ollama-intern-mcp 1.0.0

package/dist/prewarm.js

@@ -0,0 +1,69 @@
+/**
+ * Targeted prewarm — pulls profile.prewarm tiers into VRAM at server startup
+ * so the first real Claude call doesn't eat cold-load latency.
+ *
+ * Behavior is profile policy, not magic:
+ *   - dev-rtx5080:        prewarm = ["instant"]
+ *   - dev-rtx5080-llama:  prewarm = ["instant"]
+ *   - m5-max:             prewarm = []  (cold-load on unified memory is ~free)
+ *
+ * Each prewarm attempt logs a {kind: "prewarm", ...} NDJSON event with
+ * model, success/failure, elapsed_ms, and residency. That keeps benchmarks
+ * able to distinguish cold from warm and keeps adoption data clean.
+ *
+ * Failures are logged but never thrown — server startup must not depend on
+ * Ollama being reachable. If the model is missing or Ollama is down, the
+ * server still comes up; the user sees the prewarm failure in the log.
+ */
+import { resolveTier } from "./tiers.js";
+import { timestamp } from "./observability.js";
+/** Hard cap on per-tier prewarm wait. Long enough for a 14B cold load, short enough not to hang server startup. */
+const PREWARM_TIMEOUT_MS = 60_000;
+/**
+ * Run prewarm for the given tiers. Each tier issues a minimal generate
+ * (`prompt: "ok", num_predict: 1`) with `keep_alive: -1` so the model
+ * stays resident afterwards.
+ *
+ * Returns the count of successful prewarms; never throws.
+ */
+export async function runPrewarm(ctx, tiers) {
+    let successes = 0;
+    for (const tier of tiers) {
+        const model = resolveTier(tier, ctx.tiers);
+        const startedAt = Date.now();
+        const controller = new AbortController();
+        const timer = setTimeout(() => controller.abort(), PREWARM_TIMEOUT_MS);
+        let success = false;
+        let error;
+        try {
+            await ctx.client.generate({
+                model,
+                prompt: "ok",
+                options: { num_predict: 1, temperature: 0 },
+                keep_alive: -1,
+            }, controller.signal);
+            success = true;
+            successes++;
+        }
+        catch (err) {
+            error = err instanceof Error ? err.message : String(err);
+        }
+        finally {
+            clearTimeout(timer);
+        }
+        const residency = success ? await ctx.client.residency(model).catch(() => null) : null;
+        await ctx.logger.log({
+            kind: "prewarm",
+            ts: timestamp(),
+            tier,
+            model,
+            hardware_profile: ctx.hardwareProfile,
+            success,
+            elapsed_ms: Date.now() - startedAt,
+            residency,
+            ...(error ? { error } : {}),
+        });
+    }
+    return successes;
+}
+//# sourceMappingURL=prewarm.js.map

package/dist/prewarm.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"prewarm.js","sourceRoot":"","sources":["../src/prewarm.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;GAgBG;AAEH,OAAO,EAAE,WAAW,EAAa,MAAM,YAAY,CAAC;AAEpD,OAAO,EAAE,SAAS,EAAE,MAAM,oBAAoB,CAAC;AAE/C,mHAAmH;AACnH,MAAM,kBAAkB,GAAG,MAAM,CAAC;AAElC;;;;;;GAMG;AACH,MAAM,CAAC,KAAK,UAAU,UAAU,CAAC,GAAe,EAAE,KAAa;IAC7D,IAAI,SAAS,GAAG,CAAC,CAAC;IAClB,KAAK,MAAM,IAAI,IAAI,KAAK,EAAE,CAAC;QACzB,MAAM,KAAK,GAAG,WAAW,CAAC,IAAI,EAAE,GAAG,CAAC,KAAK,CAAC,CAAC;QAC3C,MAAM,SAAS,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;QAC7B,MAAM,UAAU,GAAG,IAAI,eAAe,EAAE,CAAC;QACzC,MAAM,KAAK,GAAG,UAAU,CAAC,GAAG,EAAE,CAAC,UAAU,CAAC,KAAK,EAAE,EAAE,kBAAkB,CAAC,CAAC;QACvE,IAAI,OAAO,GAAG,KAAK,CAAC;QACpB,IAAI,KAAyB,CAAC;QAC9B,IAAI,CAAC;YACH,MAAM,GAAG,CAAC,MAAM,CAAC,QAAQ,CACvB;gBACE,KAAK;gBACL,MAAM,EAAE,IAAI;gBACZ,OAAO,EAAE,EAAE,WAAW,EAAE,CAAC,EAAE,WAAW,EAAE,CAAC,EAAE;gBAC3C,UAAU,EAAE,CAAC,CAAC;aACf,EACD,UAAU,CAAC,MAAM,CAClB,CAAC;YACF,OAAO,GAAG,IAAI,CAAC;YACf,SAAS,EAAE,CAAC;QACd,CAAC;QAAC,OAAO,GAAG,EAAE,CAAC;YACb,KAAK,GAAG,GAAG,YAAY,KAAK,CAAC,CAAC,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC;QAC3D,CAAC;gBAAS,CAAC;YACT,YAAY,CAAC,KAAK,CAAC,CAAC;QACtB,CAAC;QAED,MAAM,SAAS,GAAG,OAAO,CAAC,CAAC,CAAC,MAAM,GAAG,CAAC,MAAM,CAAC,SAAS,CAAC,KAAK,CAAC,CAAC,KAAK,CAAC,GAAG,EAAE,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC;QAEvF,MAAM,GAAG,CAAC,MAAM,CAAC,GAAG,CAAC;YACnB,IAAI,EAAE,SAAS;YACf,EAAE,EAAE,SAAS,EAAE;YACf,IAAI;YACJ,KAAK;YACL,gBAAgB,EAAE,GAAG,CAAC,eAAe;YACrC,OAAO;YACP,UAAU,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,SAAS;YAClC,SAAS;YACT,GAAG,CAAC,KAAK,CAAC,CAAC,CAAC,EAAE,KAAK,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;SAC5B,CAAC,CAAC;IACL,CAAC;IACD,OAAO,SAAS,CAAC;AACnB,CAAC"}

