lynkr 9.14.1 → 9.14.3

package/config/model-capability-seeds.json

@@ -0,0 +1,75 @@
+{
+  "version": 1,
+  "updatedAt": "2026-09-08",
+  "notes": "Reviewed per-family capability seeds. Values are curated estimates triangulated from SWE-Bench Verified, TerminalBench, LiveCodeBench/BigCodeBench and Artificial Analysis coding index (see scripts/seed-capabilities.js to regenerate from live sources). Heads: reasoning/codegen/debugging/tool_use in [0,1]. Keys are normalized family ids (see capability-seeds/family.js) with trailing-* wildcards matched longest-first. Quantized self-hosted servings take an automatic -0.02 haircut at resolve time. Unknown families fall through to the family heuristic, then tier-slot caps.",
+  "seeds": {
+    "claude-opus-4*": { "caps": { "reasoning": 0.9, "codegen": 0.9, "debugging": 0.9, "tool_use": 0.9 }, "sources": ["swe-bench-verified", "artificial-analysis"], "note": "curated-estimate: flagship tier" },
+    "claude-opus-3*": { "caps": { "reasoning": 0.8, "codegen": 0.82, "debugging": 0.8, "tool_use": 0.78 }, "sources": ["swe-bench-verified"], "note": "curated-estimate" },
+    "claude-sonnet-4*": { "caps": { "reasoning": 0.78, "codegen": 0.8, "debugging": 0.78, "tool_use": 0.78 }, "sources": ["swe-bench-verified", "artificial-analysis"], "note": "curated-estimate" },
+    "claude-sonnet-3.5": { "caps": { "reasoning": 0.72, "codegen": 0.75, "debugging": 0.72, "tool_use": 0.7 }, "sources": ["swe-bench-verified"], "note": "curated-estimate" },
+    "claude-haiku*": { "caps": { "reasoning": 0.42, "codegen": 0.48, "debugging": 0.42, "tool_use": 0.45 }, "sources": ["swe-bench-verified"], "note": "curated-estimate" },
+    "gpt-5.6*": { "caps": { "reasoning": 0.85, "codegen": 0.85, "debugging": 0.85, "tool_use": 0.82 }, "sources": ["swe-bench-verified"], "note": "curated-estimate: flagship tier" },
+    "gpt-5.4*": { "caps": { "reasoning": 0.85, "codegen": 0.85, "debugging": 0.85, "tool_use": 0.82 }, "sources": ["swe-bench-verified"], "note": "curated-estimate: flagship tier" },
+    "gpt-5.3*": { "caps": { "reasoning": 0.82, "codegen": 0.85, "debugging": 0.82, "tool_use": 0.8 }, "sources": ["swe-bench-verified"], "note": "curated-estimate" },
+    "gpt-5.2*": { "caps": { "reasoning": 0.8, "codegen": 0.83, "debugging": 0.8, "tool_use": 0.78 }, "sources": ["swe-bench-verified"], "note": "curated-estimate" },
+    "gpt-5.1*": { "caps": { "reasoning": 0.7, "codegen": 0.72, "debugging": 0.7, "tool_use": 0.68 }, "sources": ["swe-bench-verified"], "note": "curated-estimate" },
+    "gpt-5-mini*": { "caps": { "reasoning": 0.6, "codegen": 0.65, "debugging": 0.6, "tool_use": 0.6 }, "sources": ["swe-bench-verified"], "note": "curated-estimate" },
+    "gpt-5-nano*": { "caps": { "reasoning": 0.45, "codegen": 0.5, "debugging": 0.45, "tool_use": 0.45 }, "sources": ["swe-bench-verified"], "note": "curated-estimate" },
+    "gpt-4o": { "caps": { "reasoning": 0.65, "codegen": 0.7, "debugging": 0.65, "tool_use": 0.68 }, "sources": ["swe-bench-verified", "livecodebench"], "note": "curated-estimate" },
+    "gpt-4o-mini": { "caps": { "reasoning": 0.45, "codegen": 0.5, "debugging": 0.45, "tool_use": 0.48 }, "sources": ["swe-bench-verified"], "note": "curated-estimate" },
+    "gpt-4.1": { "caps": { "reasoning": 0.68, "codegen": 0.72, "debugging": 0.68, "tool_use": 0.7 }, "sources": ["swe-bench-verified"], "note": "curated-estimate" },
+    "gpt-4.1-mini": { "caps": { "reasoning": 0.5, "codegen": 0.55, "debugging": 0.5, "tool_use": 0.52 }, "sources": ["swe-bench-verified"], "note": "curated-estimate" },
+    "o3": { "caps": { "reasoning": 0.88, "codegen": 0.8, "debugging": 0.85, "tool_use": 0.7 }, "sources": ["swe-bench-verified"], "note": "curated-estimate: reasoning tilt" },
+    "o3-mini": { "caps": { "reasoning": 0.72, "codegen": 0.68, "debugging": 0.7, "tool_use": 0.6 }, "sources": ["swe-bench-verified"], "note": "curated-estimate: reasoning tilt" },
+    "o4-mini": { "caps": { "reasoning": 0.7, "codegen": 0.66, "debugging": 0.68, "tool_use": 0.6 }, "sources": ["swe-bench-verified"], "note": "curated-estimate" },
+    "o1": { "caps": { "reasoning": 0.82, "codegen": 0.72, "debugging": 0.78, "tool_use": 0.6 }, "sources": ["swe-bench-verified"], "note": "curated-estimate: reasoning tilt" },
