context-management-for-antigravity

A comprehensive collection of Agent Skills for context engineering, harness engineering, multi-agent architectures, and production agent systems. Use when building, optimizing, evaluating, or debugging agent systems that require effective context management and reliable operating loops.

This skill should be used for LLM-as-a-Judge evaluation techniques — direct scoring, pairwise comparison, rubric generation, reference-based grading, bias mitigation, calibration, and evaluator reliability measurement. Route deterministic evaluation checks and pipeline design to evaluation. Route agent harness design to harness-engineering.

This skill should be used for modeling agent cognitive states using formal Belief-Desire-Intention (BDI) ontology — transforming external RDF context into explicit mental states, designing deliberative reasoning pipelines, enabling explainable agent decisions, and implementing plan libraries. Route multi-agent coordination to multi-agent-patterns, memory architecture to memory-systems, and evaluation of agent reasoning to evaluation.

This skill should be used for designing and evaluating compression strategies for long-running agent sessions — hierarchical summarization, selective retention, compaction triggers, quality thresholds, and handoff summaries. Route foundational context concepts to context-fundamentals, failure diagnosis to context-degradation, and token-level efficiency tactics to context-optimization.

This skill should be used for diagnosing and mitigating context degradation — lost-in-middle failures, context poisoning, context clash, context confusion, attention-pattern issues, and agent performance degradation caused by accumulated or conflicting context. Route foundational conceptual work to context-fundamentals, token-efficiency tactics to context-optimization, and compression strategy design to context-compression.

This skill should be used to explain or reason about the foundational concepts of context engineering — what context is, the anatomy of a context window, how attention mechanics work, the U-shaped attention curve, why context quality matters more than quantity, and the mental models needed to interpret every other context-engineering decision. Use this for conceptual explanation, onboarding, and background reading. Route operational work to the specialized skills — debugging attention failures goes to context-degradation, token-efficiency work goes to context-optimization, conversation summarization goes to context-compression, and project-shape decisions go to project-development.

This skill should be used for token-level efficiency tactics — observation masking, prefix caching, context partitioning, budget allocation, retrieval precision, format optimization, and trajectory-level token reduction. Route foundational concepts to context-fundamentals, failure diagnosis to context-degradation, compression strategies to context-compression, and filesystem offloading to filesystem-context.

This skill should be used for building deterministic evaluation frameworks for agent systems — assertion-based checks, regression testing, pass/fail criteria, metric design, and structured evaluation pipelines. Route LLM-as-judge techniques, rubric generation, and evaluator bias mitigation to advanced-evaluation. Route agent harness design to harness-engineering.

This skill should be used for filesystem-based context management — scratch pads, plan persistence, tool output offloading, sub-agent communication via shared files, dynamic skill loading, and just-in-time context discovery using standard file operations. Route memory architecture decisions to memory-systems, compression strategies to context-compression, and hosted runtime infrastructure to hosted-agents.

This skill should be used for designing autonomous agent harnesses — the operating loop that manages agent execution with locked evaluation metrics, durable logs, novelty gates, rollback mechanisms, human approval boundaries, and safety constraints. Route evaluation framework design to evaluation, LLM-as-judge techniques to advanced-evaluation, and multi-agent coordination to multi-agent-patterns.

This skill should be used when building background coding agents that run in remote sandboxed environments — pre-built images, warm sandbox pools, filesystem snapshots, multiplayer support, multi-client interfaces, and self-spawning patterns. Route multi-agent coordination patterns to multi-agent-patterns, individual tool design to tool-design, and filesystem offloading patterns to filesystem-context.

This skill should be used when sharing orchestrator trajectory state with workers via task-guided KV cache compaction — applicable when the worker runtime is controllable and the models share compatible architectures. Route multi-agent coordination patterns to multi-agent-patterns, general context optimization to context-optimization, and compression strategies to context-compression.

This skill should be used when designing agent memory architectures — short-term scratchpads, long-term persistence, entity tracking, vector RAG, knowledge graphs, and the file-system-as-memory pattern. Route filesystem-specific offloading patterns to filesystem-context, cross-session handoff summaries to context-compression, and multi-agent state sharing to multi-agent-patterns.

This skill should be used when designing multi-agent systems that need context isolation, supervisor or swarm coordination, explicit handoffs, parallel execution, or a decision on whether multiple agents are justified. Route project-level pipeline decisions to project-development, hosted sandbox infrastructure to hosted-agents, KV-cache state sharing to latent-briefing, and individual tool design to tool-design.

This skill should be used for LLM project lifecycle decisions — task-model fit analysis, pipeline architecture design, structured output schemas, batch processing strategies, cost estimation, and deployment planning. Route individual tool design to tool-design, multi-agent topology decisions to multi-agent-patterns, and evaluation framework design to evaluation.

This skill should be used for the tool-interface layer of an agent system — writing tool descriptions agents can route on, designing tool schemas and response formats, naming conventions, actionable error recovery messages, MCP server design, tool-set consolidation, and deciding when to add or remove an individual tool. Route project-shape and pipeline architecture decisions to project-development; route deciding whether to introduce sub-agents to multi-agent-patterns.

Template for creating new Agent Skills for context engineering. Use this template when adding new skills to the collection.