akm-cli 0.9.0-rc.0 → 0.9.0-rc.13

package/dist/core/config/config-schema.js

@@ -4,6 +4,15 @@
 /**
  * Zod schema for AkmConfig — the single source of truth for the on-disk shape.
  *
+ * ASSEMBLY BARREL: the individual section schemas live under `./schema/*` (one
+ * cohesive module per config area — engines, embedding, improve, sources/
+ * bundles, index, search, output, feedback, workflow, setup, plus shared
+ * `primitives`). This file assembles them into the top-level {@link AkmConfigShape}
+ * / {@link AkmConfigSchema}, owns the cross-field {@link AkmConfigSchema.superRefine},
+ * and RE-EXPORTS every section schema so all existing `./config-schema` imports
+ * keep working unchanged. The decomposition is a pure structural refactor — the
+ * generated `schemas/akm-config.json` is byte-identical.
+ *
  * Two responsibilities:
  * 1. **Validate + transform** the raw JSON-parsed config object into the runtime
  *    `AkmConfig` shape consumed by the rest of the codebase. Replaces the
@@ -34,891 +43,281 @@
  *   enforced at save time via `superRefine` on the top-level schema.
  */
 import { z } from "zod";
-import { VALID_HARNESS_IDS } from "./config-types.js";
-// ── Reusable atomic schemas ─────────────────────────────────────────────────
-/** Positive integer (used for tokens, timeouts, batch sizes). */
-const positiveInt = z.number().int().positive();
-/** Non-negative finite number (used for scores, weights, days). */
-const nonNegativeNumber = z.number().finite().min(0);
-/** Non-empty string (rejects "" and whitespace-only). */
-const nonEmptyString = z
-    .string()
-    .min(1)
-    .refine((v) => v.trim().length > 0, { message: "expected a non-empty string" });
-/** HTTP(S) URL string. */
-const httpUrl = z.string().refine((v) => v.startsWith("http://") || v.startsWith("https://"), {
-    message: "endpoint must start with http:// or https://",
-});
-// ── Feedback failure modes (F-3 / #384) ─────────────────────────────────────
-/**
- * Curated taxonomy of failure modes for negative feedback.
- *
- * Structured failure modes enable aggregation across feedback events so the
- * distill pipeline can detect that "5 assets failed for the same reason" and
- * act on it — free-text strings about the same issue are not aggregatable.
- */
-export const FEEDBACK_FAILURE_MODES = [
-    "incorrect", // Factually wrong or logically flawed content
-    "outdated", // Correct at some point but now stale
-    "dangerous", // Could cause harm if followed (security, safety)
-    "incomplete", // Missing key steps, context, or caveats
-    "redundant", // Duplicates another asset without adding value
-];
-// ── Connection configs (LLM / embedding) ────────────────────────────────────
-const LlmCapabilitiesSchema = z
+import { BUILTIN_IMPROVE_STRATEGY_NAMES, IMPROVE_PROCESS_ENGINE_CAPABILITIES } from "./engine-semantics.js";
+import { EmbeddingConnectionConfigSchema } from "./schema/embedding.js";
+import { EnginesSchema } from "./schema/engines.js";
+import { ExperimentalConfigSchema } from "./schema/experimental.js";
+import { FeedbackConfigSchema } from "./schema/feedback.js";
+import { ImproveConfigSchema } from "./schema/improve.js";
+import { IndexConfigSchema } from "./schema/index-config.js";
+import { OutputConfigSchema } from "./schema/output.js";
+import { CURRENT_CONFIG_VERSION, engineName, GlobalModelAliasesSchema, nonEmptyString, nonNegativeNumber, } from "./schema/primitives.js";
+import { SearchConfigSchema } from "./schema/search.js";
+import { SetupConfigSchema } from "./schema/setup.js";
+import { BundlesConfigSchema, RegistryConfigEntrySchema } from "./schema/sources-bundles.js";
+import { WorkflowConfigSchema } from "./schema/workflow.js";
+// ── Section re-exports (keep every `./config-schema` import path working) ────
+export { EmbeddingConnectionConfigSchema } from "./schema/embedding.js";
+export { EngineConfigSchema, EnginesSchema, LlmConnectionConfigSchema, LlmProfileConfigSchema } from "./schema/engines.js";
+export { ExperimentalConfigSchema } from "./schema/experimental.js";
+export { FEEDBACK_FAILURE_MODES, FeedbackConfigSchema } from "./schema/feedback.js";
+export { ImproveConfigSchema } from "./schema/improve.js";
+export { ConsolidateProcessConfigSchema, DistillProcessConfigSchema, ExtractProcessConfigSchema, GraphExtractionProcessConfigSchema, ImproveProcessConfigSchema, ImproveProfileConfigSchema, MemoryInferenceProcessConfigSchema, ProactiveMaintenanceProcessConfigSchema, ReflectProcessConfigSchema, TriageProcessConfigSchema, ValidationProcessConfigSchema, } from "./schema/improve-processes.js";