package/dist/profiles.d.ts

@@ -0,0 +1,54 @@
+/**
+ * Hardware profiles — pick the right tier ladder for the machine running Ollama.
+ *
+ * The same code runs against very different hardware. This module encodes
+ * the "which model belongs on which tier for which box" decision explicitly
+ * so it's a product choice, not an env-var scavenger hunt.
+ *
+ * - dev-rtx5080 (default): Qwen ladder. Coherent day-to-day dogfooding.
+ *   Same family top-to-bottom means a bad output is a tool/design problem,
+ *   not a cross-family mismatch.
+ * - dev-rtx5080-llama: parity rail. Same instant/workhorse as default, but
+ *   Llama 8B on Deep. Use this to measure whether Llama-family drift buys
+ *   anything real before committing to it on the M5 Max.
+ * - m5-max: prod target. Real tier ladder once the box arrives.
+ *
+ * Per-tier env vars (INTERN_TIER_INSTANT, etc.) still override a profile's
+ * model picks, so one-off experiments don't require a new profile.
+ */
+import type { Tier, TierConfig } from "./tiers.js";
+export type ProfileName = "dev-rtx5080" | "dev-rtx5080-llama" | "m5-max";
+export interface Profile {
+    name: ProfileName;
+    description: string;
+    tiers: TierConfig;
+    /**
+     * Per-tier timeouts in ms. Lives on the profile (not as a global constant)
+     * because cold-load behavior is hardware-bound — Instant needs 15s of
+     * margin on a 16GB-VRAM dev box but only 5s on M5 Max unified memory.
+     * Found via the first live dogfood pass: all Instant calls on RTX 5080
+     * timed out at 5s before first token.
+     */
+    timeouts: Record<Tier, number>;
+    /**
+     * Tiers to prewarm on server startup. Targeted adoption aid for tiers
+     * most likely to become habit (Instant), where cold-load drag would
+     * poison early product feel. NOT a blanket "make everything hot" knob —
+     * Workhorse and Deep are deliberately excluded so VRAM pressure and
+     * unintended residency churn don't start mattering.
+     */
+    prewarm: Tier[];
+}
+export declare const PROFILES: Record<ProfileName, Profile>;
+export declare const DEFAULT_PROFILE: ProfileName;
+/**
+ * Resolve the active profile from env. Selection order:
+ *   1. INTERN_PROFILE env var, if a known name
+ *   2. DEFAULT_PROFILE (dev-rtx5080)
+ *
+ * Per-tier env vars (INTERN_TIER_INSTANT, etc.) override the profile's picks.
+ * Profile.timeouts are not env-overridable — they are a hardware property,
+ * not a one-off tuning knob.
+ */
+export declare function loadProfile(env?: NodeJS.ProcessEnv): Profile;
+//# sourceMappingURL=profiles.d.ts.map

package/dist/profiles.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"profiles.d.ts","sourceRoot":"","sources":["../src/profiles.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;GAiBG;AAEH,OAAO,KAAK,EAAE,IAAI,EAAE,UAAU,EAAE,MAAM,YAAY,CAAC;AAEnD,MAAM,MAAM,WAAW,GAAG,aAAa,GAAG,mBAAmB,GAAG,QAAQ,CAAC;AAEzE,MAAM,WAAW,OAAO;IACtB,IAAI,EAAE,WAAW,CAAC;IAClB,WAAW,EAAE,MAAM,CAAC;IACpB,KAAK,EAAE,UAAU,CAAC;IAClB;;;;;;OAMG;IACH,QAAQ,EAAE,MAAM,CAAC,IAAI,EAAE,MAAM,CAAC,CAAC;IAC/B;;;;;;OAMG;IACH,OAAO,EAAE,IAAI,EAAE,CAAC;CACjB;AAkBD,eAAO,MAAM,QAAQ,EAAE,MAAM,CAAC,WAAW,EAAE,OAAO,CAuCjD,CAAC;AAEF,eAAO,MAAM,eAAe,EAAE,WAA2B,CAAC;AAM1D;;;;;;;;GAQG;AACH,wBAAgB,WAAW,CAAC,GAAG,GAAE,MAAM,CAAC,UAAwB,GAAG,OAAO,CAUzE"}

