@pentoshi/clai 3.8.20 → 3.8.21

package/README.md

@@ -220,7 +220,7 @@ Coding and general sysadmin work use the same agent (scaffold, debug, packages)
 | Provider | Default model | Free tier | Key prefix |
 |----------|---------------|-----------|------------|
 | Groq | `llama-3.3-70b-versatile` | ✓ | `gsk_` |
-| Gemini | `gemini-2.0-flash` | ✓ | `AIza` |
+| Gemini | `gemini-2.0-flash` | ✓ | `AIza` / `AQ.` |
 | OpenRouter | `meta-llama/llama-3.3-70b-instruct:free` | ✓ | `sk-or-` |
 | OpenAI | `gpt-4o-mini` | — | `sk-` |
 | Anthropic | `claude-3-5-haiku-latest` | — | `sk-ant-` |
@@ -232,16 +232,19 @@ Coding and general sysadmin work use the same agent (scaffold, debug, packages)
 | Bynara | `mimo-v2.5-free` | ✓ | `sk_nry_` |
 
 ```sh
-clai set groq gsk_...
+clai set groq gsk_...              # store / append an API key (multi-key supported)
+clai set groq gsk_another_...      # add a second key for the same provider
 clai set gemini --from-env GEMINI_API_KEY
 echo "gsk_..." | clai set groq --stdin
 clai set ollama --url http://localhost:11434
-clai keys
+clai keys                          # lists each key (masked) + sticky active ★
 clai use groq
-clai provider          # picker shows selected model per provider
-clai unset groq
+clai provider                      # picker shows selected model per provider
+clai unset groq                    # removes ALL keys for that provider
 ```
 
+In the TUI, **`/set`** opens a multi-row key editor (like `/scope`): add with `+`, remove rows, **Save**, or **Reset all**. **`/unset`** clears every key for the chosen provider. When several keys are stored, clai **rotates** on rate limits / auth / transient errors (one retry per key, then the next key in a circle) and shows a non-stacking toast (`using groq · …ab12` / `switching …`).
+
 Env overrides: `GROQ_API_KEY`, `GEMINI_API_KEY`, `OPENROUTER_API_KEY`, `OPENAI_API_KEY`, `ANTHROPIC_API_KEY`, `NVIDIA_API_KEY`, `OLLAMA_HOST`, …
 
 ---

package/dist/agent/context-breakdown.d.ts

@@ -0,0 +1,37 @@
+/**
+ * Estimate how a model request's context is composed (system / history / tools).
+ * Metadata-only: no prompt text is stored — only role counts and token estimates.
+ *
+ * Used for reliability ops and regression harnesses. Does not truncate context.
+ */
+import type { ChatMessage, ToolDefinition } from "../types.js";
+export interface ContextBreakdown {
+    /** Rough total estimated tokens for messages (+ tools if provided). */
+    readonly estimatedTotalTokens: number;
+    readonly systemTokens: number;
+    readonly userTokens: number;
+    readonly assistantTokens: number;
+    readonly toolResultTokens: number;
+    readonly toolSchemaTokens: number;
+    readonly messageCount: number;
+    readonly systemMessageCount: number;
+    readonly userMessageCount: number;
+    readonly assistantMessageCount: number;
+    readonly toolMessageCount: number;
+    /** High-level system subsections when detected by prefix. */
+    readonly systemParts: {
+        readonly constitutionTokens: number;
+        readonly planTokens: number;
+        readonly scopeTokens: number;
+        readonly sessionStateTokens: number;
+        readonly compactionMemoryTokens: number;
+        readonly otherSystemTokens: number;
+    };
+    readonly toolDefinitionCount: number;
+}
+/**
+ * Build a metadata-only breakdown of an assembled completion request.
+ */
+export declare function buildContextBreakdown(messages: readonly ChatMessage[], tools?: readonly ToolDefinition[] | undefined): ContextBreakdown;
+/** Flatten for audit/diagnostic logs (numeric keys only — no content). */
+export declare function contextBreakdownAuditPayload(breakdown: ContextBreakdown): Record<string, number>;

package/dist/agent/context-breakdown.js

@@ -0,0 +1,145 @@
+/**
+ * Estimate how a model request's context is composed (system / history / tools).
+ * Metadata-only: no prompt text is stored — only role counts and token estimates.
+ *
+ * Used for reliability ops and regression harnesses. Does not truncate context.
+ */
+import { estimateMessagesTokens, estimateTokens, isCompactionMemoryMessage, COMPACTION_MEMORY_PREFIX, PLAN_IMPLEMENT_MEMORY_PREFIX, MECHANICAL_MEMORY_PREFIX, } from "./context-manager.js";
+import { SESSION_STATE_PREFIX } from "./session-state.js";
+function isPlanSystem(content) {
+    return content.startsWith("ACTIVE PLAN") || content.includes("\nACTIVE PLAN");
+}
+function isScopeSystem(content) {
+    return (content.startsWith("ENGAGEMENT SCOPE") ||
+        content.includes("\nENGAGEMENT SCOPE"));
+}
+function isSessionStateSystem(content) {
+    return (content.startsWith(SESSION_STATE_PREFIX) ||
+        content.includes(SESSION_STATE_PREFIX));
+}
+function isCompactionSystem(content) {
+    return (content.startsWith(COMPACTION_MEMORY_PREFIX) ||
+        content.startsWith(PLAN_IMPLEMENT_MEMORY_PREFIX) ||
+        content.startsWith(MECHANICAL_MEMORY_PREFIX) ||
+        // Message-shaped helper when role is system.
+        isCompactionMemoryMessage({ role: "system", content }));
+}
+/**
+ * Build a metadata-only breakdown of an assembled completion request.
