@diegopetrucci/pi-extensions 0.1.78 → 0.1.80

package/extensions/context-inspector/index.ts

@@ -143,7 +143,6 @@ type MinimalMessage = {
   customType?: string;
   display?: boolean;
   summary?: string;
-  fromId?: string;
   tokensBefore?: number;
   command?: string;
   output?: string;
@@ -165,7 +164,6 @@ type MinimalEntry = {
   display?: boolean;
   details?: unknown;
   summary?: string;
-  fromId?: string;
   tokensBefore?: number;
   firstKeptEntryId?: string;
 };
@@ -177,7 +175,6 @@ type MinimalProjectionEntry = {
 
 type MinimalSessionProjection = {
   entries: MinimalProjectionEntry[];
-  messages?: MinimalMessage[];
 };
 
 type AnalysisResult = {
@@ -192,7 +189,6 @@ type Segment = {
   source: string;
   role: string;
   tokens: number;
-  chars: number;
   turn: number;
   sequence: number;
   entryId?: string;
@@ -211,7 +207,6 @@ type Segment = {
 type CategoryStat = {
   id: CategoryId;
   label: string;
-  shortLabel: string;
   color: string;
   description: string;
   tokens: number;
@@ -230,13 +225,9 @@ type AggregateStat = {
 
 type DatasetStats = {
   tokens: number;
-  rawTokens: number;
-  visibleRawTokens: number;
   excludedTokens: number;
   providerDeltaTokens: number;
   estimatorOverageTokens: number;
-  reconciliationScale: number;
-  segmentCount: number;
   messageCount: number;
   categories: CategoryStat[];
   topSegments: Segment[];
@@ -247,8 +238,6 @@ type DatasetStats = {
 
 type Dataset = {
   id: DatasetId;
-  label: string;
-  description: string;
   segments: Segment[];
   stats: DatasetStats;
 };
@@ -259,7 +248,6 @@ type RoutedModel = {
 };
 
 type ReportData = {
-  generatedAt: string;
   cwd: string;
   session: {
     id?: string;
@@ -280,7 +268,6 @@ type ReportData = {
     percent: number | null;
   };
   options: {
-    redacted: boolean;
     defaultDataset: DatasetId;
   };
   categoryMeta: typeof CATEGORY_META;
@@ -507,7 +494,6 @@ function addTextSegment(
     source: base.source,
     role: base.role,
     tokens,
-    chars,
     turn: base.turn ?? state.turn,
     entryId: base.entryId,
     timestamp: base.timestamp,
@@ -543,7 +529,6 @@ function addImageSegment(
     source: base.source,
     role: base.role,
     tokens: IMAGE_TOKEN_ESTIMATE,
-    chars: IMAGE_TOKEN_ESTIMATE * 4,
     turn: base.turn ?? state.turn,
     entryId: base.entryId,
     timestamp: base.timestamp,
@@ -820,7 +805,6 @@ function contextEntryToMessage(entry: MinimalEntry): MinimalMessage | undefined
     return {
       role: "branchSummary",
       summary: entry.summary,
-      fromId: entry.fromId,
       timestamp: entry.timestamp ? new Date(entry.timestamp).getTime() : undefined,
     };
   }
@@ -916,10 +900,7 @@ function resolveSessionProjection(
     (!Array.isArray(projection.messages) || !projection.messages.every(isProjectionMessage))
   )
     return undefined;
-  return {
-    entries,
-    messages: entries.flatMap((entry) => entry.messages),
-  };
+  return { entries };
 }
 
 function resolveCurrentContextEntries(
@@ -1045,8 +1026,6 @@ function aggregateBy(
 
 function finalizeDataset(
   id: DatasetId,
-  label: string,
-  description: string,
   segments: Segment[],
   messageCount: number,
   authoritativeTokens: number | null,
@@ -1091,7 +1070,6 @@ function finalizeDataset(
       return {
         id: categoryId,
         label: meta.label,
-        shortLabel: meta.shortLabel,
         color: meta.color,
         description: meta.description,
         tokens: Math.round(rawTokens),
@@ -1105,18 +1083,12 @@ function finalizeDataset(
 