package/dist/profiles.js

@@ -0,0 +1,95 @@
+/**
+ * Hardware profiles — pick the right tier ladder for the machine running Ollama.
+ *
+ * The same code runs against very different hardware. This module encodes
+ * the "which model belongs on which tier for which box" decision explicitly
+ * so it's a product choice, not an env-var scavenger hunt.
+ *
+ * - dev-rtx5080 (default): Qwen ladder. Coherent day-to-day dogfooding.
+ *   Same family top-to-bottom means a bad output is a tool/design problem,
+ *   not a cross-family mismatch.
+ * - dev-rtx5080-llama: parity rail. Same instant/workhorse as default, but
+ *   Llama 8B on Deep. Use this to measure whether Llama-family drift buys
+ *   anything real before committing to it on the M5 Max.
+ * - m5-max: prod target. Real tier ladder once the box arrives.
+ *
+ * Per-tier env vars (INTERN_TIER_INSTANT, etc.) still override a profile's
+ * model picks, so one-off experiments don't require a new profile.
+ */
+/** Timeouts sized for a 16GB-VRAM discrete-GPU box where cold load costs 3-5s. */
+const DEV_RTX5080_TIMEOUTS = {
+    instant: 15_000,
+    workhorse: 20_000,
+    deep: 90_000,
+    embed: 10_000,
+};
+/** Timeouts sized for M5 Max 128GB unified memory where cold load is ~instant. */
+const M5_MAX_TIMEOUTS = {
+    instant: 5_000,
+    workhorse: 20_000,
+    deep: 90_000,
+    embed: 10_000,
+};
+export const PROFILES = {
+    "dev-rtx5080": {
+        name: "dev-rtx5080",
+        description: "RTX 5080 16GB VRAM — Qwen ladder. Default dev profile for dogfooding the delegation spine.",
+        tiers: {
+            instant: "qwen2.5:7b-instruct-q4_K_M",
+            workhorse: "qwen2.5-coder:7b-instruct-q4_K_M",
+            deep: "qwen2.5:14b-instruct-q4_K_M",
+            embed: "nomic-embed-text",
+        },
+        timeouts: DEV_RTX5080_TIMEOUTS,
+        prewarm: ["instant"],
+    },
+    "dev-rtx5080-llama": {
+        name: "dev-rtx5080-llama",
+        description: "RTX 5080 with Llama 8B on Deep — parity comparison lane for future Llama-family Deep migration.",
+        tiers: {
+            instant: "qwen2.5:7b-instruct-q4_K_M",
+            workhorse: "qwen2.5-coder:7b-instruct-q4_K_M",
+            deep: "llama3.1:8b-instruct-q4_K_M",
+            embed: "nomic-embed-text",
+        },
+        timeouts: DEV_RTX5080_TIMEOUTS,
+        prewarm: ["instant"],
+    },
+    "m5-max": {
+        name: "m5-max",
+        description: "M5 Max 128GB unified — full tier ladder. Prod target.",
+        tiers: {
+            instant: "qwen2.5:14b-instruct-q4_K_M",
+            workhorse: "qwen2.5-coder:32b-instruct-q4_K_M",
+            deep: "llama3.3:70b-instruct-q4_K_M",
+            embed: "nomic-embed-text",
+        },
+        timeouts: M5_MAX_TIMEOUTS,
+        prewarm: [],
+    },
+};
+export const DEFAULT_PROFILE = "dev-rtx5080";
+function isProfileName(x) {
+    return x === "dev-rtx5080" || x === "dev-rtx5080-llama" || x === "m5-max";
+}
+/**
+ * Resolve the active profile from env. Selection order:
+ *   1. INTERN_PROFILE env var, if a known name
+ *   2. DEFAULT_PROFILE (dev-rtx5080)
+ *
+ * Per-tier env vars (INTERN_TIER_INSTANT, etc.) override the profile's picks.
+ * Profile.timeouts are not env-overridable — they are a hardware property,
+ * not a one-off tuning knob.
+ */
+export function loadProfile(env = process.env) {
+    const name = isProfileName(env.INTERN_PROFILE) ? env.INTERN_PROFILE : DEFAULT_PROFILE;
+    const base = PROFILES[name];
+    const tiers = {
+        instant: env.INTERN_TIER_INSTANT || base.tiers.instant,
+        workhorse: env.INTERN_TIER_WORKHORSE || base.tiers.workhorse,
+        deep: env.INTERN_TIER_DEEP || base.tiers.deep,
+        embed: env.INTERN_EMBED_MODEL || base.tiers.embed,
+    };
+    return { name, description: base.description, tiers, timeouts: base.timeouts, prewarm: base.prewarm };
+}
+//# sourceMappingURL=profiles.js.map

package/dist/profiles.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"profiles.js","sourceRoot":"","sources":["../src/profiles.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;GAiBG;AA4BH,kFAAkF;AAClF,MAAM,oBAAoB,GAAyB;IACjD,OAAO,EAAE,MAAM;IACf,SAAS,EAAE,MAAM;IACjB,IAAI,EAAE,MAAM;IACZ,KAAK,EAAE,MAAM;CACd,CAAC;AAEF,kFAAkF;AAClF,MAAM,eAAe,GAAyB;IAC5C,OAAO,EAAE,KAAK;IACd,SAAS,EAAE,MAAM;IACjB,IAAI,EAAE,MAAM;IACZ,KAAK,EAAE,MAAM;CACd,CAAC;AAEF,MAAM,CAAC,MAAM,QAAQ,GAAiC;IACpD,aAAa,EAAE;QACb,IAAI,EAAE,aAAa;QACnB,WAAW,EACT,4FAA4F;QAC9F,KAAK,EAAE;YACL,OAAO,EAAE,4BAA4B;YACrC,SAAS,EAAE,kCAAkC;YAC7C,IAAI,EAAE,6BAA6B;YACnC,KAAK,EAAE,kBAAkB;SAC1B;QACD,QAAQ,EAAE,oBAAoB;QAC9B,OAAO,EAAE,CAAC,SAAS,CAAC;KACrB;IACD,mBAAmB,EAAE;QACnB,IAAI,EAAE,mBAAmB;QACzB,WAAW,EACT,iGAAiG;QACnG,KAAK,EAAE;YACL,OAAO,EAAE,4BAA4B;YACrC,SAAS,EAAE,kCAAkC;YAC7C,IAAI,EAAE,6BAA6B;YACnC,KAAK,EAAE,kBAAkB;SAC1B;QACD,QAAQ,EAAE,oBAAoB;QAC9B,OAAO,EAAE,CAAC,SAAS,CAAC;KACrB;IACD,QAAQ,EAAE;QACR,IAAI,EAAE,QAAQ;QACd,WAAW,EAAE,uDAAuD;QACpE,KAAK,EAAE;YACL,OAAO,EAAE,6BAA6B;YACtC,SAAS,EAAE,mCAAmC;YAC9C,IAAI,EAAE,8BAA8B;YACpC,KAAK,EAAE,kBAAkB;SAC1B;QACD,QAAQ,EAAE,eAAe;QACzB,OAAO,EAAE,EAAE;KACZ;CACF,CAAC;AAEF,MAAM,CAAC,MAAM,eAAe,GAAgB,aAAa,CAAC;AAE1D,SAAS,aAAa,CAAC,CAAqB;IAC1C,OAAO,CAAC,KAAK,aAAa,IAAI,CAAC,KAAK,mBAAmB,IAAI,CAAC,KAAK,QAAQ,CAAC;AAC5E,CAAC;AAED;;;;;;;;GAQG;AACH,MAAM,UAAU,WAAW,CAAC,MAAyB,OAAO,CAAC,GAAG;IAC9D,MAAM,IAAI,GAAgB,aAAa,CAAC,GAAG,CAAC,cAAc,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,cAAc,CAAC,CAAC,CAAC,eAAe,CAAC;IACnG,MAAM,IAAI,GAAG,QAAQ,CAAC,IAAI,CAAC,CAAC;IAC5B,MAAM,KAAK,GAAe;QACxB,OAAO,EAAE,GAAG,CAAC,mBAAmB,IAAI,IAAI,CAAC,KAAK,CAAC,OAAO;QACtD,SAAS,EAAE,GAAG,CAAC,qBAAqB,IAAI,IAAI,CAAC,KAAK,CAAC,SAAS;QAC5D,IAAI,EAAE,GAAG,CAAC,gBAAgB,IAAI,IAAI,CAAC,KAAK,CAAC,IAAI;QAC7C,KAAK,EAAE,GAAG,CAAC,kBAAkB,IAAI,IAAI,CAAC,KAAK,CAAC,KAAK;KAClD,CAAC;IACF,OAAO,EAAE,IAAI,EAAE,WAAW,EAAE,IAAI,CAAC,WAAW,EAAE,KAAK,EAAE,QAAQ,EAAE,IAAI,CAAC,QAAQ,EAAE,OAAO,EAAE,IAAI,CAAC,OAAO,EAAE,CAAC;AACxG,CAAC"}