+    "o1-mini": { "caps": { "reasoning": 0.62, "codegen": 0.6, "debugging": 0.6, "tool_use": 0.5 }, "sources": ["swe-bench-verified"], "note": "curated-estimate" },
+    "gemini-2.5-pro": { "caps": { "reasoning": 0.85, "codegen": 0.83, "debugging": 0.85, "tool_use": 0.78 }, "sources": ["swe-bench-verified", "artificial-analysis"], "note": "curated-estimate" },
+    "gemini-3-pro*": { "caps": { "reasoning": 0.85, "codegen": 0.83, "debugging": 0.85, "tool_use": 0.78 }, "sources": ["swe-bench-verified"], "note": "curated-estimate" },
+    "gemini-2.5-flash": { "caps": { "reasoning": 0.55, "codegen": 0.6, "debugging": 0.55, "tool_use": 0.58 }, "sources": ["swe-bench-verified"], "note": "curated-estimate" },
+    "gemini-2.0-flash": { "caps": { "reasoning": 0.45, "codegen": 0.5, "debugging": 0.45, "tool_use": 0.5 }, "sources": ["swe-bench-verified"], "note": "curated-estimate" },
+    "gemini-1.5-pro": { "caps": { "reasoning": 0.6, "codegen": 0.62, "debugging": 0.6, "tool_use": 0.6 }, "sources": ["swe-bench-verified"], "note": "curated-estimate" },
+    "deepseek-r1": { "caps": { "reasoning": 0.82, "codegen": 0.78, "debugging": 0.8, "tool_use": 0.6 }, "sources": ["swe-bench-verified", "livecodebench"], "note": "curated-estimate: reasoning tilt, weak tool-use" },
+    "deepseek-reasoner": { "caps": { "reasoning": 0.82, "codegen": 0.78, "debugging": 0.8, "tool_use": 0.6 }, "sources": ["swe-bench-verified"], "note": "curated-estimate" },
+    "deepseek-chat": { "caps": { "reasoning": 0.65, "codegen": 0.7, "debugging": 0.65, "tool_use": 0.62 }, "sources": ["livecodebench"], "note": "curated-estimate" },
+    "deepseek-v3*": { "caps": { "reasoning": 0.68, "codegen": 0.72, "debugging": 0.68, "tool_use": 0.62 }, "sources": ["swe-bench-verified", "livecodebench"], "note": "curated-estimate" },
+    "qwen3-max": { "caps": { "reasoning": 0.78, "codegen": 0.8, "debugging": 0.78, "tool_use": 0.72 }, "sources": ["swe-bench-verified", "livecodebench"], "note": "curated-estimate" },
+    "qwen3-235b*": { "caps": { "reasoning": 0.78, "codegen": 0.8, "debugging": 0.78, "tool_use": 0.7 }, "sources": ["livecodebench"], "note": "curated-estimate" },
+    "qwen3-32b": { "caps": { "reasoning": 0.65, "codegen": 0.7, "debugging": 0.65, "tool_use": 0.6 }, "sources": ["livecodebench"], "note": "curated-estimate" },
+    "qwen3-coder*": { "caps": { "reasoning": 0.68, "codegen": 0.78, "debugging": 0.72, "tool_use": 0.62 }, "sources": ["swe-bench-verified", "livecodebench"], "note": "curated-estimate: coder tilt" },
+    "qwen2.5-coder-32b*": { "caps": { "reasoning": 0.55, "codegen": 0.68, "debugging": 0.6, "tool_use": 0.55 }, "sources": ["livecodebench"], "note": "curated-estimate: coder tilt" },
+    "qwen2.5-coder-14b*": { "caps": { "reasoning": 0.45, "codegen": 0.58, "debugging": 0.5, "tool_use": 0.45 }, "sources": ["livecodebench"], "note": "curated-estimate" },
+    "qwen2.5-coder-7b*": { "caps": { "reasoning": 0.35, "codegen": 0.48, "debugging": 0.4, "tool_use": 0.38 }, "sources": ["livecodebench"], "note": "curated-estimate" },
+    "qwen3-8b": { "caps": { "reasoning": 0.45, "codegen": 0.5, "debugging": 0.45, "tool_use": 0.45 }, "sources": ["livecodebench"], "note": "curated-estimate" },
+    "qwen3-4b": { "caps": { "reasoning": 0.35, "codegen": 0.42, "debugging": 0.38, "tool_use": 0.38 }, "sources": ["livecodebench"], "note": "curated-estimate" },
+    "glm-5.2": { "caps": { "reasoning": 0.68, "codegen": 0.72, "debugging": 0.68, "tool_use": 0.66 }, "sources": ["swe-bench-verified", "artificial-analysis"], "note": "curated-estimate: provider-agnostic (z.ai/Baidu/local share this entry)" },
+    "glm-5*": { "caps": { "reasoning": 0.65, "codegen": 0.68, "debugging": 0.65, "tool_use": 0.62 }, "sources": ["artificial-analysis"], "note": "curated-estimate: GLM-5 family default" },
+    "glm-4.7": { "caps": { "reasoning": 0.62, "codegen": 0.66, "debugging": 0.62, "tool_use": 0.6 }, "sources": ["swe-bench-verified"], "note": "curated-estimate" },
+    "glm-4.6": { "caps": { "reasoning": 0.6, "codegen": 0.64, "debugging": 0.6, "tool_use": 0.58 }, "sources": ["swe-bench-verified"], "note": "curated-estimate" },