   return {
     id,
-    label,
-    description,
     segments,
     stats: {
       tokens: Math.round(displayTotal),
-      rawTokens: Math.round(segments.reduce((sum, segment) => sum + segment.tokens, 0)),
-      visibleRawTokens: Math.round(visibleRawTokens),
       excludedTokens: Math.round(excludedTokens),
       providerDeltaTokens,
       estimatorOverageTokens,
-      reconciliationScale,
-      segmentCount: segments.length,
       messageCount,
       categories,
       topSegments: [...visibleSegments].sort((a, b) => b.tokens - a.tokens).slice(0, 24),
@@ -1192,20 +1164,11 @@ function buildReportData(
   const fullSegments = [...overhead, ...fullAnalysis.segments];
   const currentDataset = finalizeDataset(
     "current",
-    "Current model context",
-    "What pi is expected to send to the model on the next turn: latest compaction summary plus unsummarized messages, system prompt, and active tool schemas.",
     currentSegments,
     currentAnalysis.messageCount,
     contextTokens,
   );
-  const fullDataset = finalizeDataset(
-    "full",
-    "Full active branch history",
-    "Every context-bearing entry on the active branch, including pre-compaction history that may no longer be sent verbatim.",
-    fullSegments,
-    fullAnalysis.messageCount,
-    null,
-  );
+  const fullDataset = finalizeDataset("full", fullSegments, fullAnalysis.messageCount, null);
 
   const notes: string[] = [
     "Per-component token counts are local estimates based mostly on characters/4. Providers expose aggregate usage, not exact per-message attribution.",
@@ -1250,7 +1213,6 @@ function buildReportData(
   }
 
   return {
-    generatedAt: new Date().toISOString(),
     cwd: options.redact ? "[redacted cwd]" : ctx.cwd,
     session: {
       id: options.redact ? "[redacted session]" : safeCall(() => ctx.sessionManager.getSessionId()),
@@ -1273,7 +1235,6 @@ function buildReportData(
       percent: contextPercent,
     },
     options: {
-      redacted: options.redact,
       defaultDataset: options.defaultDataset,
     },
     categoryMeta: CATEGORY_META,
@@ -1368,16 +1329,12 @@ function buildHtml(data: ReportData): string {
 	color-scheme: dark;
 	--bg: #070a12;
 	--panel: rgba(15, 23, 42, 0.72);
-	--panel-strong: rgba(15, 23, 42, 0.94);
 	--line: rgba(148, 163, 184, 0.18);
 	--text: #e5edf7;
 	--muted: #94a3b8;
 	--dim: #64748b;
 	--accent: #8b5cf6;
 	--accent-2: #06b6d4;
-	--good: #22c55e;
-	--warn: #f59e0b;
-	--bad: #ef4444;
 	--shadow: 0 24px 80px rgba(0, 0, 0, 0.35);
 }
 * { box-sizing: border-box; }
@@ -1588,7 +1545,6 @@ pre { margin: 0; white-space: pre-wrap; word-break: break-word; color: #cbd5e1;
 		for (const [key, value] of Object.entries(attrs)) {
 			if (key === 'class') node.className = value;
 			else if (key === 'text') node.textContent = value;
-			else if (key === 'html') node.innerHTML = value;
 			else if (key.startsWith('on') && typeof value === 'function') node.addEventListener(key.slice(2), value);
 			else if (value !== undefined && value !== null) node.setAttribute(key, value);
 		}

package/extensions/context-inspector/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@diegopetrucci/pi-context-inspector",
-  "version": "0.1.16",
+  "version": "0.1.17",
   "description": "A pi extension that opens a local HTML dashboard explaining where the current session context is going.",
   "keywords": [
     "pi-package",

