{
  "schema_version": "newruntime-topic-hub-v0.1",
  "type": "topic_hub",
  "slug": "context-engineering",
  "title": "Context engineering - New Runtime",
  "description": "How teams package memory, rules, traces, tools, and task state so agents can work without restarting from raw chat context.",
  "answer": [
    "Context engineering is the work of turning scattered organizational knowledge into durable inputs an agent can inspect and act on.",
    "The strongest systems separate reusable rules, task state, source evidence, and operator corrections instead of burying everything in one prompt.",
    "The practical payoff is lower context waste and more reproducible agent behavior across long-running tasks."
  ],
  "search_intents": [
    "context engineering",
    "agent context architecture",
    "LLM context systems"
  ],
  "status": "featured",
  "last_updated": "2026-07-21",
  "counts": {
    "total": 16,
    "signals": 9,
    "patterns": 1,
    "posts": 3,
    "atlas": 3,
    "sources": 27
  },
  "routes": {
    "html": "https://newruntime.com/topics/context-engineering/",
    "markdown": "https://newruntime.com/topics/context-engineering.md",
    "json": "https://newruntime.com/topics/context-engineering.json"
  },
  "top_sources": [
    "https://aclanthology.org/2024.acl-long.747",
    "https://aihero.dev/how-to-kill-the-bloat-in-claude-codes-system-prompt",
    "https://arxiv.org/abs/2410.10813",
    "https://arxiv.org/abs/2602.12670",
    "https://arxiv.org/abs/2603.29919",
    "https://arxiv.org/abs/2605.24050",
    "https://claude.com/blog/steering-claude-code-skills-hooks-rules-subagents-and-more",
    "https://code.claude.com/docs/en/sub-agents",
    "https://dropbox.tech/security/dropbox-mcp-dash-design-code-security",
    "https://engineering.atspotify.com/2026/6/encoding-your-domain-expert-the-context-layer-behind-spotifys-data-assistant",
    "https://engineering.grab.com/scaling-out-distroless-adoption-with-ai",
    "https://github.com/AI-Builder-Club/skills"
  ],
  "patterns": [
    {
      "kind": "Pattern",
      "slug": "skills-become-portable-capability-layer",
      "title": "Skills become a portable capability layer",
      "description": "Agent skills are emerging as a portable capability layer, but their value depends on progressive disclosure, provenance, security review, and behavioral evaluation.",
      "date": "2026-07-16",
      "topics": [
        "agent-skills",
        "evals",
        "context-engineering"
      ],
      "source_count": 4,
      "metric": "medium",
      "url": "https://newruntime.com/patterns/skills-become-portable-capability-layer/"
    }
  ],
  "field_notes": [
    {
      "kind": "Field Note",
      "slug": "ai-native-engineering-team",
      "title": "AI-Native Engineering Teams Are Smaller and Stricter",
      "description": "AI-native engineering changes team size, roles, onboarding, metrics, and review: code is no longer written only by humans, but context and quality ownership get stricter.",
      "date": "2026-07-21",
      "topics": [
        "ai-native-teams",
        "engineering-management",
        "context-engineering"
      ],
      "source_count": 1,
      "url": "https://newruntime.com/posts/ai-native-engineering-team/"
    },
    {
      "kind": "Field Note",
      "slug": "claude-code-subagents-memory",
      "title": "Claude Code Subagents Split Memory Instead of Inflating One Session",
      "description": "Claude Code subagents show a practical form of agent memory hygiene: move noisy work into a separate context and return only the compressed result to the main session.",
      "date": "2026-07-21",
      "topics": [
        "claude-code",
        "subagents",
        "context-engineering"
      ],
      "source_count": 1,
      "url": "https://newruntime.com/posts/claude-code-subagents-memory/"
    },
    {
      "kind": "Field Note",
      "slug": "unabyss-shared-context-layer",
      "title": "Unabyss Builds a Shared Context Layer Across Claude, Codex, and Cursor",
      "description": "Unabyss points to where agent tooling is going: context stops being one client's internal memory and becomes a separate product layer with governance and portability.",
      "date": "2026-07-21",
      "topics": [
        "context-engineering",
        "mcp",
        "agent-tools"
      ],
      "source_count": 2,
      "url": "https://newruntime.com/posts/unabyss-shared-context-layer/"
    }
  ],
  "raw_signals": [
    {
      "kind": "Raw Signal",
      "slug": "agent-skills-need-behavioral-evals",
      "title": "Agent skills need behavioral evals, not prose review",
      "description": "Benchmarks show that expert-authored skills can help while self-generated skills can underperform a no-skill baseline.",
      "date": "2026-07-16",
      "topics": [
        "agent-skills",
        "evals",
        "context-engineering"
      ],
      "source_count": 5,
      "metric": "structural",
      "url": "https://newruntime.com/signals/agent-skills-need-behavioral-evals/"
    },
    {
      "kind": "Raw Signal",
      "slug": "retrieval-does-not-create-company-memory",
      "title": "Retrieval and long context do not create company memory",
      "description": "Finding a relevant fragment does not establish whether it is current, authorized, trustworthy, or applicable to the present task.",
      "date": "2026-07-15",
      "topics": [
        "company-memory",
        "retrieval",
        "context-engineering"
      ],
      "source_count": 2,
      "metric": "structural",
      "url": "https://newruntime.com/signals/retrieval-does-not-create-company-memory/"
    },
    {
      "kind": "Raw Signal",
      "slug": "claude-code-startup-context-can-be-reduced-to-13k-tokens",
      "title": "Claude Code Startup Context Can Be Reduced to 13K Tokens",