package/dist/protectedPaths.d.ts

@@ -0,0 +1,33 @@
+/**
+ * Protected-path list — explicit, versioned, one place.
+ *
+ * Writes from `ollama_draft` that target these paths require explicit
+ * `confirm_write: true`, enforced server-side (never prompt-side).
+ *
+ * Do NOT scatter conditionals across tool handlers. Add a new protected
+ * path here, bump PROTECTED_PATHS_VERSION, and the whole system picks it up.
+ */
+export declare const PROTECTED_PATHS_VERSION = 1;
+export interface ProtectedPathRule {
+    /** Glob-ish pattern, evaluated by matchesProtectedPath(). Use POSIX separators. */
+    pattern: string;
+    /** Human-readable reason shown in the error. */
+    reason: string;
+}
+/**
+ * Rules — kept explicit so a reviewer can eyeball them.
+ *
+ * Matching is substring-against-normalized-path with "/" as separator.
+ * A pattern ending in "/" matches any descendant.
+ */
+export declare const PROTECTED_PATHS: ProtectedPathRule[];
+/**
+ * Normalize a path for comparison: forward slashes, no leading ./, lowercase on Windows.
+ */
+export declare function normalizePath(p: string): string;
+export interface ProtectedMatch {
+    protected: boolean;
+    rule?: ProtectedPathRule;
+}
+export declare function matchesProtectedPath(path: string, rules?: ProtectedPathRule[]): ProtectedMatch;
+//# sourceMappingURL=protectedPaths.d.ts.map

package/dist/protectedPaths.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"protectedPaths.d.ts","sourceRoot":"","sources":["../src/protectedPaths.ts"],"names":[],"mappings":"AAAA;;;;;;;;GAQG;AAEH,eAAO,MAAM,uBAAuB,IAAI,CAAC;AAEzC,MAAM,WAAW,iBAAiB;IAChC,mFAAmF;IACnF,OAAO,EAAE,MAAM,CAAC;IAChB,gDAAgD;IAChD,MAAM,EAAE,MAAM,CAAC;CAChB;AAED;;;;;GAKG;AACH,eAAO,MAAM,eAAe,EAAE,iBAAiB,EAW9C,CAAC;AAEF;;GAEG;AACH,wBAAgB,aAAa,CAAC,CAAC,EAAE,MAAM,GAAG,MAAM,CAI/C;AAED,MAAM,WAAW,cAAc;IAC7B,SAAS,EAAE,OAAO,CAAC;IACnB,IAAI,CAAC,EAAE,iBAAiB,CAAC;CAC1B;AAED,wBAAgB,oBAAoB,CAClC,IAAI,EAAE,MAAM,EACZ,KAAK,GAAE,iBAAiB,EAAoB,GAC3C,cAAc,CAchB"}

package/dist/protectedPaths.js

@@ -0,0 +1,54 @@
+/**
+ * Protected-path list — explicit, versioned, one place.
+ *
+ * Writes from `ollama_draft` that target these paths require explicit
+ * `confirm_write: true`, enforced server-side (never prompt-side).
+ *
+ * Do NOT scatter conditionals across tool handlers. Add a new protected
+ * path here, bump PROTECTED_PATHS_VERSION, and the whole system picks it up.
+ */
+export const PROTECTED_PATHS_VERSION = 1;
+/**
+ * Rules — kept explicit so a reviewer can eyeball them.
+ *
+ * Matching is substring-against-normalized-path with "/" as separator.
+ * A pattern ending in "/" matches any descendant.
+ */
+export const PROTECTED_PATHS = [
+    { pattern: "memory/", reason: "Canon-adjacent memory — human judgment only." },
+    { pattern: "MEMORY.md", reason: "Memory index — only the memory system updates this." },
+    { pattern: ".claude/", reason: "Claude configuration and rules — human-owned." },
+    { pattern: "docs/canon/", reason: "Game canon — authored by humans, never drafted." },
+    { pattern: "canon/", reason: "Game canon — authored by humans, never drafted." },
+    { pattern: "doctrine/", reason: "Combat/systems doctrine — human-owned." },
+    { pattern: "games/", reason: "Proprietary game data root." },
+    { pattern: ".git/", reason: "Git internals." },
+    { pattern: "SECURITY.md", reason: "Security policy — reviewed changes only." },
+    { pattern: "LICENSE", reason: "License — do not auto-modify." },
+];
+/**
+ * Normalize a path for comparison: forward slashes, no leading ./, lowercase on Windows.
+ */
+export function normalizePath(p) {
+    let n = p.replace(/\\/g, "/");
+    if (n.startsWith("./"))
+        n = n.slice(2);
+    return n;
+}
+export function matchesProtectedPath(path, rules = PROTECTED_PATHS) {
+    const n = normalizePath(path);
+    for (const rule of rules) {
+        const pat = rule.pattern;
+        if (pat.endsWith("/")) {
+            // Directory rule: any segment boundary containing the dir name matches.
+            if (n.startsWith(pat) || n.includes("/" + pat)) {
+                return { protected: true, rule };
+            }
+        }
+        else if (n === pat || n.endsWith("/" + pat)) {
+            return { protected: true, rule };
+        }
+    }
+    return { protected: false };
+}
+//# sourceMappingURL=protectedPaths.js.map