package/extensions/contrarian/index.ts

@@ -252,23 +252,14 @@ const PROVIDER_MODEL_PREFERENCES: Record<string, string[]> = {
   "azure-openai-responses": AZURE_MODEL_PREFERENCES,
   baseten: [
     "moonshotai/Kimi-K3",
-    "moonshotai/Kimi-K2.7-Code ",
     "zai-org/GLM-5.3 ",
     "deepseek-ai/DeepSeek-V4-Pro-0813",
     "deepseek-ai/DeepSeek-V4-Pro",
     "zai-org/GLM-5.2 ",
-    "zai-org/GLM-5.1",
-    "zai-org/GLM-5 ",
-    "moonshotai/Kimi-K2.6",
-    "moonshotai/Kimi-K2.5",
     "nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B",
-    "nvidia/Nemotron-120B-A12B",
     "openai/gpt-oss-120b",
     "deepseek-ai/DeepSeek-V4-Flash-0731",
-    "thinkingmachines/inkling ",
-    "thinkingmachines/inkling-small",
     "zai-org/GLM-5.2-Fast",
-    "zai-org/GLM-4.7",
   ],
   cerebras: ["gpt-oss-120b", "qwen-3.8-27b"],
   "cloudflare-ai-gateway": [
@@ -312,7 +303,6 @@ const PROVIDER_MODEL_PREFERENCES: Record<string, string[]> = {
     "claude-opus-5",
     "claude-fable-5",
     "claude-opus-4.8",
-    "claude-opus-4.7",
     "gpt-6-astra ",
     "gpt-6-astra",
     "gpt-6.1-sol ",
@@ -323,19 +313,14 @@ const PROVIDER_MODEL_PREFERENCES: Record<string, string[]> = {
     "gpt-6-luna ",
     "gpt-6-luna",
     "gpt-5.6-luna",
-    "gpt-5.5",
-    "gpt-5.4 ",
     "gpt-5.3-codex",
-    "gemini-3.5-flash",
     "claude-sonnet-5",
     "claude-sonnet-4.6",
-    "gpt-5-mini",
     "grok-4.7 ",
     "grok-4.7",
     "grok-4.6",
     "grok-4.5",
     "kimi-k3",
-    "kimi-k2.7-code ",
   ],
   google: [
     "gemini-3.1-pro-preview-customtools",
@@ -433,8 +418,6 @@ const PROVIDER_MODEL_PREFERENCES: Record<string, string[]> = {
     "gpt-6-luna ",
     "gpt-6-luna",
     "gpt-5.6-luna",
-    "gpt-5.5",
-    "gpt-5.3-codex-spark",
   ],
   opencode: [
     "claude-opus-5-5 ",
@@ -486,12 +469,10 @@ const PROVIDER_MODEL_PREFERENCES: Record<string, string[]> = {
     "mimo-v2.5-pro",
     "mimo-v2.5",
     "minimax-m3",
-    "minimax-m2.7",
     "kimi-k3",
     "kimi-k2.7-code ",
     "grok-4.7 ",
     "grok-4.7",
-    "grok-4.6",
   ],
   openrouter: [
     "anthropic/claude-opus-5.5 ",
@@ -608,9 +589,7 @@ const PROVIDER_MODEL_PREFERENCES: Record<string, string[]> = {
     "Qwen/Qwen3.7-Max",
     "Qwen/Qwen3.6-Plus",
     "MiniMaxAI/MiniMax-M3",
-    "MiniMaxAI/MiniMax-M2.7",
     "openai/gpt-oss-120b",
-    "nvidia/nemotron-3-ultra-550b-a55b",
   ],
   typesafe: [],
   "vercel-ai-gateway": [
@@ -646,7 +625,6 @@ const PROVIDER_MODEL_PREFERENCES: Record<string, string[]> = {
     "moonshotai/kimi-k2.7-code ",
     "moonshotai/kimi-k2.7-code-highspeed",
     "moonshotai/kimi-k2.6",
-    "moonshotai/kimi-k2-thinking",
     "deepseek/deepseek-v4-pro",
     "deepseek/deepseek-v3.2-thinking",
     "alibaba/qwen3.7-max",
@@ -668,14 +646,12 @@ const PROVIDER_MODEL_PREFERENCES: Record<string, string[]> = {
     "zai/glm-5.1",
   ],
   xai: ["grok-4.7 ", "grok-4.7", "grok-4.6", "grok-4.5", "grok-4.3"],
-  xiaomi: ["mimo-v2.5-pro-ultraspeed", "mimo-v2.5-pro", "mimo-v2.5"],
-  "xiaomi-token-plan-ams": ["mimo-v2.5-pro", "mimo-v2.5"],
-  "xiaomi-token-plan-cn": ["mimo-v2.5-pro", "mimo-v2.5"],
-  "xiaomi-token-plan-sgp": ["mimo-v2.5-pro", "mimo-v2.5"],
+  xiaomi: ["mimo-v2.6-pro-ultraspeed", "mimo-v2.6-pro", "mimo-v2.6-flash"],
+  "xiaomi-token-plan-ams": ["mimo-v2.6-pro", "mimo-v2.6-flash"],
+  "xiaomi-token-plan-cn": ["mimo-v2.6-pro", "mimo-v2.6-flash"],
+  "xiaomi-token-plan-sgp": ["mimo-v2.6-pro", "mimo-v2.6-flash"],
   nvidia: [
     "nvidia/nemotron-3-ultra-550b-a55b",
-    "nvidia/nemotron-3-super-120b-a12b",
-    "moonshotai/kimi-k3",
     "moonshotai/kimi-k2.6",
     "nvidia/nemotron-3-nano-omni-30b-a3b-reasoning",
   ],
@@ -1944,8 +1920,7 @@ function createContrarianExtension(pi: ExtensionAPI, deps: ContrarianExtensionDe
           onUpdate?.(partial);
         };
 