+    "glm-4.5*": { "caps": { "reasoning": 0.5, "codegen": 0.55, "debugging": 0.5, "tool_use": 0.52 }, "sources": ["swe-bench-verified"], "note": "curated-estimate" },
+    "glm-4-flash": { "caps": { "reasoning": 0.35, "codegen": 0.4, "debugging": 0.35, "tool_use": 0.38 }, "sources": ["artificial-analysis"], "note": "curated-estimate" },
+    "kimi-k3": { "caps": { "reasoning": 0.72, "codegen": 0.75, "debugging": 0.72, "tool_use": 0.7 }, "sources": ["swe-bench-verified", "artificial-analysis"], "note": "curated-estimate" },
+    "kimi-k2.6": { "caps": { "reasoning": 0.62, "codegen": 0.66, "debugging": 0.62, "tool_use": 0.62 }, "sources": ["swe-bench-verified"], "note": "curated-estimate" },
+    "kimi-k2.5": { "caps": { "reasoning": 0.6, "codegen": 0.64, "debugging": 0.6, "tool_use": 0.6 }, "sources": ["swe-bench-verified"], "note": "curated-estimate" },
+    "kimi-k2*": { "caps": { "reasoning": 0.6, "codegen": 0.63, "debugging": 0.6, "tool_use": 0.6 }, "sources": ["swe-bench-verified"], "note": "curated-estimate: Kimi-K2 family default" },
+    "llama-3.3-70b*": { "caps": { "reasoning": 0.6, "codegen": 0.65, "debugging": 0.6, "tool_use": 0.58 }, "sources": ["swe-bench-verified", "livecodebench"], "note": "curated-estimate" },
+    "llama-3.1-70b*": { "caps": { "reasoning": 0.55, "codegen": 0.6, "debugging": 0.55, "tool_use": 0.52 }, "sources": ["livecodebench"], "note": "curated-estimate" },
+    "llama-4-maverick*": { "caps": { "reasoning": 0.6, "codegen": 0.62, "debugging": 0.6, "tool_use": 0.58 }, "sources": ["swe-bench-verified"], "note": "curated-estimate" },
+    "llama-4-scout*": { "caps": { "reasoning": 0.5, "codegen": 0.52, "debugging": 0.5, "tool_use": 0.5 }, "sources": ["swe-bench-verified"], "note": "curated-estimate" },
+    "llama-3.2*": { "caps": { "reasoning": 0.25, "codegen": 0.3, "debugging": 0.25, "tool_use": 0.28 }, "sources": ["livecodebench"], "note": "curated-estimate" },
+    "mistral-large*": { "caps": { "reasoning": 0.6, "codegen": 0.65, "debugging": 0.6, "tool_use": 0.62 }, "sources": ["swe-bench-verified"], "note": "curated-estimate" },
+    "codestral*": { "caps": { "reasoning": 0.5, "codegen": 0.62, "debugging": 0.55, "tool_use": 0.5 }, "sources": ["livecodebench"], "note": "curated-estimate: coder tilt" },
+    "mistral-small*": { "caps": { "reasoning": 0.45, "codegen": 0.5, "debugging": 0.45, "tool_use": 0.48 }, "sources": ["artificial-analysis"], "note": "curated-estimate" },
+    "devstral*": { "caps": { "reasoning": 0.5, "codegen": 0.6, "debugging": 0.55, "tool_use": 0.5 }, "sources": ["swe-bench-verified"], "note": "curated-estimate: coder tilt" },
+    "gpt-oss-120b": { "caps": { "reasoning": 0.65, "codegen": 0.68, "debugging": 0.65, "tool_use": 0.55 }, "sources": ["swe-bench-verified"], "note": "curated-estimate" },
+    "gpt-oss-20b": { "caps": { "reasoning": 0.45, "codegen": 0.5, "debugging": 0.45, "tool_use": 0.42 }, "sources": ["swe-bench-verified"], "note": "curated-estimate" },
+    "gemma-3-27b": { "caps": { "reasoning": 0.45, "codegen": 0.48, "debugging": 0.45, "tool_use": 0.45 }, "sources": ["livecodebench"], "note": "curated-estimate" },
+    "gemma-2*": { "caps": { "reasoning": 0.28, "codegen": 0.32, "debugging": 0.28, "tool_use": 0.3 }, "sources": ["livecodebench"], "note": "curated-estimate" },
+    "phi-4": { "caps": { "reasoning": 0.45, "codegen": 0.48, "debugging": 0.45, "tool_use": 0.4 }, "sources": ["livecodebench"], "note": "curated-estimate" },
+    "phi-3*": { "caps": { "reasoning": 0.28, "codegen": 0.32, "debugging": 0.28, "tool_use": 0.28 }, "sources": ["livecodebench"], "note": "curated-estimate" },
+    "minimax-m2*": { "caps": { "reasoning": 0.65, "codegen": 0.68, "debugging": 0.65, "tool_use": 0.65 }, "sources": ["swe-bench-verified"], "note": "curated-estimate" },
+    "muse-spark-1.3": { "caps": { "reasoning": 0.85, "codegen": 0.87, "debugging": 0.85, "tool_use": 0.75 }, "sources": ["deepswe-1.1@75.4% (vendor-reported, max setting)", "terminal-bench-2.1@88.8% (vendor-reported)"], "note": "curated-estimate: full 1.3 at max reasoning. Vendor self-reported; absent from public DeepSWE leaderboard (only 1.2 listed); hands-on gap reported by third parties. tool_use capped — no direct tool-use data. Revisit on independent rows." },