+ */
+export function buildContextBreakdown(messages, tools) {
+    let systemTokens = 0;
+    let userTokens = 0;
+    let assistantTokens = 0;
+    let toolResultTokens = 0;
+    let systemMessageCount = 0;
+    let userMessageCount = 0;
+    let assistantMessageCount = 0;
+    let toolMessageCount = 0;
+    let constitutionTokens = 0;
+    let planTokens = 0;
+    let scopeTokens = 0;
+    let sessionStateTokens = 0;
+    let compactionMemoryTokens = 0;
+    let otherSystemTokens = 0;
+    for (const message of messages) {
+        const tokens = estimateTokens(message.content) + 4;
+        switch (message.role) {
+            case "system": {
+                systemMessageCount += 1;
+                systemTokens += tokens;
+                if (isCompactionSystem(message.content)) {
+                    compactionMemoryTokens += tokens;
+                }
+                else if (isPlanSystem(message.content)) {
+                    planTokens += tokens;
+                }
+                else if (isScopeSystem(message.content)) {
+                    scopeTokens += tokens;
+                }
+                else if (isSessionStateSystem(message.content)) {
+                    sessionStateTokens += tokens;
+                }
+                else if (systemMessageCount === 1 ||
+                    /CURRENT MODE:|OUTCOME CONTRACT|# ROLE|# HONESTY/i.test(message.content)) {
+                    // Primary system / composed constitution (often message[0]).
+                    constitutionTokens += tokens;
+                }
+                else {
+                    otherSystemTokens += tokens;
+                }
+                break;
+            }
+            case "user":
+                userMessageCount += 1;
+                userTokens += tokens;
+                break;
+            case "assistant":
+                assistantMessageCount += 1;
+                assistantTokens += tokens;
+                if (message.toolCalls?.length) {
+                    assistantTokens += estimateTokens(JSON.stringify(message.toolCalls));
+                }
+                break;
+            case "tool":
+                toolMessageCount += 1;
+                toolResultTokens += tokens;
+                break;
+            default:
+                otherSystemTokens += tokens;
+                break;
+        }
+    }
+    let toolSchemaTokens = 0;
+    const toolDefinitionCount = tools?.length ?? 0;
+    if (tools?.length) {
+        toolSchemaTokens = estimateTokens(JSON.stringify(tools));
+    }
+    const estimatedTotalTokens = estimateMessagesTokens([...messages]) + toolSchemaTokens;
+    return {
+        estimatedTotalTokens,
+        systemTokens,
+        userTokens,
+        assistantTokens,
+        toolResultTokens,
+        toolSchemaTokens,
+        messageCount: messages.length,
+        systemMessageCount,
+        userMessageCount,
+        assistantMessageCount,
+        toolMessageCount,
+        systemParts: {
+            constitutionTokens,
+            planTokens,
+            scopeTokens,
+            sessionStateTokens,
+            compactionMemoryTokens,
+            otherSystemTokens,
+        },
+        toolDefinitionCount,
+    };
+}
+/** Flatten for audit/diagnostic logs (numeric keys only — no content). */
+export function contextBreakdownAuditPayload(breakdown) {
+    return {
+        estimatedTotalTokens: breakdown.estimatedTotalTokens,
+        systemTokens: breakdown.systemTokens,
+        userTokens: breakdown.userTokens,
+        assistantTokens: breakdown.assistantTokens,
+        toolResultTokens: breakdown.toolResultTokens,
+        toolSchemaTokens: breakdown.toolSchemaTokens,