package/dist/protectedPaths.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"protectedPaths.js","sourceRoot":"","sources":["../src/protectedPaths.ts"],"names":[],"mappings":"AAAA;;;;;;;;GAQG;AAEH,MAAM,CAAC,MAAM,uBAAuB,GAAG,CAAC,CAAC;AASzC;;;;;GAKG;AACH,MAAM,CAAC,MAAM,eAAe,GAAwB;IAClD,EAAE,OAAO,EAAE,SAAS,EAAE,MAAM,EAAE,8CAA8C,EAAE;IAC9E,EAAE,OAAO,EAAE,WAAW,EAAE,MAAM,EAAE,qDAAqD,EAAE;IACvF,EAAE,OAAO,EAAE,UAAU,EAAE,MAAM,EAAE,+CAA+C,EAAE;IAChF,EAAE,OAAO,EAAE,aAAa,EAAE,MAAM,EAAE,iDAAiD,EAAE;IACrF,EAAE,OAAO,EAAE,QAAQ,EAAE,MAAM,EAAE,iDAAiD,EAAE;IAChF,EAAE,OAAO,EAAE,WAAW,EAAE,MAAM,EAAE,wCAAwC,EAAE;IAC1E,EAAE,OAAO,EAAE,QAAQ,EAAE,MAAM,EAAE,6BAA6B,EAAE;IAC5D,EAAE,OAAO,EAAE,OAAO,EAAE,MAAM,EAAE,gBAAgB,EAAE;IAC9C,EAAE,OAAO,EAAE,aAAa,EAAE,MAAM,EAAE,0CAA0C,EAAE;IAC9E,EAAE,OAAO,EAAE,SAAS,EAAE,MAAM,EAAE,+BAA+B,EAAE;CAChE,CAAC;AAEF;;GAEG;AACH,MAAM,UAAU,aAAa,CAAC,CAAS;IACrC,IAAI,CAAC,GAAG,CAAC,CAAC,OAAO,CAAC,KAAK,EAAE,GAAG,CAAC,CAAC;IAC9B,IAAI,CAAC,CAAC,UAAU,CAAC,IAAI,CAAC;QAAE,CAAC,GAAG,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC;IACvC,OAAO,CAAC,CAAC;AACX,CAAC;AAOD,MAAM,UAAU,oBAAoB,CAClC,IAAY,EACZ,QAA6B,eAAe;IAE5C,MAAM,CAAC,GAAG,aAAa,CAAC,IAAI,CAAC,CAAC;IAC9B,KAAK,MAAM,IAAI,IAAI,KAAK,EAAE,CAAC;QACzB,MAAM,GAAG,GAAG,IAAI,CAAC,OAAO,CAAC;QACzB,IAAI,GAAG,CAAC,QAAQ,CAAC,GAAG,CAAC,EAAE,CAAC;YACtB,wEAAwE;YACxE,IAAI,CAAC,CAAC,UAAU,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC,QAAQ,CAAC,GAAG,GAAG,GAAG,CAAC,EAAE,CAAC;gBAC/C,OAAO,EAAE,SAAS,EAAE,IAAI,EAAE,IAAI,EAAE,CAAC;YACnC,CAAC;QACH,CAAC;aAAM,IAAI,CAAC,KAAK,GAAG,IAAI,CAAC,CAAC,QAAQ,CAAC,GAAG,GAAG,GAAG,CAAC,EAAE,CAAC;YAC9C,OAAO,EAAE,SAAS,EAAE,IAAI,EAAE,IAAI,EAAE,CAAC;QACnC,CAAC;IACH,CAAC;IACD,OAAO,EAAE,SAAS,EAAE,KAAK,EAAE,CAAC;AAC9B,CAAC"}

package/dist/runContext.d.ts

@@ -0,0 +1,21 @@
+/**
+ * Shared dependencies every tool handler needs.
+ *
+ * Passing one RunContext is cleaner than threading 3+ positional args,
+ * and gives us a stable shape to extend later (metrics, tracing, etc.)
+ * without touching 8 call sites.
+ */
+import type { OllamaClient } from "./ollama.js";
+import type { Tier, TierConfig } from "./tiers.js";
+import type { Logger } from "./observability.js";
+export interface RunContext {
+    client: OllamaClient;
+    /** Concrete tier→model picks from the active Profile. */
+    tiers: TierConfig;
+    /** Per-tier timeouts in ms, sized for the active profile's hardware. */
+    timeouts: Record<Tier, number>;
+    /** Profile name written onto every envelope + NDJSON line. */
+    hardwareProfile: string;
+    logger: Logger;
+}
+//# sourceMappingURL=runContext.d.ts.map

package/dist/runContext.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"runContext.d.ts","sourceRoot":"","sources":["../src/runContext.ts"],"names":[],"mappings":"AAAA;;;;;;GAMG;AAEH,OAAO,KAAK,EAAE,YAAY,EAAE,MAAM,aAAa,CAAC;AAChD,OAAO,KAAK,EAAE,IAAI,EAAE,UAAU,EAAE,MAAM,YAAY,CAAC;AACnD,OAAO,KAAK,EAAE,MAAM,EAAE,MAAM,oBAAoB,CAAC;AAEjD,MAAM,WAAW,UAAU;IACzB,MAAM,EAAE,YAAY,CAAC;IACrB,yDAAyD;IACzD,KAAK,EAAE,UAAU,CAAC;IAClB,wEAAwE;IACxE,QAAQ,EAAE,MAAM,CAAC,IAAI,EAAE,MAAM,CAAC,CAAC;IAC/B,8DAA8D;IAC9D,eAAe,EAAE,MAAM,CAAC;IACxB,MAAM,EAAE,MAAM,CAAC;CAChB"}