+    "muse-spark-1.2*": { "caps": { "reasoning": 0.42, "codegen": 0.46, "debugging": 0.42, "tool_use": 0.44 }, "sources": ["lynkr-telemetry:4176 scored rows, avg quality 52.7 @ complexity 51.7"], "note": "measured: operator's own traffic. Below the 70 evidence bar — economize trivial turns only." },
+    "muse-spark-1.3-contributor*": { "caps": { "reasoning": 0.4, "codegen": 0.45, "debugging": 0.4, "tool_use": 0.45 }, "sources": ["lynkr-telemetry:29 scored rows (thin) + 1.2 family history"], "note": "provisional: free/contributor serving grades far below full-1.3 published scores (variant, effort setting, or benchmark inflation — undetermined). Do NOT cover with muse-spark*: the full model and the free serving must never share caps." }
+  }
+}

package/config/model-tiers.json

@@ -54,6 +54,9 @@
         ],
         "moonshot": [
           "kimi-k2.6"
+        ],
+        "fireworks": [
+          "accounts/fireworks/models/llama-3.1-8b-instruct"
         ]
       }
     },
@@ -109,6 +112,9 @@
         ],
         "moonshot": [
           "kimi-k2.6"
+        ],
+        "fireworks": [
+          "accounts/fireworks/models/kimi-k2-instruct-0905"
         ]
       }
     },
@@ -158,6 +164,9 @@
         ],
         "moonshot": [
           "kimi-k2.6"
+        ],
+        "fireworks": [
+          "accounts/fireworks/models/deepseek-v3p1"
         ]
       }
     },
@@ -205,6 +214,9 @@
         "moonshot": [
           "kimi-k3",
           "kimi-k2.6"
+        ],
+        "fireworks": [
+          "accounts/fireworks/models/glm-5p2"
         ]
       }
     }
@@ -233,4 +245,4 @@
     "kimi": "moonshot",
     "z-ai": "zai"
   }
-}
+}

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "lynkr",
-  "version": "9.14.1",
+  "version": "9.14.3",
   "files": [
     "index.js",
     "install.sh",
@@ -38,13 +38,14 @@
     "start:supervised": "while true; do node index.js 2>&1 | npx pino-pretty --sync; code=$?; echo \"[supervisor] lynkr exited ($code) — restarting in 3s\"; sleep 3; done",
     "lint": "eslint src index.js",
     "test": "npm run test:unit && npm run test:performance",
-    "test:unit": "DATABRICKS_API_KEY=test-key DATABRICKS_API_BASE=http://test.com LOG_FILE_ENABLED=false node --test test/routing.test.js test/hybrid-routing-integration.test.js test/retry-logic.test.js test/sse-transformer.test.js test/passthrough-stream.test.js test/passthrough-mode.test.js test/openrouter-error-resilience.test.js test/format-conversion.test.js test/azure-openai-config.test.js test/azure-openai-format-conversion.test.js test/azure-openai-routing.test.js test/azure-openai-streaming.test.js test/azure-openai-error-resilience.test.js test/azure-openai-integration.test.js test/openai-integration.test.js test/atlas-integration.test.js test/toon-compression.test.js test/gcf-compression.test.js test/llamacpp-integration.test.js test/resilience.test.js test/telemetry-routing.test.js test/memory/store.test.js test/memory/surprise.test.js test/memory/extractor.test.js test/memory/search.test.js test/memory/retriever.test.js test/memory/distiller.test.js test/memory/distiller-freeze.test.js test/memory/wiki.test.js test/memory/skills-cache.test.js test/memory/tencentdb-launcher.test.js test/distill.test.js test/large-payload.test.js test/prompt-cache-injection.test.js test/risk-analyzer.test.js test/interaction-block.test.js test/preflight.test.js test/token-reduction.test.js test/session-affinity.test.js test/cache-state.test.js test/cache-switch-cost.test.js test/lens-recommendations.test.js test/model-registry-cost.test.js test/output-format-guard.test.js test/tier-fallback.test.js test/wrap.test.js test/init.test.js test/tool-call-response-metadata.test.js test/degradation.test.js test/routing-telemetry-columns.test.js test/sticky-routing.test.js test/knn-ambiguous-escalate.test.js test/deescalator.test.js test/client-profiles.test.js test/strip-internal-fields.test.js test/complexity-tool-subtraction.test.js test/bandit.test.js test/routing-propensity.test.js test/reward-pipeline.test.js test/knn-cold-start.test.js test/calibration.test.js test/feedback-loop.test.js test/session-fingerprint.test.js test/side-request-guards.test.js test/verifier.test.js test/intent-score.test.js test/difficulty-classifier.test.js test/classifier-setup.test.js test/usage-stats.test.js test/loop-guard.test.js test/moonshot-model-mapping.test.js test/baidu-model-mapping.test.js",