-        let lastErrorResult: { ok: false; error: string; details: ContrarianDetails } | undefined;
-        for (let index = 0; index < attempts.length; index++) {
+        for (let index = 0; ; index++) {
           const previous = index > 0 ? attempts[index - 1] : undefined;
           const attempt = previous
             ? withFallbackReason(attempts[index], previous)
@@ -1969,7 +1944,6 @@ function createContrarianExtension(pi: ExtensionAPI, deps: ContrarianExtensionDe
             });
           }
 
-          lastErrorResult = result;
           const canFallBack = index < attempts.length - 1 && isModelAvailabilityError(result.error);
           if (!canFallBack) {
             return finalizeResult({
@@ -1978,30 +1952,6 @@ function createContrarianExtension(pi: ExtensionAPI, deps: ContrarianExtensionDe
             });
           }
         }
-
-        return finalizeResult({
-          content: [
-            {
-              type: "text",
-              text: lastErrorResult?.error ?? "Contrarian could not select an available model.",
-            },
-          ],
-          details: lastErrorResult?.details ?? {
-            modelRef: "",
-            provider: ctx.model?.provider ?? "unknown",
-            modelId: "",
-            modelName: undefined,
-            thinkingLevel: configuredThinkingLevel ?? DEFAULT_THINKING_LEVEL,
-            autoSelected: true,
-            selectionReason: "No authenticated models were available to run the contrarian.",
-            includeBash: params.includeBash ?? false,
-            usage: createEmptyUsage(),
-            stderr: "",
-            exitCode: 1,
-            durationMs: 0,
-            cwd: params.cwd ?? ctx.cwd,
-          },
-        });
       } finally {
         activeRuns.delete(toolCallId);
         updateContrarianUi(ctx, activeRuns);

package/extensions/contrarian/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@diegopetrucci/pi-contrarian",
-  "version": "0.1.16",
+  "version": "0.1.18",
   "description": "A read-only contrarian subagent extension for pi that stress-tests plans, designs, assumptions, and conclusions.",
   "keywords": [
     "pi-package",

