context windows, and orchestrates sub-agents. This plugin captures those patterns from real API traces and helps you migrate them to your own agent codebase.
| Skill | Purpose |
|---|---|
| cc-trace | Capture Claude Code's real API requests using claude-trace and store raw trace data for analysis. |
| cc-learn | Extract patterns and raw examples from traces, compare with your project code through collaborative dialogue, and produce a detailed migration plan (docs/cc-context/plans/YYYY-MM-DD-migration-plan.md). |
| cc-apply | Execute the migration plan step by step, modifying your project code with confirmation at each step. |
| cc-verify | Capture your project's runtime API traces and verify that migrated patterns are actually working — not just present in code. |
cc-trace → cc-learn → cc-apply → cc-verify
(capture) (analyze (execute) (verify)
+ plan) ↑ |
└────────────┘
(iterate on gaps)
Each skill works independently. Use them together for a full learning loop, or individually as needed:
- Just curious? Run
cc-tracealone to see what Claude Code sends to the API. - Have existing knowledge? Skip to
cc-learnwith a knowledge base from a previous session to plan your migration. - Already migrated? Run
cc-verifyto confirm runtime behavior matches.
# Add the marketplace
/plugin marketplace add claude89757/claude-code-context-skills
# Install the plugin
/plugin install cc-context-skills@cc-context-skills# Add a locally cloned copy as marketplace
/plugin marketplace add /path/to/cc-context-skills
# Install the plugin
/plugin install cc-context-skills@cc-context-skills- Node.js 16+
- jq
- claude-trace (auto-installed by cc-trace)
# Run the cc-trace skill
/cc-traceThis captures a real API request from the latest Claude Code version and saves raw trace data to docs/cc-context/traces/.
/cc-learnExtracts patterns and raw reference examples from traces, compares with your project code through collaborative dialogue, and generates a migration plan (docs/cc-context/plans/YYYY-MM-DD-migration-plan.md) with:
- Alignment score
- Prioritized gaps (HIGH / MED / LOW)
- Concrete migration steps with file references
/cc-applyReads the migration plan and executes changes step by step, with confirmation at each step.
/cc-verifyCaptures your project's API traces and checks whether patterns are actually present in runtime behavior.
From a single Claude Code trace, you can discover patterns like:
- System prompt architecture — 3-block structure with selective cache control
- Stable-first content ordering — maximizes prompt cache hit rates
- System-reminder injection — dynamic context in user messages, not system prompt
- Rich tool descriptions — behavioral guides embedded in tool definitions
- Adaptive thinking — effort level adapts to task complexity
- Lazy context management — activated only when context window fills up
.claude-plugin/
plugin.json # Plugin metadata
skills/
cc-trace/ # Trace capture & raw data storage
SKILL.md
scripts/ # Shell scripts for capture & analysis
references/ # Troubleshooting & version analysis guides
cc-learn/ # Pattern extraction + example curation + migration planning
SKILL.md
references/ # Pattern taxonomy
cc-apply/ # Migration plan execution
SKILL.md
references/ # API migration strategies
cc-verify/ # Runtime verification
SKILL.md
references/ # Verification criteria
data-contracts.md # Inter-skill data paths & format conventions
MIT