+        messageCount: breakdown.messageCount,
+        systemMessageCount: breakdown.systemMessageCount,
+        userMessageCount: breakdown.userMessageCount,
+        assistantMessageCount: breakdown.assistantMessageCount,
+        toolMessageCount: breakdown.toolMessageCount,
+        constitutionTokens: breakdown.systemParts.constitutionTokens,
+        planTokens: breakdown.systemParts.planTokens,
+        scopeTokens: breakdown.systemParts.scopeTokens,
+        sessionStateTokens: breakdown.systemParts.sessionStateTokens,
+        compactionMemoryTokens: breakdown.systemParts.compactionMemoryTokens,
+        otherSystemTokens: breakdown.systemParts.otherSystemTokens,
+        toolDefinitionCount: breakdown.toolDefinitionCount,
+    };
+}
+//# sourceMappingURL=context-breakdown.js.map

package/dist/agent/context-breakdown.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"context-breakdown.js","sourceRoot":"","sources":["../../src/agent/context-breakdown.ts"],"names":[],"mappings":"AAAA;;;;;GAKG;AAGH,OAAO,EACL,sBAAsB,EACtB,cAAc,EACd,yBAAyB,EACzB,wBAAwB,EACxB,4BAA4B,EAC5B,wBAAwB,GACzB,MAAM,sBAAsB,CAAC;AAC9B,OAAO,EAAE,oBAAoB,EAAE,MAAM,oBAAoB,CAAC;AA2B1D,SAAS,YAAY,CAAC,OAAe;IACnC,OAAO,OAAO,CAAC,UAAU,CAAC,aAAa,CAAC,IAAI,OAAO,CAAC,QAAQ,CAAC,eAAe,CAAC,CAAC;AAChF,CAAC;AAED,SAAS,aAAa,CAAC,OAAe;IACpC,OAAO,CACL,OAAO,CAAC,UAAU,CAAC,kBAAkB,CAAC;QACtC,OAAO,CAAC,QAAQ,CAAC,oBAAoB,CAAC,CACvC,CAAC;AACJ,CAAC;AAED,SAAS,oBAAoB,CAAC,OAAe;IAC3C,OAAO,CACL,OAAO,CAAC,UAAU,CAAC,oBAAoB,CAAC;QACxC,OAAO,CAAC,QAAQ,CAAC,oBAAoB,CAAC,CACvC,CAAC;AACJ,CAAC;AAED,SAAS,kBAAkB,CAAC,OAAe;IACzC,OAAO,CACL,OAAO,CAAC,UAAU,CAAC,wBAAwB,CAAC;QAC5C,OAAO,CAAC,UAAU,CAAC,4BAA4B,CAAC;QAChD,OAAO,CAAC,UAAU,CAAC,wBAAwB,CAAC;QAC5C,6CAA6C;QAC7C,yBAAyB,CAAC,EAAE,IAAI,EAAE,QAAQ,EAAE,OAAO,EAAE,CAAC,CACvD,CAAC;AACJ,CAAC;AAED;;GAEG;AACH,MAAM,UAAU,qBAAqB,CACnC,QAAgC,EAChC,KAA6C;IAE7C,IAAI,YAAY,GAAG,CAAC,CAAC;IACrB,IAAI,UAAU,GAAG,CAAC,CAAC;IACnB,IAAI,eAAe,GAAG,CAAC,CAAC;IACxB,IAAI,gBAAgB,GAAG,CAAC,CAAC;IACzB,IAAI,kBAAkB,GAAG,CAAC,CAAC;IAC3B,IAAI,gBAAgB,GAAG,CAAC,CAAC;IACzB,IAAI,qBAAqB,GAAG,CAAC,CAAC;IAC9B,IAAI,gBAAgB,GAAG,CAAC,CAAC;IAEzB,IAAI,kBAAkB,GAAG,CAAC,CAAC;IAC3B,IAAI,UAAU,GAAG,CAAC,CAAC;IACnB,IAAI,WAAW,GAAG,CAAC,CAAC;IACpB,IAAI,kBAAkB,GAAG,CAAC,CAAC;IAC3B,IAAI,sBAAsB,GAAG,CAAC,CAAC;IAC/B,IAAI,iBAAiB,GAAG,CAAC,CAAC;IAE1B,KAAK,MAAM,OAAO,IAAI,QAAQ,EAAE,CAAC;QAC/B,MAAM,MAAM,GAAG,cAAc,CAAC,OAAO,CAAC,OAAO,CAAC,GAAG,CAAC,CAAC;QACnD,QAAQ,OAAO,CAAC,IAAI,EAAE,CAAC;YACrB,KAAK,QAAQ,CAAC,CAAC,CAAC;gBACd,kBAAkB,IAAI,CAAC,CAAC;gBACxB,YAAY,IAAI,MAAM,CAAC;gBACvB,IAAI,kBAAkB,CAAC,OAAO,CAAC,OAAO,CAAC,EAAE,CAAC;oBACxC,sBAAsB,IAAI,MAAM,CAAC;gBACnC,CAAC;qBAAM,IAAI,YAAY,CAAC,OAAO,CAAC,OAAO,CAAC,EAAE,CAAC;oBACzC,UAAU,IAAI,MAAM,CAAC;gBACvB,CAAC;qBAAM,IAAI,aAAa,CAAC,OAAO,CAAC,OAAO,CAAC,EAAE,CAAC;oBAC1C,WAAW,IAAI,MAAM,CAAC;gBACxB,CAAC;qBAAM,IAAI,oBAAoB,CAAC,OAAO,CAAC,OAAO,CAAC,EAAE,CAAC;oBACjD,kBAAkB,IAAI,MAAM,CAAC;gBAC/B,CAAC;qBAAM,IACL,kBAAkB,KAAK,CAAC;oBACxB,kDAAkD,CAAC,IAAI,CACrD,OAAO,CAAC,OAAO,CAChB,EACD,CAAC;oBACD,6DAA6D;oBAC7D,kBAAkB,IAAI,MAAM,CAAC;gBAC/B,CAAC;qBAAM,CAAC;oBACN,iBAAiB,IAAI,MAAM,CAAC;gBAC9B,CAAC;gBACD,MAAM;YACR,CAAC;YACD,KAAK,MAAM;gBACT,gBAAgB,IAAI,CAAC,CAAC;gBACtB,UAAU,IAAI,MAAM,CAAC;gBACrB,MAAM;YACR,KAAK,WAAW;gBACd,qBAAqB,IAAI,CAAC,CAAC;gBAC3B,eAAe,IAAI,MAAM,CAAC;gBAC1B,IAAI,OAAO,CAAC,SAAS,EAAE,MAAM,EAAE,CAAC;oBAC9B,eAAe,IAAI,cAAc,CAC/B,IAAI,CAAC,SAAS,CAAC,OAAO,CAAC,SAAS,CAAC,CAClC,CAAC;gBACJ,CAAC;gBACD,MAAM;YACR,KAAK,MAAM;gBACT,gBAAgB,IAAI,CAAC,CAAC;gBACtB,gBAAgB,IAAI,MAAM,CAAC;gBAC3B,MAAM;YACR;gBACE,iBAAiB,IAAI,MAAM,CAAC;gBAC5B,MAAM;QACV,CAAC;IACH,CAAC;IAED,IAAI,gBAAgB,GAAG,CAAC,CAAC;IACzB,MAAM,mBAAmB,GAAG,KAAK,EAAE,MAAM,IAAI,