-    "test:memory": "DATABRICKS_API_KEY=test-key DATABRICKS_API_BASE=http://test.com node --test test/memory/store.test.js test/memory/surprise.test.js test/memory/extractor.test.js test/memory/search.test.js test/memory/retriever.test.js test/memory/distiller.test.js test/memory/distiller-freeze.test.js test/memory/wiki.test.js test/memory/skills-cache.test.js test/memory/tencentdb-launcher.test.js",
-    "test:new-features": "DATABRICKS_API_KEY=test-key DATABRICKS_API_BASE=http://test.com node --test test/passthrough-mode.test.js test/openrouter-error-resilience.test.js test/format-conversion.test.js",
-    "test:performance": "DATABRICKS_API_KEY=test-key DATABRICKS_API_BASE=http://test.com node test/hybrid-routing-performance.test.js && DATABRICKS_API_KEY=test-key DATABRICKS_API_BASE=http://test.com node test/performance-tests.js",
-    "test:benchmark": "DATABRICKS_API_KEY=test-key DATABRICKS_API_BASE=http://test.com node test/performance-benchmark.js",
-    "test:quick": "DATABRICKS_API_KEY=test-key DATABRICKS_API_BASE=http://test.com node --test test/routing.test.js",
-    "test:all": "npm run test:unit && npm run test:performance && npm run test:benchmark"
+    "test:unit": "LYNKR_KNN_DIR=/tmp/lynkr-test-knn DATABRICKS_API_KEY=test-key DATABRICKS_API_BASE=http://test.com LOG_FILE_ENABLED=false node --test test/routing.test.js test/hybrid-routing-integration.test.js test/retry-logic.test.js test/sse-transformer.test.js test/passthrough-stream.test.js test/passthrough-mode.test.js test/openrouter-error-resilience.test.js test/format-conversion.test.js test/azure-openai-config.test.js test/azure-openai-format-conversion.test.js test/azure-openai-routing.test.js test/azure-openai-streaming.test.js test/azure-openai-error-resilience.test.js test/azure-openai-integration.test.js test/openai-integration.test.js test/atlas-integration.test.js test/toon-compression.test.js test/gcf-compression.test.js test/llamacpp-integration.test.js test/resilience.test.js test/telemetry-routing.test.js test/memory/store.test.js test/memory/surprise.test.js test/memory/extractor.test.js test/memory/search.test.js test/memory/retriever.test.js test/memory/distiller.test.js test/memory/distiller-freeze.test.js test/memory/wiki.test.js test/memory/skills-cache.test.js test/memory/tencentdb-launcher.test.js test/distill.test.js test/large-payload.test.js test/prompt-cache-injection.test.js test/risk-analyzer.test.js test/interaction-block.test.js test/preflight.test.js test/token-reduction.test.js test/session-affinity.test.js test/cache-state.test.js test/cache-switch-cost.test.js test/lens-recommendations.test.js test/model-registry-cost.test.js test/output-format-guard.test.js test/tier-fallback.test.js test/wrap.test.js test/init.test.js test/tool-call-response-metadata.test.js test/degradation.test.js test/routing-telemetry-columns.test.js test/sticky-routing.test.js test/knn-ambiguous-escalate.test.js test/deescalator.test.js test/client-profiles.test.js test/strip-internal-fields.test.js test/complexity-tool-subtraction.test.js test/bandit.test.js test/routing-propensity.test.js test/reward-pipeline.test.js test/knn-cold-start.test.js test/calibration.test.js test/feedback-loop.test.js test/session-fingerprint.test.js test/side-request-guards.test.js test/verifier.test.js test/intent-score.test.js test/difficulty-classifier.test.js test/classifier-setup.test.js test/usage-stats.test.js test/loop-guard.test.js test/moonshot-model-mapping.test.js test/baidu-model-mapping.test.js test/tenant-policy-ingress-parity.test.js test/decide.test.js test/embeddings-degradation.test.js test/health-probe.test.js test/stuck-detector.test.js test/onnx-embedder.test.js test/ope.test.js test/hierarchical-budget.test.js test/token-rate-limit.test.js test/otel-export.test.js test/mcp-broker.test.js test/compression-budget.test.js test/gpt-utils.test.js test/dedup-observe-only.test.js test/context-window-header.test.js test/token-budget-auto.test.js test/opencode-setup.test.js test/auth-mode-first-party.test.js",
