{
  "schema_version": "1.1",
  "id": "s9:https://github.com/mksglu/context-mode",
  "slug": "context-mode-1olsb2y",
  "url": "https://feed7.dev/p/context-mode-1olsb2y",
  "title": "mksglu/context-mode",
  "why_included": "Context Mode keeps bulky tool output outside an agent’s prompt and restores relevant session state after compaction. Its MCP tools and hooks support 17 clients, though enforcement varies by platform.",
  "summary": "Context Mode reports shrinking **315 KB to 5.4 KB** by running tool work in a sandbox and returning only selected results. It records edits, Git operations, tasks, errors, and decisions in SQLite, then uses **FTS5 with BM25** to retrieve relevant state after compaction.",
  "practical_implication": "For long coding-agent sessions, route large reads, searches, and fetches through its execution and indexing tools, then run its doctor checks. Hook-capable clients can enforce routing automatically; MCP-only installs expose **11 tools** but rely on the model to choose them.",
  "agent_context": "Context Mode reports shrinking **315 KB to 5.4 KB** by running tool work in a sandbox and returning only selected results. It records edits, Git operations, tasks, errors, and decisions in SQLite, then uses **FTS5 with BM25** to retrieve relevant state after compaction.\n\nFor long coding-agent sessions, route large reads, searches, and fetches through its execution and indexing tools, then run its doctor checks. Hook-capable clients can enforce routing automatically; MCP-only installs expose **11 tools** but rely on the model to choose them.\n\nThe **98% reduction** is a project-reported example, not an independent benchmark. Installation and hook coverage differ across the 17 supported clients, and Cursor’s marketplace plugin was still awaiting review in the supplied material.",
  "source": {
    "name": "GitHub",
    "url": "https://github.com/mksglu/context-mode",
    "published_at": null
  },
  "source_class": "tool",
  "content_type": "GitHub Repo",
  "layer": "context",
  "domains": [
    "coding"
  ],
  "topics": [
    "context-engineering",
    "context-caching",
    "mcp"
  ],
  "verification": {
    "status": "needs_review",
    "label": "Needs Review",
    "method": "unverified",
    "verified_at": null
  },
  "uncertainty": [
    "The **98% reduction** is a project-reported example, not an independent benchmark. Installation and hook coverage differ across the 17 supported clients, and Cursor’s marketplace plugin was still awaiting review in the supplied material."
  ],
  "connected_context": {
    "meaning": "This combines context reduction with durable, searchable session state: bulky tool work stays in a sandbox while edits, errors, and decisions remain retrievable after compaction. Relative to the candidates, it joins compression and retrieval in one coding-agent layer, but its reported reduction is workload-specific and MCP-only clients still depend on voluntary tool routing.",
    "corpus_size": 703,
    "generated_at": "2026-09-08T10:04:31.773Z",
    "connections": [
      {
        "title": "Blast Radius",
        "source_name": "arXiv",
        "source_url": "https://arxiv.org/abs/2608.07440v1",
        "feed7_url": "https://feed7.dev/p/2608-07440v1-1gi5t78",
        "reason": "Blast Radius reversibly removes predicted-unused history, whereas Context Mode externalizes bulky work and indexes selected state; together they frame reduction and later recovery as separate policies that need long-horizon testing."
      },
      {
        "title": "JuliusBrussee/caveman",
        "source_name": "GitHub",
        "source_url": "https://github.com/JuliusBrussee/caveman",
        "feed7_url": "https://feed7.dev/p/caveman-0yoqowc",
        "reason": "Both target token reduction, but Caveman warns that recurring compression instructions can outweigh savings; that makes Context Mode’s headline reduction subject to session-level cost and exactness measurements."
      },
      {
        "title": "tirth8205/code-review-graph",
        "source_name": "GitHub",
        "source_url": "https://github.com/tirth8205/code-review-graph",
        "feed7_url": "https://feed7.dev/p/code-review-graph-0ag5r2t",
        "reason": "code-review-graph offers dependency-shaped repository retrieval, while Context Mode preserves broader operational history such as edits and decisions; the graph could narrow what code is fetched from the durable session record."
      },
      {
        "title": "Building Agents Is Trivial Now, Context Is the Next Frontier — Jeff Ng, Unblocked",
        "source_name": "AI Engineer",
        "source_url": "https://www.youtube.com/watch?v=HvMyYLTfvhg",
        "feed7_url": "https://feed7.dev/p/building-agents-is-trivial-now-context-is-the-next-frontier-jeff-ng-unbl-073pk7b",
        "reason": "Unblocked shows that repository evidence alone can miss decisions and incidents; Context Mode’s indexed decision and error history partly addresses that gap, though it does not itself reconcile conflicts or organizational sources."
      }
    ]
  },
  "lifecycle": "Current",
  "published_at": null,
  "modified_at": null,
  "supersedes": [],
  "expires_at": null,
  "formats": {
    "html": "https://feed7.dev/p/context-mode-1olsb2y",
    "json": "https://feed7.dev/p/context-mode-1olsb2y.json",
    "markdown": "https://feed7.dev/p/context-mode-1olsb2y.md"
  }
}