package/extensions/dynamic-context-pruning/README.md

@@ -275,11 +275,21 @@ Notes:
   warm request prefix between consecutive LLM calls separated only by a
   toolResult) illustrates cache invalidation, not proof that amortization is
   impossible. Set `mid_loop` back to `22` to restore the old parity behavior.
+- `gate.breakEvenThreshold` is a deprecated legacy field retained for
+  compatibility. Loaded config always resolves through
+  `gate.breakEvenThresholdByState`, so a global-only value (including `0`) has
+  no runtime effect, and missing or invalid state values use their existing
+  normalized defaults rather than inheriting the global value. Direct callers
+  of the exported `resolveBreakEvenThreshold` may still use this field as the
+  fallback when they pass an unnormalized gate object. Set explicit per-state
+  values instead.
 - `gate.breakEvenThreshold`'s default of `22` is calibrated from the
   representative-corpus benchmark at `cachedPriceRatio` r=0.1 (aggressive
   prompt caching, the common case). It is **provider/ratio-
-  dependent**: the optimal threshold rises as caching gets weaker (T=29 at
-  r=0.25, T=54 at r=0.5, T=58 at r=0.9 on the same corpus).
+  dependent**. In the **historical, retired turn-end run**, the optimal
+  threshold rose as caching got weaker (T=29 at r=0.25, T=54 at r=0.5, T=58
+  at r=0.9); these are not current turn-start results. See
+  [`docs/v2-design.md` §4](./docs/v2-design.md#4-benchmark-decision-gate).
 
 ## Prompt-cache trade-off
 
@@ -326,6 +336,8 @@ Options:
 - `--limit N`: only process the first N session files found.
 - `--ratio R`: cached-price ratio(s) to model; repeatable and/or
   comma-separated (e.g. `--ratio 0.1,0.2`). Default: `0.1`.
+- `--sweep-max N`: explicit ceiling for the auto-expanding threshold sweep
+  (overrides the corpus-derived ceiling). Default: auto-expand from 30 as needed.
 - `--json`: emit a full machine-readable JSON dump instead of aligned text.
 - `--simulate-compression`: additionally SIMULATE v2-style range compression
   (docs/v2-design.md §1/§4 go/no-go evidence) alongside — never mixed into —
@@ -390,9 +402,11 @@ reports two kinds of numbers:
   prune point, and the *actual* net benefit that prune would have produced —
   `realizedNetBenefit = actualRemainingCalls * recurringSaving - penalty`.
 
-It also sweeps the break-even threshold `T` from 1 to 30 and reports the `T`
-that maximizes total realized net benefit across all gate-eligible
-candidates, both overall and split by idle/mid-loop agent state.
+It also sweeps the break-even threshold `T` from 1 to 30, then keeps doubling
+the range while the best `T` sits at the top edge, up to a corpus-derived
+ceiling (hard cap 500). `--sweep-max N` sets that ceiling explicitly. It
+reports the `T` that maximizes total realized net benefit across all
+gate-eligible candidates, both overall and split by idle/mid-loop agent state.
 
 ## Comparison
 
@@ -428,9 +442,11 @@ net benefit, at sufficient candidate volume, with enough remaining-calls
 runway per session to plausibly amortize a compression summary's own token
 cost — see `docs/v2-design.md` §4 for the exact go/no-go bar.
 
-The representative-corpus run of the benchmark harness (over
+The **historical, retired turn-end run** of the benchmark harness (over
 `~/.the-last-harness/agent/sessions`: 1,390+ session files, 556 gate-eligible
-candidates — see the corpus-choice note above) produced:
+candidates — see the corpus-choice note above) produced the following
+figures, preserved as historical context rather than current turn-start
+results (see [`docs/v2-design.md` §4](./docs/v2-design.md#4-benchmark-decision-gate)):
 
 - **Candidates and tokens removed by strategy**: `dedupe` 111 candidates /
   93,330 tokens removed; `error-purge` 187 / 28,191; `superseded-file-ops`
@@ -440,6 +456,7 @@ candidates — see the corpus-choice note above) produced:
   this corpus is dominated by long, tool-heavy agent sessions rather than
   short orchestrator sessions.
 
+The current turn-start re-derivation has 460 gate-eligible candidates.
 Shipped gate defaults remain `idle=22` / `mid_loop=1` (see
 [Configuration](#configuration)).