+    "test:memory": "LYNKR_KNN_DIR=/tmp/lynkr-test-knn DATABRICKS_API_KEY=test-key DATABRICKS_API_BASE=http://test.com node --test test/memory/store.test.js test/memory/surprise.test.js test/memory/extractor.test.js test/memory/search.test.js test/memory/retriever.test.js test/memory/distiller.test.js test/memory/distiller-freeze.test.js test/memory/wiki.test.js test/memory/skills-cache.test.js test/memory/tencentdb-launcher.test.js",
+    "test:new-features": "LYNKR_KNN_DIR=/tmp/lynkr-test-knn DATABRICKS_API_KEY=test-key DATABRICKS_API_BASE=http://test.com node --test test/passthrough-mode.test.js test/openrouter-error-resilience.test.js test/format-conversion.test.js",
+    "test:performance": "LYNKR_KNN_DIR=/tmp/lynkr-test-knn DATABRICKS_API_KEY=test-key DATABRICKS_API_BASE=http://test.com node test/hybrid-routing-performance.test.js && DATABRICKS_API_KEY=test-key DATABRICKS_API_BASE=http://test.com node test/performance-tests.js",
+    "test:benchmark": "LYNKR_KNN_DIR=/tmp/lynkr-test-knn DATABRICKS_API_KEY=test-key DATABRICKS_API_BASE=http://test.com node test/performance-benchmark.js",
+    "test:quick": "LYNKR_KNN_DIR=/tmp/lynkr-test-knn DATABRICKS_API_KEY=test-key DATABRICKS_API_BASE=http://test.com node --test test/routing.test.js",
+    "test:all": "npm run test:unit && npm run test:performance && npm run test:benchmark",
+    "seed:capabilities": "node scripts/seed-capabilities.js --refresh"
   },
   "keywords": [
     "llm",
@@ -90,32 +91,23 @@
     "node": ">=20.0.0"
   },
   "dependencies": {
-    "@azure/openai": "^2.0.0",
-    "@babel/parser": "^7.29.0",
-    "@babel/traverse": "^7.29.0",
     "@blackwell-systems/gcf": "2.4.0",
-    "@tencentdb-agent-memory/memory-tencentdb": "^1.0.1",
     "@toon-format/toon": "^2.1.0",
     "cockatiel": "^3.2.1",
-    "compression": "^1.7.4",
     "diff": "^5.2.0",
     "dotenv": "^16.4.5",
     "express": "^5.1.0",
     "express-rate-limit": "^8.2.1",
-    "fast-glob": "^3.3.2",
-    "graphify": "^1.0.0",
     "hnswlib-node": "^3.0.0",
     "js-tiktoken": "^1.0.20",
-    "js-yaml": "^4.1.1",
-    "openai": "^6.14.0",
     "pino": "^8.17.2",
     "pino-http": "^8.6.0",
-    "pino-roll": "^4.0.0",
-    "undici": "^6.22.0"
+    "pino-roll": "^4.0.0"
   },
   "optionalDependencies": {
+    "@huggingface/transformers": "^4.2.0",
     "better-sqlite3": "^12.11.1",
-    "dockerode": "^4.0.2",
+    "dockerode": "^5.0.1",
     "tree-sitter": "^0.21.1",
     "tree-sitter-javascript": "^0.21.0",
     "tree-sitter-python": "^0.21.0",
@@ -127,7 +119,9 @@
     "nodemon": "^3.1.0",
     "pino-pretty": "^10.2.0"
   },
-  "allowScripts": {
-    "@tencentdb-agent-memory/memory-tencentdb@1.0.1": true
+  "allowScripts": {},
+  "overrides": {
+    "sharp": "^0.35.4",
+    "adm-zip": "^0.6.0"
   }
 }

package/scripts/claude-desktop.js

@@ -0,0 +1,212 @@
+#!/usr/bin/env node
+/**
+ * Install or remove the Lynkr profile for Claude Desktop's native
+ * third-party inference mode (macOS only).
+ *
+ * Claude Desktop supports a "3p" deployment mode in which it sends its
+ * Anthropic Messages API traffic to a local gateway instead of
+ * api.anthropic.com. The mechanism (reverse-engineered from Ollama's
+ * cmd/launch/claude_desktop.go, which uses the same route) is three JSON
+ * files:
+ *
+ *   ~/Library/Application Support/Claude/claude_desktop_config.json
+ *       deploymentMode: "1p" | "3p"
+ *   ~/Library/Application Support/Claude-3p/claude_desktop_config.json
+ *       deploymentMode again, for the third-party profile root
+ *   ~/Library/Application Support/Claude-3p/configLibrary/_meta.json
+ *       profile registry: { appliedId, entries: [{id, name}] }
+ *   ~/Library/Application Support/Claude-3p/configLibrary/<uuid>.json
+ *       the gateway profile itself (base URL, bearer key, display name)
+ *
+ * Usage:
+ *   node scripts/claude-desktop.js --install [--url http://127.0.0.1:8081] [--key lynkr]
+ *   node scripts/claude-desktop.js --restore
+ *   node scripts/claude-desktop.js --status
+ *
+ * WARNING: this is undocumented Claude Desktop surface observed via
+ * Ollama's public integration; a Desktop update may change the key names.
+ * --restore (or `ollama launch claude-desktop --restore`) always brings
+ * back the stock Anthropic profile by flipping deploymentMode to "1p".