package/dist/runContext.js

@@ -0,0 +1,9 @@
+/**
+ * Shared dependencies every tool handler needs.
+ *
+ * Passing one RunContext is cleaner than threading 3+ positional args,
+ * and gives us a stable shape to extend later (metrics, tracing, etc.)
+ * without touching 8 call sites.
+ */
+export {};
+//# sourceMappingURL=runContext.js.map

package/dist/runContext.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"runContext.js","sourceRoot":"","sources":["../src/runContext.ts"],"names":[],"mappings":"AAAA;;;;;;GAMG"}

package/dist/semaphore.d.ts

@@ -0,0 +1,16 @@
+/**
+ * Global concurrency gate around the local Ollama instance.
+ * Local inference is bottlenecked on VRAM and compute; letting 20 calls
+ * race into llama.cpp just thrashes. Default to 2 concurrent calls.
+ */
+export declare class Semaphore {
+    private permits;
+    private queue;
+    constructor(permits: number);
+    acquire(): Promise<() => void>;
+    private release;
+    /** For tests. */
+    get pending(): number;
+}
+export declare const ollamaSemaphore: Semaphore;
+//# sourceMappingURL=semaphore.d.ts.map

package/dist/semaphore.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"semaphore.d.ts","sourceRoot":"","sources":["../src/semaphore.ts"],"names":[],"mappings":"AAAA;;;;GAIG;AAEH,qBAAa,SAAS;IACpB,OAAO,CAAC,OAAO,CAAS;IACxB,OAAO,CAAC,KAAK,CAAyB;gBAE1B,OAAO,EAAE,MAAM;IAIrB,OAAO,IAAI,OAAO,CAAC,MAAM,IAAI,CAAC;IAapC,OAAO,CAAC,OAAO;IAMf,iBAAiB;IACjB,IAAI,OAAO,IAAI,MAAM,CAEpB;CACF;AAID,eAAO,MAAM,eAAe,WAAqC,CAAC"}

package/dist/semaphore.js

@@ -0,0 +1,37 @@
+/**
+ * Global concurrency gate around the local Ollama instance.
+ * Local inference is bottlenecked on VRAM and compute; letting 20 calls
+ * race into llama.cpp just thrashes. Default to 2 concurrent calls.
+ */
+export class Semaphore {
+    permits;
+    queue = [];
+    constructor(permits) {
+        this.permits = permits;
+    }
+    async acquire() {
+        if (this.permits > 0) {
+            this.permits--;
+            return () => this.release();
+        }
+        return new Promise((resolve) => {
+            this.queue.push(() => {
+                this.permits--;
+                resolve(() => this.release());
+            });
+        });
+    }
+    release() {
+        this.permits++;
+        const next = this.queue.shift();
+        if (next)
+            next();
+    }
+    /** For tests. */
+    get pending() {
+        return this.queue.length;
+    }
+}
+const DEFAULT_CONCURRENCY = Number(process.env.INTERN_MAX_CONCURRENT ?? 2);
+export const ollamaSemaphore = new Semaphore(DEFAULT_CONCURRENCY);
+//# sourceMappingURL=semaphore.js.map

package/dist/semaphore.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"semaphore.js","sourceRoot":"","sources":["../src/semaphore.ts"],"names":[],"mappings":"AAAA;;;;GAIG;AAEH,MAAM,OAAO,SAAS;IACZ,OAAO,CAAS;IAChB,KAAK,GAAsB,EAAE,CAAC;IAEtC,YAAY,OAAe;QACzB,IAAI,CAAC,OAAO,GAAG,OAAO,CAAC;IACzB,CAAC;IAED,KAAK,CAAC,OAAO;QACX,IAAI,IAAI,CAAC,OAAO,GAAG,CAAC,EAAE,CAAC;YACrB,IAAI,CAAC,OAAO,EAAE,CAAC;YACf,OAAO,GAAG,EAAE,CAAC,IAAI,CAAC,OAAO,EAAE,CAAC;QAC9B,CAAC;QACD,OAAO,IAAI,OAAO,CAAa,CAAC,OAAO,EAAE,EAAE;YACzC,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,GAAG,EAAE;gBACnB,IAAI,CAAC,OAAO,EAAE,CAAC;gBACf,OAAO,CAAC,GAAG,EAAE,CAAC,IAAI,CAAC,OAAO,EAAE,CAAC,CAAC;YAChC,CAAC,CAAC,CAAC;QACL,CAAC,CAAC,CAAC;IACL,CAAC;IAEO,OAAO;QACb,IAAI,CAAC,OAAO,EAAE,CAAC;QACf,MAAM,IAAI,GAAG,IAAI,CAAC,KAAK,CAAC,KAAK,EAAE,CAAC;QAChC,IAAI,IAAI;YAAE,IAAI,EAAE,CAAC;IACnB,CAAC;IAED,iBAAiB;IACjB,IAAI,OAAO;QACT,OAAO,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC;IAC3B,CAAC;CACF;AAED,MAAM,mBAAmB,GAAG,MAAM,CAAC,OAAO,CAAC,GAAG,CAAC,qBAAqB,IAAI,CAAC,CAAC,CAAC;AAE3E,MAAM,CAAC,MAAM,eAAe,GAAG,IAAI,SAAS,CAAC,mBAAmB,CAAC,CAAC"}