CAAC,CAAC;IAC/C,IAAI,KAAK,EAAE,MAAM,EAAE,CAAC;QAClB,gBAAgB,GAAG,cAAc,CAAC,IAAI,CAAC,SAAS,CAAC,KAAK,CAAC,CAAC,CAAC;IAC3D,CAAC;IAED,MAAM,oBAAoB,GACxB,sBAAsB,CAAC,CAAC,GAAG,QAAQ,CAAC,CAAC,GAAG,gBAAgB,CAAC;IAE3D,OAAO;QACL,oBAAoB;QACpB,YAAY;QACZ,UAAU;QACV,eAAe;QACf,gBAAgB;QAChB,gBAAgB;QAChB,YAAY,EAAE,QAAQ,CAAC,MAAM;QAC7B,kBAAkB;QAClB,gBAAgB;QAChB,qBAAqB;QACrB,gBAAgB;QAChB,WAAW,EAAE;YACX,kBAAkB;YAClB,UAAU;YACV,WAAW;YACX,kBAAkB;YAClB,sBAAsB;YACtB,iBAAiB;SAClB;QACD,mBAAmB;KACpB,CAAC;AACJ,CAAC;AAED,0EAA0E;AAC1E,MAAM,UAAU,4BAA4B,CAC1C,SAA2B;IAE3B,OAAO;QACL,oBAAoB,EAAE,SAAS,CAAC,oBAAoB;QACpD,YAAY,EAAE,SAAS,CAAC,YAAY;QACpC,UAAU,EAAE,SAAS,CAAC,UAAU;QAChC,eAAe,EAAE,SAAS,CAAC,eAAe;QAC1C,gBAAgB,EAAE,SAAS,CAAC,gBAAgB;QAC5C,gBAAgB,EAAE,SAAS,CAAC,gBAAgB;QAC5C,YAAY,EAAE,SAAS,CAAC,YAAY;QACpC,kBAAkB,EAAE,SAAS,CAAC,kBAAkB;QAChD,gBAAgB,EAAE,SAAS,CAAC,gBAAgB;QAC5C,qBAAqB,EAAE,SAAS,CAAC,qBAAqB;QACtD,gBAAgB,EAAE,SAAS,CAAC,gBAAgB;QAC5C,kBAAkB,EAAE,SAAS,CAAC,WAAW,CAAC,kBAAkB;QAC5D,UAAU,EAAE,SAAS,CAAC,WAAW,CAAC,UAAU;QAC5C,WAAW,EAAE,SAAS,CAAC,WAAW,CAAC,WAAW;QAC9C,kBAAkB,EAAE,SAAS,CAAC,WAAW,CAAC,kBAAkB;QAC5D,sBAAsB,EAAE,SAAS,CAAC,WAAW,CAAC,sBAAsB;QACpE,iBAAiB,EAAE,SAAS,CAAC,WAAW,CAAC,iBAAiB;QAC1D,mBAAmB,EAAE,SAAS,CAAC,mBAAmB;KACnD,CAAC;AACJ,CAAC"}

package/dist/agent/reliability-policy.d.ts

@@ -0,0 +1,85 @@
+/**
+ * Reliability experiments from the architecture audit (E1–E6).
+ *
+ * Defaults are capability-preserving: they reduce unnecessary context / fragile
+ * free-tier overload without hard-truncating critical evidence or blocking free
+ * users. Every knob is config-backed and can be disabled independently.
+ */
+import type { ProviderId } from "../types.js";
+/** Hard auto-compact ceiling — never raise soft past this. */
+export declare const HARD_COMPACT_TOKEN_BUDGET = 100000;
+/** E1 default: start compact earlier than the hard ceiling. */
+export declare const DEFAULT_SOFT_COMPACT_TOKEN_BUDGET = 70000;
+/** E2 default fs passthrough (was 400k). Full body always on disk when truncated. */
+export declare const DEFAULT_FS_PASSTHROUGH_CAP_CHARS = 64000;
+/** E4: warn when free-cloud context is this large. */
+export declare const DEFAULT_FREE_TIER_WARN_TOKENS = 40000;
+/** E4: stronger notice after this many consecutive free-tier stream failures. */
+export declare const DEFAULT_FREE_TIER_FAIL_THRESHOLD = 2;
+/** E3 floors — large enough for fs.write / salvage, lower than prior 32k default. */
+export declare const ADAPTIVE_MAX_TOKENS_TOOL_STEP = 24576;
+export declare const ADAPTIVE_MAX_TOKENS_LIGHT = 12288;
+export declare const ADAPTIVE_MAX_TOKENS_FLOOR = 8192;
+export declare const LEGACY_MAX_TOKENS = 32768;
+export interface ReliabilityPolicy {
+    readonly softEarlyCompact: boolean;
+    readonly softCompactTokenBudget: number;
+    readonly hardCompactTokenBudget: number;
+    readonly fsPassthroughCapChars: number;
+    readonly adaptiveMaxTokens: boolean;
+    readonly freeTierContextGuard: boolean;
+    readonly freeTierWarnTokens: number;
+    readonly freeTierFailThreshold: number;
+    readonly toolResultDedup: boolean;
+    readonly slimNativePrompt: boolean;
+}
+/** Resolve effective policy from config + optional env overrides. */
+export declare function getReliabilityPolicy(): ReliabilityPolicy;
+/**
+ * E1: token threshold that triggers auto-compact.
+ * Soft path fires earlier; hard budget is always the ceiling used when soft is off.
+ */
+export declare function autoCompactTriggerTokens(policy?: ReliabilityPolicy): number;
+/**
+ * E3: max completion tokens for a model step.
+ * Tool-heavy steps keep room for large writes; light steps use a lower cap.
+ * Never goes below {@link ADAPTIVE_MAX_TOKENS_FLOOR}.