+ */
+
+const fs = require("fs");
+const os = require("os");
+const path = require("path");
+const { execFileSync } = require("child_process");
+
+const PROFILE_ID = "00000000-0000-4000-8000-000000008081";
+const PROFILE_NAME = "Lynkr";
+const DEFAULT_URL = "http://127.0.0.1:8081";
+const DEFAULT_KEY = "lynkr";
+
+const appSupport = path.join(os.homedir(), "Library", "Application Support");
+const PATHS = {
+  normalConfig: path.join(appSupport, "Claude", "claude_desktop_config.json"),
+  tpConfig: path.join(appSupport, "Claude-3p", "claude_desktop_config.json"),
+  meta: path.join(appSupport, "Claude-3p", "configLibrary", "_meta.json"),
+  profile: path.join(appSupport, "Claude-3p", "configLibrary", `${PROFILE_ID}.json`),
+  backup: path.join(appSupport, "Claude-3p", "configLibrary", ".lynkr-backup.json"),
+};
+
+function readJson(file) {
+  try {
+    return JSON.parse(fs.readFileSync(file, "utf8"));
+  } catch (err) {
+    if (err.code === "ENOENT") return {};
+    throw new Error(`Cannot parse ${file}: ${err.message}`);
+  }
+}
+
+function writeJson(file, obj) {
+  fs.mkdirSync(path.dirname(file), { recursive: true });
+  fs.writeFileSync(file, JSON.stringify(obj, null, 2) + "\n");
+}
+
+function setDeploymentMode(file, mode) {
+  const cfg = readJson(file);
+  cfg.deploymentMode = mode;
+  writeJson(file, cfg);
+}
+
+function claudeDesktopRunning() {
+  try {
+    execFileSync("pgrep", ["-x", "Claude"], { stdio: "pipe" });
+    return true;
+  } catch {
+    return false;
+  }
+}
+
+function install(opts) {
+  if (!fs.existsSync(path.join(appSupport, "Claude"))) {
+    fail("Claude Desktop does not appear to be installed (no ~/Library/Application Support/Claude).");
+  }
+
+  // First install: remember what we're replacing so --restore is faithful.
+  if (!fs.existsSync(PATHS.backup)) {
+    const meta = readJson(PATHS.meta);
+    const normal = readJson(PATHS.normalConfig);
+    writeJson(PATHS.backup, {
+      previousAppliedId: meta.appliedId ?? null,
+      previousDeploymentMode: normal.deploymentMode ?? "1p",
+    });
+  }
+
+  // Profile: same keys Ollama's launcher writes, pointed at Lynkr.
+  const profile = readJson(PATHS.profile);
+  Object.assign(profile, {
+    inferenceProvider: "gateway",
+    inferenceGatewayBaseUrl: opts.url,
+    inferenceGatewayApiKey: opts.key,
+    inferenceGatewayAuthScheme: "bearer",
+    deploymentDisplayName: PROFILE_NAME,
+    chatTabEnabled: true,
+    disableDeploymentModeChooser: true,
+    coworkEgressAllowedHosts: ["*"],
+    disableEssentialTelemetry: true,
+    disableNonessentialTelemetry: true,
+    autoModeEnabled: true,
+  });
+  delete profile.inferenceModels;
+  writeJson(PATHS.profile, profile);
+
+  // Registry: apply our profile, dedupe any stale entry for our id.
+  const meta = readJson(PATHS.meta);
+  meta.appliedId = PROFILE_ID;
+  meta.entries = (Array.isArray(meta.entries) ? meta.entries : [])
+    .filter((e) => !(e && e.id === PROFILE_ID))
+    .concat([{ id: PROFILE_ID, name: PROFILE_NAME }]);
+  writeJson(PATHS.meta, meta);
+
+  setDeploymentMode(PATHS.normalConfig, "3p");
+  setDeploymentMode(PATHS.tpConfig, "3p");
+
+  console.log(`Claude Desktop profile changed to ${PROFILE_NAME}.`);
+  console.log(`Gateway: ${opts.url} (bearer key: ${opts.key})`);
+  console.log("To restore the usual Claude profile: node scripts/claude-desktop.js --restore");
+  if (claudeDesktopRunning()) {
+    console.log("\nClaude Desktop is running — quit and reopen it for the change to take effect.");
+  }
+  console.log("\nReminder: Lynkr must be running and restarted with the claude-desktop-gateway");
+  console.log("models endpoint (src/api/claude-desktop-gateway.js) for model discovery to work.");
+}
+
+function restore() {
+  const backup = readJson(PATHS.backup);
+
+  setDeploymentMode(PATHS.normalConfig, backup.previousDeploymentMode || "1p");
+  setDeploymentMode(PATHS.tpConfig, backup.previousDeploymentMode || "1p");
+
+  const meta = readJson(PATHS.meta);
+  meta.entries = (Array.isArray(meta.entries) ? meta.entries : []).filter(
+    (e) => !(e && e.id === PROFILE_ID)
+  );
+  if (meta.appliedId === PROFILE_ID) {
+    if (backup.previousAppliedId) meta.appliedId = backup.previousAppliedId;
+    else delete meta.appliedId;
+  }
+  writeJson(PATHS.meta, meta);
+
+  for (const file of [PATHS.profile, PATHS.backup]) {
+    try {
+      fs.unlinkSync(file);
+    } catch (err) {
+      if (err.code !== "ENOENT") {
+        console.warn(`Couldn't remove ${file}: ${err.message}`);
+      }
+    }
+  }
+
+  console.log("Claude Desktop restored to the usual Claude profile.");
+  if (claudeDesktopRunning()) {
+    console.log("Claude Desktop is running — quit and reopen it for the change to take effect.");
+  }
+}
+
+function status() {
+  const normal = readJson(PATHS.normalConfig);
+  const meta = readJson(PATHS.meta);
+  const profile = readJson(PATHS.profile);
+  const mode = normal.deploymentMode || "1p";
+  const applied = meta.appliedId === PROFILE_ID;
+  console.log(`deploymentMode: ${mode}`);
+  console.log(`Lynkr profile installed: ${fs.existsSync(PATHS.profile)}`);
+  console.log(`Lynkr profile applied: ${applied}`);
+  if (profile.inferenceGatewayBaseUrl) {
+    console.log(`gateway URL: ${profile.inferenceGatewayBaseUrl}`);
+  }
+  console.log(
+    mode === "3p" && applied
+      ? "\nClaude Desktop is routed through Lynkr."