package/dist/sources.d.ts

@@ -0,0 +1,19 @@
+/**
+ * Shared source-file loader used by tools that accept `source_paths[]`.
+ *
+ * The point of a path-based tool input is context preservation — Claude does
+ * not pre-read the file, the server does. Keep this module fast and boring.
+ */
+export interface LoadedSource {
+    path: string;
+    body: string;
+}
+/**
+ * Read each path, slice to `perFileMax` chars per file, return in input order.
+ * Throws SOURCE_PATH_NOT_FOUND on the first missing/unreadable path so the
+ * caller fails loud instead of getting a partial answer.
+ */
+export declare function loadSources(paths: string[], perFileMax: number): Promise<LoadedSource[]>;
+/** Format loaded sources as a single prompt block with begin/end markers per file. */
+export declare function formatSourcesBlock(sources: LoadedSource[]): string;
+//# sourceMappingURL=sources.d.ts.map

package/dist/sources.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"sources.d.ts","sourceRoot":"","sources":["../src/sources.ts"],"names":[],"mappings":"AAAA;;;;;GAKG;AAMH,MAAM,WAAW,YAAY;IAC3B,IAAI,EAAE,MAAM,CAAC;IACb,IAAI,EAAE,MAAM,CAAC;CACd;AAED;;;;GAIG;AACH,wBAAsB,WAAW,CAC/B,KAAK,EAAE,MAAM,EAAE,EACf,UAAU,EAAE,MAAM,GACjB,OAAO,CAAC,YAAY,EAAE,CAAC,CA2BzB;AAED,sFAAsF;AACtF,wBAAgB,kBAAkB,CAAC,OAAO,EAAE,YAAY,EAAE,GAAG,MAAM,CAIlE"}

package/dist/sources.js

@@ -0,0 +1,41 @@
+/**
+ * Shared source-file loader used by tools that accept `source_paths[]`.
+ *
+ * The point of a path-based tool input is context preservation — Claude does
+ * not pre-read the file, the server does. Keep this module fast and boring.
+ */
+import { readFile, stat } from "node:fs/promises";
+import { resolve } from "node:path";
+import { InternError } from "./errors.js";
+/**
+ * Read each path, slice to `perFileMax` chars per file, return in input order.
+ * Throws SOURCE_PATH_NOT_FOUND on the first missing/unreadable path so the
+ * caller fails loud instead of getting a partial answer.
+ */
+export async function loadSources(paths, perFileMax) {
+    const loaded = [];
+    for (const p of paths) {
+        const abs = resolve(p);
+        try {
+            const st = await stat(abs);
+            if (!st.isFile()) {
+                throw new InternError("SOURCE_PATH_NOT_FOUND", `Not a file: ${p}`, "Pass file paths only, not directories.", false);
+            }
+            const raw = await readFile(abs, "utf8");
+            loaded.push({ path: p, body: raw.slice(0, perFileMax) });
+        }
+        catch (err) {
+            if (err instanceof InternError)
+                throw err;
+            throw new InternError("SOURCE_PATH_NOT_FOUND", `Cannot read source path: ${p} — ${err.message}`, "Check the path exists and is readable.", false);
+        }
+    }
+    return loaded;
+}
+/** Format loaded sources as a single prompt block with begin/end markers per file. */
+export function formatSourcesBlock(sources) {
+    return sources
+        .map((s) => `=== BEGIN ${s.path} ===\n${s.body}\n=== END ${s.path} ===`)
+        .join("\n\n");
+}
+//# sourceMappingURL=sources.js.map

package/dist/sources.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"sources.js","sourceRoot":"","sources":["../src/sources.ts"],"names":[],"mappings":"AAAA;;;;;GAKG;AAEH,OAAO,EAAE,QAAQ,EAAE,IAAI,EAAE,MAAM,kBAAkB,CAAC;AAClD,OAAO,EAAE,OAAO,EAAE,MAAM,WAAW,CAAC;AACpC,OAAO,EAAE,WAAW,EAAE,MAAM,aAAa,CAAC;AAO1C;;;;GAIG;AACH,MAAM,CAAC,KAAK,UAAU,WAAW,CAC/B,KAAe,EACf,UAAkB;IAElB,MAAM,MAAM,GAAmB,EAAE,CAAC;IAClC,KAAK,MAAM,CAAC,IAAI,KAAK,EAAE,CAAC;QACtB,MAAM,GAAG,GAAG,OAAO,CAAC,CAAC,CAAC,CAAC;QACvB,IAAI,CAAC;YACH,MAAM,EAAE,GAAG,MAAM,IAAI,CAAC,GAAG,CAAC,CAAC;YAC3B,IAAI,CAAC,EAAE,CAAC,MAAM,EAAE,EAAE,CAAC;gBACjB,MAAM,IAAI,WAAW,CACnB,uBAAuB,EACvB,eAAe,CAAC,EAAE,EAClB,wCAAwC,EACxC,KAAK,CACN,CAAC;YACJ,CAAC;YACD,MAAM,GAAG,GAAG,MAAM,QAAQ,CAAC,GAAG,EAAE,MAAM,CAAC,CAAC;YACxC,MAAM,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,CAAC,EAAE,IAAI,EAAE,GAAG,CAAC,KAAK,CAAC,CAAC,EAAE,UAAU,CAAC,EAAE,CAAC,CAAC;QAC3D,CAAC;QAAC,OAAO,GAAG,EAAE,CAAC;YACb,IAAI,GAAG,YAAY,WAAW;gBAAE,MAAM,GAAG,CAAC;YAC1C,MAAM,IAAI,WAAW,CACnB,uBAAuB,EACvB,4BAA4B,CAAC,MAAO,GAAa,CAAC,OAAO,EAAE,EAC3D,wCAAwC,EACxC,KAAK,CACN,CAAC;QACJ,CAAC;IACH,CAAC;IACD,OAAO,MAAM,CAAC;AAChB,CAAC;AAED,sFAAsF;AACtF,MAAM,UAAU,kBAAkB,CAAC,OAAuB;IACxD,OAAO,OAAO;SACX,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,aAAa,CAAC,CAAC,IAAI,SAAS,CAAC,CAAC,IAAI,aAAa,CAAC,CAAC,IAAI,MAAM,CAAC;SACvE,IAAI,CAAC,MAAM,CAAC,CAAC;AAClB,CAAC"}