      "description": "A measured context audit shows that tool descriptions and default instructions consume a material share of the initial window before repository work begins.",
      "date": "2026-07-09",
      "topics": [
        "claude-code",
        "context-engineering",
        "token-economics"
      ],
      "source_count": 2,
      "metric": "notable",
      "url": "https://newruntime.com/signals/claude-code-startup-context-can-be-reduced-to-13k-tokens/"
    },
    {
      "kind": "Raw Signal",
      "slug": "codebase-memory-combines-search-with-a-repository-map",
      "title": "Codebase Memory Combines Search with a Repository Map",
      "description": "A code retrieval tool adds structural repository context to text search so an agent can locate relevant code without loading the whole project into context.",
      "date": "2026-07-03",
      "topics": [
        "code-search",
        "agent-memory",
        "context-engineering"
      ],
      "source_count": 2,
      "metric": "notable",
      "url": "https://newruntime.com/signals/codebase-memory-combines-search-with-a-repository-map/"
    },
    {
      "kind": "Raw Signal",
      "slug": "pydantic-ai-capabilities-progressive-disclosure",
      "title": "Pydantic AI loads capabilities only when needed",
      "description": "A capability can bundle instructions, tools, model settings, and hooks while exposing only a compact description until activation.",
      "date": "2026-06-28",
      "topics": [
        "agent-skills",
        "context-engineering",
        "tool-use"
      ],
      "source_count": 2,
      "metric": "notable",
      "url": "https://newruntime.com/signals/pydantic-ai-capabilities-progressive-disclosure/"
    },
    {
      "kind": "Raw Signal",
      "slug": "claude-md-shrinks-as-skills-hooks-and-subagents-mature",
      "title": "CLAUDE.md Shrinks as Skills, Hooks, and Subagents Mature",
      "description": "Agent behavior moves out of one giant context file into scoped mechanisms that load only for the relevant task and lifecycle stage.",
      "date": "2026-06-23",
      "topics": [
        "claude-code",
        "skills",
        "context-engineering"
      ],
      "source_count": 1,
      "metric": "notable",
      "url": "https://newruntime.com/signals/claude-md-shrinks-as-skills-hooks-and-subagents-mature/"
    },
    {
      "kind": "Raw Signal",
      "slug": "production-ai-engineering-is-the-system-around-the-model",
      "title": "Production AI Engineering Is the System Around the Model",
      "description": "Examples from large engineering teams point to testing, context layers, security, and workflow integration as the durable production work.",
      "date": "2026-06-16",
      "topics": [
        "production-ai",
        "agent-evals",
        "context-engineering"
      ],
      "source_count": 5,
      "metric": "notable",
      "url": "https://newruntime.com/signals/production-ai-engineering-is-the-system-around-the-model/"
    },
    {
      "kind": "Raw Signal",
      "slug": "codex-subagents-reduce-context-pollution",
      "title": "Codex Subagents Reduce Context Pollution",
      "description": "Delegating bounded work to subagents preserves the parent agent's attention and creates clearer evidence boundaries than loading every exploration step into one conversation.",
      "date": "2026-05-24",
      "topics": [
        "codex",
        "subagents",
        "context-engineering"
      ],
      "source_count": 1,
      "metric": "notable",
      "url": "https://newruntime.com/signals/codex-subagents-reduce-context-pollution/"
    },
    {
      "kind": "Raw Signal",
      "slug": "skill-md-acts-as-a-behavior-loader-not-a-better-prompt",
      "title": "SKILL.md Acts as a Behavior Loader, Not a Better Prompt",
      "description": "Skills package procedures, tools, and progressive context so an agent can load behavior only when the task requires it.",
      "date": "2026-05-07",
      "topics": [
        "skills",
        "context-engineering",
        "agent-harness"
      ],
      "source_count": 3,
      "metric": "notable",
      "url": "https://newruntime.com/signals/skill-md-acts-as-a-behavior-loader-not-a-better-prompt/"
    }
  ],
  "atlas_records": [
    {
      "kind": "Shift",
      "slug": "shared-context-infrastructure",
      "title": "Client-local memory -> shared context infrastructure",
      "description": "Context is shifting from private memory inside one AI client into a governed layer between agents, IDEs, documents, and tools.",
      "date": "2026-07-21",
      "topics": [
        "context-engineering",
        "agent-memory",
        "mcp"
      ],
      "source_count": 2,
      "metric": "emerging",
      "url": "https://newruntime.com/shifts/shared-context-infrastructure/"
    },
    {
      "kind": "Graveyard",
      "slug": "manual-context-handoff",
      "title": "Manual context handoff between AI clients",
      "description": "The ritual of retelling one AI client what another already understood becomes optional when context moves into a shared layer.",
      "date": "2026-07-21",
      "topics": [
        "context-engineering",
        "agent-memory"
      ],
      "source_count": 1,
      "metric": "challenged",
      "url": "https://newruntime.com/graveyard/manual-context-handoff/"
    },
    {
      "kind": "Delta",
      "slug": "shared-context-portability",
      "title": "Shared context lets AI clients stop starting from zero",
      "description": "A separate context layer lets multiple AI clients read shared project context without a human manually copying memory.",
      "date": "2026-07-21",
      "topics": [
        "context-engineering",
        "mcp"
      ],
      "source_count": 2,
      "metric": "B1",
      "url": "https://newruntime.com/deltas/shared-context-portability/"
    }
  ]
}