+      : "\nClaude Desktop is on the stock Anthropic profile."
+  );
+}
+
+function fail(msg) {
+  console.error(msg);
+  process.exit(1);
+}
+
+function main() {
+  if (process.platform !== "darwin") {
+    fail("Claude Desktop profile management is only supported on macOS (matching the upstream integration).");
+  }
+  const args = process.argv.slice(2);
+  const getFlag = (name, fallback) => {
+    const i = args.indexOf(name);
+    return i !== -1 && args[i + 1] ? args[i + 1] : fallback;
+  };
+
+  if (args.includes("--install")) {
+    install({ url: getFlag("--url", DEFAULT_URL), key: getFlag("--key", DEFAULT_KEY) });
+  } else if (args.includes("--restore")) {
+    restore();
+  } else if (args.includes("--status")) {
+    status();
+  } else {
+    console.log("Usage: node scripts/claude-desktop.js --install [--url URL] [--key KEY] | --restore | --status");
+    process.exit(args.length === 0 ? 0 : 1);
+  }
+}
+
+main();

package/scripts/ope-report.js

@@ -0,0 +1,78 @@
+#!/usr/bin/env node
+/**
+ * Off-policy evaluation report (ROUTING-NOTES §3 + §4.10.3).
+ *
+ * Scores counterfactual routing policies against logged production
+ * decisions — no live traffic involved. Reads routing_telemetry rows
+ * (propensity + candidates + quality_score, and context where the bandit
+ * ran) and prints IPS / SNIPS / DR / WDR estimates for each reference
+ * policy, alongside the logged policy's actual mean reward.
+ *
+ * Usage:
+ *   node scripts/ope-report.js [--days 30] [--limit 50000]
+ *
+ * Reading the output:
+ *   - "logged" is what the live policy actually achieved (ground truth).
+ *   - A candidate policy whose WDR estimate beats "logged" — with a healthy
+ *     effective sample size — is evidence worth acting on. A tiny ESS means
+ *     a few high-weight rows dominate; distrust the number regardless of
+ *     how good it looks.
+ *   - Rows logged before the `context` column existed contribute to
+ *     IPS/SNIPS but not to the DR regression term (drRows shows coverage).
+ */
+
+const telemetry = require('../src/routing/telemetry');
+const { evaluatePolicy, policies } = require('../src/routing/ope');
+
+function arg(name, fallback) {
+  const idx = process.argv.indexOf(`--${name}`);
+  if (idx === -1 || idx + 1 >= process.argv.length) return fallback;
+  const n = Number.parseInt(process.argv[idx + 1], 10);
+  return Number.isNaN(n) ? fallback : n;
+}
+
+const days = arg('days', 30);
+const limit = arg('limit', 50_000);
+
+const rows = telemetry.query({
+  since: Date.now() - days * 24 * 60 * 60 * 1000,
+  limit,
+});
+
+if (rows.length === 0) {
+  console.log(`No telemetry rows in the last ${days} days.`);
+  process.exit(0);
+}
+
+// Multi-candidate rows are where policies can actually differ; single-
+// candidate rows contribute identically to every policy's estimate.
+const multiCandidate = rows.filter((r) => {
+  try {
+    const c = typeof r.candidates === 'string' ? JSON.parse(r.candidates) : r.candidates;
+    return Array.isArray(c) && c.length > 1;
+  } catch {
+    return false;
+  }
+});
+
+console.log(`\nOff-policy evaluation — last ${days} days`);
+console.log(`rows: ${rows.length} total, ${multiCandidate.length} with a real choice (>1 candidate)\n`);
+
+const fmt = (v) => (v == null ? '   n/a' : (v * 100).toFixed(2).padStart(6));
+
+const results = [];
+for (const [name, policyFn] of Object.entries(policies)) {
+  const r = evaluatePolicy(rows, policyFn);
+  results.push({ name, ...r });
+}
+
+const logged = results[0]?.loggedMeanReward;
+console.log(`logged policy actual mean reward: ${logged == null ? 'n/a' : (logged * 100).toFixed(2)} (quality points)\n`);
+console.log('policy                 |    IPS |  SNIPS |     DR |    WDR |     ESS | usable | drRows');
+console.log('-----------------------+--------+--------+--------+--------+---------+--------+-------');
+for (const r of results) {
+  console.log(
+    `${r.name.padEnd(22)} | ${fmt(r.ips)} | ${fmt(r.snips)} | ${fmt(r.dr)} | ${fmt(r.wdr)} | ${String(r.effectiveSampleSize == null ? 'n/a' : Math.round(r.effectiveSampleSize)).padStart(7)} | ${String(r.usable).padStart(6)} | ${String(r.drRows).padStart(6)}`
+  );
+}
+console.log('\nEstimates are quality points (0-100 scale). Prefer WDR; check ESS before trusting any gap.');