package/dist/tiers.d.ts

@@ -0,0 +1,39 @@
+/**
+ * Tier shape and constants — tier→model selection lives in ./profiles.ts.
+ *
+ * Claude picks the tier by picking the tool. Tools declare their tier;
+ * resolveTier() turns a Tier into the concrete model name from a TierConfig,
+ * which comes from the active Profile.
+ */
+export type Tier = "instant" | "workhorse" | "deep" | "embed";
+export interface TierConfig {
+    instant: string;
+    workhorse: string;
+    deep: string;
+    embed: string;
+}
+export declare function resolveTier(tier: Tier, config: TierConfig): string;
+/**
+ * Per-tier timeouts in ms. Used by guardrails/timeouts.ts to enforce
+ * degradation rules and decide fallback behavior.
+ */
+export declare const TIER_TIMEOUT_MS: Record<Tier, number>;
+/**
+ * Fallback tier — what to degrade to when a tier's timeout fires.
+ * Deep → workhorse → instant. Embed has no fallback (no cheaper embed tier).
+ */
+export declare const TIER_FALLBACK: Record<Tier, Tier | null>;
+/**
+ * Default temperatures by work shape. Small models reward tight scaffolding;
+ * these map to the patterns in the handoff prompt-shape notes.
+ */
+export declare const TEMPERATURE_BY_SHAPE: {
+    readonly classify: 0.1;
+    readonly extract: 0.1;
+    readonly triage: 0.1;
+    readonly summarize: 0.3;
+    readonly research: 0.3;
+    readonly draft: 0.4;
+    readonly chat: 0.7;
+};
+//# sourceMappingURL=tiers.d.ts.map

package/dist/tiers.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"tiers.d.ts","sourceRoot":"","sources":["../src/tiers.ts"],"names":[],"mappings":"AAAA;;;;;;GAMG;AAEH,MAAM,MAAM,IAAI,GAAG,SAAS,GAAG,WAAW,GAAG,MAAM,GAAG,OAAO,CAAC;AAE9D,MAAM,WAAW,UAAU;IACzB,OAAO,EAAE,MAAM,CAAC;IAChB,SAAS,EAAE,MAAM,CAAC;IAClB,IAAI,EAAE,MAAM,CAAC;IACb,KAAK,EAAE,MAAM,CAAC;CACf;AAED,wBAAgB,WAAW,CAAC,IAAI,EAAE,IAAI,EAAE,MAAM,EAAE,UAAU,GAAG,MAAM,CAElE;AAED;;;GAGG;AACH,eAAO,MAAM,eAAe,EAAE,MAAM,CAAC,IAAI,EAAE,MAAM,CAKhD,CAAC;AAEF;;;GAGG;AACH,eAAO,MAAM,aAAa,EAAE,MAAM,CAAC,IAAI,EAAE,IAAI,GAAG,IAAI,CAKnD,CAAC;AAEF;;;GAGG;AACH,eAAO,MAAM,oBAAoB;;;;;;;;CAQvB,CAAC"}

package/dist/tiers.js

@@ -0,0 +1,44 @@
+/**
+ * Tier shape and constants — tier→model selection lives in ./profiles.ts.
+ *
+ * Claude picks the tier by picking the tool. Tools declare their tier;
+ * resolveTier() turns a Tier into the concrete model name from a TierConfig,
+ * which comes from the active Profile.
+ */
+export function resolveTier(tier, config) {
+    return config[tier];
+}
+/**
+ * Per-tier timeouts in ms. Used by guardrails/timeouts.ts to enforce
+ * degradation rules and decide fallback behavior.
+ */
+export const TIER_TIMEOUT_MS = {
+    instant: 5_000,
+    workhorse: 20_000,
+    deep: 90_000,
+    embed: 10_000,
+};
+/**
+ * Fallback tier — what to degrade to when a tier's timeout fires.
+ * Deep → workhorse → instant. Embed has no fallback (no cheaper embed tier).
+ */
+export const TIER_FALLBACK = {
+    deep: "workhorse",
+    workhorse: "instant",
+    instant: null,
+    embed: null,
+};
+/**
+ * Default temperatures by work shape. Small models reward tight scaffolding;
+ * these map to the patterns in the handoff prompt-shape notes.
+ */
+export const TEMPERATURE_BY_SHAPE = {
+    classify: 0.1,
+    extract: 0.1,
+    triage: 0.1,
+    summarize: 0.3,
+    research: 0.3,
+    draft: 0.4,
+    chat: 0.7,
+};
+//# sourceMappingURL=tiers.js.map

package/dist/tiers.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"tiers.js","sourceRoot":"","sources":["../src/tiers.ts"],"names":[],"mappings":"AAAA;;;;;;GAMG;AAWH,MAAM,UAAU,WAAW,CAAC,IAAU,EAAE,MAAkB;IACxD,OAAO,MAAM,CAAC,IAAI,CAAC,CAAC;AACtB,CAAC;AAED;;;GAGG;AACH,MAAM,CAAC,MAAM,eAAe,GAAyB;IACnD,OAAO,EAAE,KAAK;IACd,SAAS,EAAE,MAAM;IACjB,IAAI,EAAE,MAAM;IACZ,KAAK,EAAE,MAAM;CACd,CAAC;AAEF;;;GAGG;AACH,MAAM,CAAC,MAAM,aAAa,GAA8B;IACtD,IAAI,EAAE,WAAW;IACjB,SAAS,EAAE,SAAS;IACpB,OAAO,EAAE,IAAI;IACb,KAAK,EAAE,IAAI;CACZ,CAAC;AAEF;;;GAGG;AACH,MAAM,CAAC,MAAM,oBAAoB,GAAG;IAClC,QAAQ,EAAE,GAAG;IACb,OAAO,EAAE,GAAG;IACZ,MAAM,EAAE,GAAG;IACX,SAAS,EAAE,GAAG;IACd,QAAQ,EAAE,GAAG;IACb,KAAK,EAAE,GAAG;IACV,IAAI,EAAE,GAAG;CACD,CAAC"}