+ */
+export declare function resolveStepMaxTokens(input: {
+    nativeToolsActive: boolean;
+    toolsAttached: boolean;
+    recoveryNudge?: boolean | undefined;
+    policy?: ReliabilityPolicy | undefined;
+}): number;
+export declare function isFreeCloudProvider(provider: ProviderId): boolean;
+/**
+ * E4: advisory notices for free-tier + large context / repeated failures.
+ * Never blocks the request.
+ */
+export declare function freeTierGuardNotices(input: {
+    provider: ProviderId;
+    estimatedInputTokens: number;
+    consecutiveFailures: number;
+    policy?: ReliabilityPolicy | undefined;
+}): string[];
+/** Stable hash for E5 tool-result dedup (content only). */
+export declare function hashToolResultContent(content: string): string;
+/**
+ * E5: if identical tool output already appeared this turn, replace with a
+ * pointer so history does not grow with duplicate dumps. Full text remains
+ * available via artifact path when present.
+ */
+export declare function dedupeToolContextOutput(input: {
+    content: string;
+    toolName: string;
+    artifactPath?: string | undefined;
+    seenHashes: Map<string, {
+        toolName: string;
+        count: number;
+    }>;
+    policy?: ReliabilityPolicy | undefined;
+}): {
+    content: string;
+    deduped: boolean;
+    hash: string;
+};

package/dist/agent/reliability-policy.js

@@ -0,0 +1,158 @@
+/**
+ * Reliability experiments from the architecture audit (E1–E6).
+ *
+ * Defaults are capability-preserving: they reduce unnecessary context / fragile
+ * free-tier overload without hard-truncating critical evidence or blocking free
+ * users. Every knob is config-backed and can be disabled independently.
+ */
+import { createHash } from "node:crypto";
+import { getConfig, providerCategory } from "../store/config.js";
+import { AUTO_COMPACT_TOKEN_BUDGET } from "./context-manager.js";
+/** Hard auto-compact ceiling — never raise soft past this. */
+export const HARD_COMPACT_TOKEN_BUDGET = AUTO_COMPACT_TOKEN_BUDGET;
+/** E1 default: start compact earlier than the hard ceiling. */
+export const DEFAULT_SOFT_COMPACT_TOKEN_BUDGET = 70_000;
+/** E2 default fs passthrough (was 400k). Full body always on disk when truncated. */
+export const DEFAULT_FS_PASSTHROUGH_CAP_CHARS = 64_000;
+/** E4: warn when free-cloud context is this large. */
+export const DEFAULT_FREE_TIER_WARN_TOKENS = 40_000;
+/** E4: stronger notice after this many consecutive free-tier stream failures. */
+export const DEFAULT_FREE_TIER_FAIL_THRESHOLD = 2;
+/** E3 floors — large enough for fs.write / salvage, lower than prior 32k default. */
+export const ADAPTIVE_MAX_TOKENS_TOOL_STEP = 24_576;
+export const ADAPTIVE_MAX_TOKENS_LIGHT = 12_288;
+export const ADAPTIVE_MAX_TOKENS_FLOOR = 8_192;
+export const LEGACY_MAX_TOKENS = 32_768;
+function boolEnv(name) {
+    const v = process.env[name]?.trim().toLowerCase();
+    if (v === undefined || v === "")
+        return undefined;
+    if (v === "1" || v === "true" || v === "yes" || v === "on")
+        return true;
+    if (v === "0" || v === "false" || v === "no" || v === "off")
+        return false;
+    return undefined;
+}
+function intEnv(name) {
+    const v = process.env[name]?.trim();
+    if (!v)
+        return undefined;
+    const n = Number.parseInt(v, 10);
+    return Number.isFinite(n) ? n : undefined;
+}
+/** Resolve effective policy from config + optional env overrides. */
+export function getReliabilityPolicy() {
+    const cfg = getConfig();
+    const softEarly = boolEnv("CLAI_SOFT_EARLY_COMPACT") ?? cfg.softEarlyCompact ?? true;
+    let softBudget = intEnv("CLAI_SOFT_COMPACT_TOKENS") ??
+        cfg.softCompactTokenBudget ??
+        DEFAULT_SOFT_COMPACT_TOKEN_BUDGET;
+    softBudget = Math.max(20_000, Math.min(softBudget, HARD_COMPACT_TOKEN_BUDGET));
+    let fsCap = intEnv("CLAI_FS_PASSTHROUGH_CHARS") ??
+        cfg.fsPassthroughCapChars ??
+        DEFAULT_FS_PASSTHROUGH_CAP_CHARS;
+    fsCap = Math.max(8_000, Math.min(fsCap, 400_000));
+    return {
+        softEarlyCompact: softEarly,
+        softCompactTokenBudget: softBudget,
+        hardCompactTokenBudget: HARD_COMPACT_TOKEN_BUDGET,
+        fsPassthroughCapChars: fsCap,
+        adaptiveMaxTokens: boolEnv("CLAI_ADAPTIVE_MAX_TOKENS") ?? cfg.adaptiveMaxTokens ?? true,
+        freeTierContextGuard: boolEnv("CLAI_FREE_TIER_GUARD") ?? cfg.freeTierContextGuard ?? true,
+        freeTierWarnTokens: intEnv("CLAI_FREE_TIER_WARN_TOKENS") ??
+            cfg.freeTierWarnTokens ??
+            DEFAULT_FREE_TIER_WARN_TOKENS,
+        freeTierFailThreshold: intEnv("CLAI_FREE_TIER_FAIL_THRESHOLD") ??
+            cfg.freeTierFailThreshold ??
