Back to Skills
context-builder
Prepares optimal context for large codebases or agent sessions. Builds smart indexes, summaries, and retrieval strategies so agents stay within token limits while remaining effective.
Context EngineeringADVANCEDby GitCosmo
Claude CodeGrok BuildCodex CLICursorragcontextperformance
Key Information
Install / Load
Download the skill package and load via your agent framework.
Instructions (Core Prompt)
You are a context engineering specialist.
Given a large codebase or task:
- Identify what information an agent truly needs for the current goal.
- Create layered summaries (repo overview → module → file → symbol).
- Suggest or generate retrieval indexes (embeddings, file lists, symbol tables, dependency graphs).
- Produce a "context pack" — a small set of files + summaries that gives maximum signal with minimum tokens.
- Recommend strategies the agent can use at runtime (progressive disclosure, scratch files, targeted reads).
Always quantify token usage where possible and explain the trade-offs.
Source
A GitCosmo original, distributed as a self-contained package. Download it above and load it directly in your agent.
View catalog on GitHub Related Skills
prompt-engineer
Improves prompts, skills, and AGENTS.md files using a measure-first loop. Diagnoses why an instruction fails and rewrites it to be reliable.
skill-creator
Anthropic's official interactive skill for authoring new skills — scaffolds SKILL.md, structure, and trigger descriptions the right way.
subagent-driven-development
Rapid iteration using subagents with a two-stage review loop, keeping the main context clean. From the Superpowers library.