+            DEFAULT_FREE_TIER_FAIL_THRESHOLD,
+        toolResultDedup: boolEnv("CLAI_TOOL_RESULT_DEDUP") ?? cfg.toolResultDedup ?? true,
+        slimNativePrompt: boolEnv("CLAI_SLIM_NATIVE_PROMPT") ?? cfg.slimNativePrompt ?? true,
+    };
+}
+/**
+ * E1: token threshold that triggers auto-compact.
+ * Soft path fires earlier; hard budget is always the ceiling used when soft is off.
+ */
+export function autoCompactTriggerTokens(policy = getReliabilityPolicy()) {
+    if (!policy.softEarlyCompact)
+        return policy.hardCompactTokenBudget;
+    return Math.min(policy.softCompactTokenBudget, policy.hardCompactTokenBudget);
+}
+/**
+ * E3: max completion tokens for a model step.
+ * Tool-heavy steps keep room for large writes; light steps use a lower cap.
+ * Never goes below {@link ADAPTIVE_MAX_TOKENS_FLOOR}.
+ */
+export function resolveStepMaxTokens(input) {
+    const policy = input.policy ?? getReliabilityPolicy();
+    if (!policy.adaptiveMaxTokens)
+        return LEGACY_MAX_TOKENS;
+    if (input.recoveryNudge) {
+        return Math.max(ADAPTIVE_MAX_TOKENS_FLOOR, ADAPTIVE_MAX_TOKENS_LIGHT);
+    }
+    if (input.toolsAttached || input.nativeToolsActive) {
+        return ADAPTIVE_MAX_TOKENS_TOOL_STEP;
+    }
+    return ADAPTIVE_MAX_TOKENS_LIGHT;
+}
+export function isFreeCloudProvider(provider) {
+    return providerCategory[provider] === "free-cloud";
+}
+/**
+ * E4: advisory notices for free-tier + large context / repeated failures.
+ * Never blocks the request.
+ */
+export function freeTierGuardNotices(input) {
+    const policy = input.policy ?? getReliabilityPolicy();
+    if (!policy.freeTierContextGuard)
+        return [];
+    if (!isFreeCloudProvider(input.provider))
+        return [];
+    const notices = [];
+    if (input.estimatedInputTokens >= policy.freeTierWarnTokens) {
+        notices.push(`Large context (~${input.estimatedInputTokens.toLocaleString()} tokens) on free-cloud model — disconnects and empty admissions are more common. Prefer /compact, a paid/local model, or shorter turns if this fails.`);
+    }
+    if (input.consecutiveFailures >= policy.freeTierFailThreshold) {
+        notices.push(`Free-cloud model failed ${input.consecutiveFailures} time(s) this turn. Switch with /model or /provider, or enable providerFallback for automatic recovery.`);
+    }
+    return notices;
+}
+/** Stable hash for E5 tool-result dedup (content only). */
+export function hashToolResultContent(content) {
+    return createHash("sha256").update(content).digest("hex").slice(0, 16);
+}
+/**
+ * E5: if identical tool output already appeared this turn, replace with a
+ * pointer so history does not grow with duplicate dumps. Full text remains
+ * available via artifact path when present.
+ */
+export function dedupeToolContextOutput(input) {
+    const policy = input.policy ?? getReliabilityPolicy();
+    const hash = hashToolResultContent(input.content);
+    if (!policy.toolResultDedup) {
+        return { content: input.content, deduped: false, hash };
+    }
+    // Tiny results are not worth pointer indirection.
+    if (input.content.length < 400) {
+        input.seenHashes.set(hash, {
+            toolName: input.toolName,
+            count: (input.seenHashes.get(hash)?.count ?? 0) + 1,
+        });
+        return { content: input.content, deduped: false, hash };
+    }
+    const prior = input.seenHashes.get(hash);
+    if (!prior) {
+        input.seenHashes.set(hash, { toolName: input.toolName, count: 1 });
+        return { content: input.content, deduped: false, hash };
+    }
+    prior.count += 1;
+    const artifact = input.artifactPath
+        ? ` Full output: ${input.artifactPath}`
+        : "";
+    const head = input.content.slice(0, 200).replace(/\s+/g, " ").trim();
+    const content = `[duplicate tool output — identical to earlier ${prior.toolName} this turn; hash=${hash}]${artifact}\n` +
+        `Preview: ${head}${input.content.length > 200 ? "…" : ""}\n` +
+        `Re-read the prior tool result or artifact if you need the full body; do not assume it changed.`;
+    return { content, deduped: true, hash };
+}
+//# sourceMappingURL=reliability-policy.js.map

package/dist/agent/reliability-policy.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"reliability-policy.js","sourceRoot":"","sources":["../../src/agent/reliability-policy.ts"],"names":[],"mappings":"AAAA;;;;;;GAMG;AAEH,OAAO,EAAE,UAAU,EAAE,MAAM,aAAa,CAAC;AACzC,OAAO,EAAE,SAAS,EAAE,gBAAgB,EAAE,MAAM,oBAAoB,CAAC;AAEjE,OAAO,EAAE,yBAAyB,EAAE,MAAM,sBAAsB,CAAC;AAEjE,8DAA8D;AAC9D,MAAM,CAAC,MAAM,yBAAyB,GAAG,yBAAyB,CAAC;AAEnE,+DAA+D;AAC/D,MAAM,CAAC,MAAM,iCAAiC,GAAG,MAAM,CAAC;AAExD,qFAAqF;AACrF,MAAM,CAAC,MAAM,gCAAgC,GAAG,MAAM,CAAC;AAEvD,sDAAsD;AACtD,MAAM,CAAC,MAAM,6BAA6B,GAAG,MAAM,CAAC;AAEpD,iFAAiF;AACjF,MAAM,CAAC,MAAM,gCAAgC,GAAG,CAAC,CAAC;AAElD,qFAAqF;AACrF,MAAM,CAAC,MAAM,6BAA6B,GAAG,MAAM,CAAC;AACpD,MAAM,CAAC,MAAM,yBAAyB,GAAG,MAAM,CAAC;AAChD,MAAM,CAAC,MAAM,yBAAyB,GAAG,KAAK,CAAC;AAC/C,MAAM,CAAC,MAAM,iBAAiB,GAAG,MAAM,CAAC;AAexC,SAAS,OAAO,CAAC,IAAY;IAC3B,MAAM,CAAC,GAAG,OAAO,CAAC,GAAG,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,CAAC,WAAW,EAAE,CAAC;IAClD,IAAI,CAAC,KAAK,SAAS,IAAI,CAAC,KAAK,EAAE;QAAE,OAAO,SAAS,CAAC;IAClD,IAAI,CAAC,KAAK,GAAG,IAAI,CAAC,KAAK,MAAM,IAAI,CAAC,KAAK,KAAK,IAAI,CAAC,KAAK,IAAI;QAAE,OAAO,IAAI,CAAC;IACxE,IAAI,CAAC,KAAK,GAAG,IAAI,CAAC,KAAK,OAAO,IAAI,CAAC,KAAK,IAAI,IAAI,CAAC,KAAK,KAAK;QAAE,OAAO,KAAK,CAAC;IAC1E,OAAO,SAAS,CAAC;AACnB,CAAC;AAED,SAAS,MAAM,CAAC,IAAY;IAC1B,MAAM,CAAC,GAAG,OAAO,CAAC,GAAG,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,CAAC;IACpC,IAAI,CAAC,CAAC;QAAE,OAAO,SAAS,CAAC;IACzB,MAAM,CAAC,GAAG,MAAM,CAAC,QAAQ,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC;IACjC,OAAO,MAAM,CAAC,QAAQ,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC;AAC5C,CAAC;AAED,qEAAqE;AACrE,MAAM,UAAU,oBAAoB;IAClC,MAAM,GAAG,GAAG,SAAS,EAAE,CAAC;IACxB,MAAM,SAAS,GACb,OAAO,CAAC,yBAAyB,CAAC,IAAI,GAAG,CAAC,gBAAgB,IAAI,IAAI,CAAC;IACrE,IAAI,UAAU,GACZ,MAAM,CAAC,0BAA0B,CAAC;QAClC,GAAG,CAAC,sBAAsB;QAC1B,iCAAiC,CAAC;IACpC,UAAU,GAAG,IAAI,CAAC,GAAG,CAAC,MAAM,EAAE,IAAI,CAAC,GAAG,CAAC,UAAU,EAAE,yBAAyB,CAAC,CAAC,CAAC;IAE/E,IAAI,KAAK,GACP,MAAM,CAAC,2BAA2B,CAAC;QACnC,GAAG,CAAC,qBAAqB;QACzB,gCAAgC,CAAC;IACnC,KAAK,GAAG,IAAI,CAAC,GAAG,CAAC,KAAK,EAAE,IAAI,CAAC,GAAG,CAAC,KAAK,EAAE,OAAO,CAAC,CAAC,CAAC;IAElD,OAAO;QACL,gBAAgB,EAAE,SAAS;QAC3B,sBAAsB,EAAE,UAAU;QAClC,sBAAsB,EAAE,yBAAyB;QACjD,qBAAqB,EAAE,KAAK;QAC5B,iBAAiB,EACf,OAAO,CAAC,0BAA0B,CAAC,IAAI,GAAG,CAAC,iBAAiB,IAAI,IAAI;QACtE,oBAAoB,EAClB,OAAO,CAAC,sBAAsB,CAAC,IAAI,GAAG,CAAC,oBAAoB,IAAI,IAAI;QACrE,kBAAkB,EAChB,MAAM,CAAC,4BAA4B,CAAC;YACpC,GAAG,CAAC,kBAAkB;YACtB,6BAA6B;QAC/B,qBAAqB,EACnB,MAAM,CAAC,+BAA+B,CAAC;YACvC,GAAG,CAAC,qBAAqB;YACzB,gCAAgC;QAClC,eAAe,EACb,OAAO,CAAC,wBAAwB,CAAC,IAAI,GAAG,CAAC,eAAe,IAAI,IAAI;QAClE,gBAAgB,EACd,OAAO,CAAC,yBAAyB,CAAC,IAAI,GAAG,CAAC,gBAAgB,IAAI,IAAI;KACrE,CAAC;AACJ,CAAC;AAED;;;GAGG;AACH,MAAM,UAAU,wBAAwB,CAAC,MAAM,GAAG,oBAAoB,EAAE;IACtE,IAAI,CAAC,MAAM,CAAC,gBAAgB;QAAE,OAAO,MAAM,CAAC,sBAAsB,CAAC;IACnE,OAAO,IAAI,CAAC,GAAG,CACb,MAAM,CAAC,sBAAsB,EAC7B,MAAM,CAAC,sBAAsB,CAC9B,CAAC;AACJ,CAAC;AAED;;;;GAIG;AACH,MAAM,UAAU,oBAAoB,CAAC,KAKpC;IACC,MAAM,MAAM,GAAG,KAAK,CAAC,MAAM,IAAI,oBAAoB,EAAE,CAAC;IACtD,IAAI,CAAC,MAAM,CAAC,iBAAiB;QAAE,OAAO,iBAAiB,CAAC;IACxD,IAAI,KAAK,CAAC,aAAa,EAAE,CAAC;QACxB,OAAO,IAAI,CAAC,GAAG,CAAC,yBAAyB,EAAE,yBAAyB,CAAC,CAAC;IACxE,CAAC;IACD,IAAI,KAAK,CAAC,aAAa,IAAI,KAAK,CAAC,iBAAiB,EAAE,CAAC;QACnD,OAAO,6BAA6B,CAAC;IACvC,CAAC;IACD,OAAO,yBAAyB,CAAC;AACnC,CAAC;AAED,MAAM,UAAU,mBAAmB,CAAC,QAAoB;IACtD,OAAO,gBAAgB,CAAC,QAAQ,CAAC,KAAK,YAAY,CAAC;AACrD,CAAC;AAED;;;GAGG;AACH,MAAM,UAAU,oBAAoB,CAAC,KAKpC;IACC,MAAM,MAAM,GAAG,KAAK,CAAC,MAAM,IAAI,oBAAoB,EAAE,CAAC;IACtD,IAAI,CAAC,MAAM,CAAC,oBAAoB;QAAE,OAAO,EAAE,CAAC;IAC5C,IAAI,CAAC,mBAAmB,CAAC,KAAK,CAAC,QAAQ,CAAC;QAAE,OAAO,EAAE,CAAC;IACpD,MAAM,OAAO,GAAa,EAAE,CAAC;IAC7B,IAAI,KAAK,CAAC,oBAAoB,IAAI,MAAM,CAAC,kBAAkB,EAAE,CAAC;QAC5D,OAAO,CAAC,IAAI,CACV,mBAAmB,KAAK,CAAC,oBAAoB,CAAC,cAAc,EAAE,uJAAuJ,CACtN,CAAC;IACJ,CAAC;IACD,IAAI,KAAK,CAAC,mBAAmB,IAAI,MAAM,CAAC,qBAAqB,EAAE,CAAC;QAC9D,OAAO,CAAC,IAAI,CACV,2BAA2B,KAAK,CAAC,mBAAmB,yGAAyG,CAC9J,CAAC;IACJ,CAAC;IACD,OAAO,OAAO,CAAC;AACjB,CAAC;AAED,2DAA2D;AAC3D,MAAM,UAAU,qBAAqB,CAAC,OAAe;IACnD,OAAO,UAAU,CAAC,QAAQ,CAAC,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,KAAK,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC;AACzE,CAAC;AAED;;;;GAIG;AACH,MAAM,UAAU,uBAAuB,CAAC,KAMvC;IACC,MAAM,MAAM,GAAG,KAAK,CAAC,MAAM,IAAI,oBAAoB,EAAE,CAAC;IACtD,MAAM,IAAI,GAAG,qBAAqB,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC;IAClD,IAAI,CAAC,MAAM,CAAC,eAAe,EAAE,CAAC;QAC5B,OAAO,EAAE,OAAO,EAAE,KAAK,CAAC,OAAO,EAAE,OAAO,EAAE,KAAK,EAAE,IAAI,EAAE,CAAC;IAC1D,CAAC;IACD,kDAAkD;IAClD,IAAI,KAAK,CAAC,OAAO,CAAC,MAAM,GAAG,GAAG,EAAE,CAAC;QAC/B,KAAK,CAAC,UAAU,CAAC,GAAG,CAAC,IAAI,EAAE;YACzB,QAAQ,EAAE,KAAK,CAAC,QAAQ;YACxB,KAAK,EAAE,CAAC,KAAK,CAAC,UAAU,CAAC,GAAG,CAAC,IAAI,CAAC,EAAE,KAAK,IAAI,CAAC,CAAC,GAAG,CAAC;SACpD,CAAC,CAAC;QACH,OAAO,EAAE,OAAO,EAAE,KAAK,CAAC,OAAO,EAAE,OAAO,EAAE,KAAK,EAAE,IAAI,EAAE,CAAC;IAC1D,CAAC;IACD,MAAM,KAAK,GAAG,KAAK,CAAC,UAAU,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC;IACzC,IAAI,CAAC,KAAK,EAAE,CAAC;QACX,KAAK,CAAC,UAAU,CAAC,GAAG,CAAC,IAAI,EAAE,EAAE,QAAQ,EAAE,KAAK,CAAC,QAAQ,EAAE,KAAK,EAAE,CAAC,EAAE,CAAC,CAAC;QACnE,OAAO,EAAE,OAAO,EAAE,KAAK,CAAC,OAAO,EAAE,OAAO,EAAE,KAAK,EAAE,IAAI,EAAE,CAAC;IAC1D,CAAC;IACD,KAAK,CAAC,KAAK,IAAI,CAAC,CAAC;IACjB,MAAM,QAAQ,GAAG,KAAK,CAAC,YAAY;QACjC,CAAC,CAAC,iBAAiB,KAAK,CAAC,YAAY,EAAE;QACvC,CAAC,CAAC,EAAE,CAAC;IACP,MAAM,IAAI,GAAG,KAAK,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC,EAAE,GAAG,CAAC,CAAC,OAAO,CAAC,MAAM,EAAE,GAAG,CAAC,CAAC,IAAI,EAAE,CAAC;IACrE,MAAM,OAAO,GACX,iDAAiD,KAAK,CAAC,QAAQ,oBAAoB,IAAI,IAAI,QAAQ,IAAI;QACvG,YAAY,IAAI,GAAG,KAAK,CAAC,OAAO,CAAC,MAAM,GAAG,GAAG,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,IAAI;QAC5D,gGAAgG,CAAC;IACnG,OAAO,EAAE,OAAO,EAAE,OAAO,EAAE,IAAI,EAAE,IAAI,EAAE,CAAC;AAC1C,CAAC"}