+export { IndexConfigSchema, IndexPassConfigSchema } from "./schema/index-config.js";
+export { OutputConfigSchema } from "./schema/output.js";
+export { CURRENT_CONFIG_VERSION, LlmInvocationOverridesSchema } from "./schema/primitives.js";
+export { SearchConfigSchema } from "./schema/search.js";
+export { SetupConfigSchema } from "./schema/setup.js";
+export { BundleConfigEntrySchema, BundlesConfigSchema, RegistryConfigEntrySchema, SourceConfigEntrySchema, } from "./schema/sources-bundles.js";
+export { WorkflowConfigSchema } from "./schema/workflow.js";
+// ── Defaults ───────────────────────────────────────────────────────────────
+export const DefaultsSchema = z
     .object({
-    structuredOutput: z.boolean().optional(),
+    engine: engineName.optional(),
+    llmEngine: engineName.optional(),
+    improveStrategy: engineName.optional(),
 })
     .passthrough();
+// ── Top-level AkmConfig ────────────────────────────────────────────────────
 /**
- * Connection config used for both top-level `llm` (after migration) and
- * `profiles.llm[*]`. `model` is required at schema level — partial entries
- * created by `akm config set llm.endpoint <url>` (where model is left absent)
- * are normalized to `model: ""` *before* Zod sees them by the load-time
- * pre-Zod migrator hook, so this strict shape gates CLI writes without
- * breaking legacy load-time partial configs.
+ * Base object schema used both as the top-level shape and as the source of
+ * truth for {@link listTopLevelConfigKeys}. {@link AkmConfigSchema} wraps this
+ * with cross-field refinements (`.superRefine()`).
+ *
+ * All fields validate loudly — typos and shape errors throw at load time. The
+ * legacy parser's warn-and-drop tolerance was a frequent source of silent
+ * configuration loss; the migration module ({@link migrateConfigShape}) handles
+ * one-time 0.7→0.8 input transforms before the schema sees the value.
  */
-export const LlmConnectionConfigSchema = z
-    .object({
-    provider: z.string().optional(),
-    endpoint: z.string(),
-    model: z.string(),
-    apiKey: z.string().optional(),
-    temperature: z.number().finite().optional(),
-    maxTokens: positiveInt.optional(),
-    timeoutMs: positiveInt.optional(),
-    concurrency: positiveInt.optional(),
-    capabilities: LlmCapabilitiesSchema.optional(),
-    extraParams: z.record(z.unknown()).optional(),
-    contextLength: positiveInt.optional(),
-    judgeModel: z.string().min(1).optional(),
-    enableThinking: z.boolean().optional(),
-})
-    .passthrough();
-export const LlmProfileConfigSchema = LlmConnectionConfigSchema.extend({
-    supportsJsonSchema: z.boolean().optional(),
-}).passthrough();
-const EmbeddingOllamaOptionsSchema = z
-    .object({
-    num_ctx: positiveInt.optional(),
-})
-    .passthrough();
+export const AkmConfigShape = {
+    configVersion: z.literal(CURRENT_CONFIG_VERSION),
+    engines: EnginesSchema.optional(),
+    defaults: DefaultsSchema.optional(),
+    // Global model-alias tiers: alias → platform → exact model string, with a
+    // reserved `"*"` platform key as fallback. Lets workflows/callers name a
+    // semantic tier ("fast", "deep") that resolves per-harness at dispatch
+    // time. Values are literal model strings, never other aliases (one
+    // resolution level). Platform keys match the platform string a command
+    // builder resolves against ("claude", "opencode", "opencode-sdk", or a
+    // custom profile's name for the default builder) — unknown keys are inert.
+    // Precedence: profile modelAliases > this table > built-in aliases.
+    modelAliases: GlobalModelAliasesSchema.optional(),
+    semanticSearchMode: z.enum(["off", "auto"]).default("off"),
+    embedding: EmbeddingConnectionConfigSchema.optional(),
+    index: IndexConfigSchema.optional(),
+    registries: z.array(RegistryConfigEntrySchema).optional(),
+    // 0.9.0 config-shape cutover (spec §10.1 / D-R5). `bundles` + `defaultBundle`
+    // are the ONLY source shape — the retired `stashDir`/`sources[]`/`installed[]`
+    // trio is hard-rejected at load (see the top-level superRefine). The migrator
+    // ({@link migrateConfigSourcesToBundles}) converts a pre-cutover config to this
+    // shape before validation. `defaultBundle` names the primary bundle (spec
+    // §11.1 short-ref resolution / D-R4).
+    bundles: BundlesConfigSchema.optional(),
+    defaultBundle: nonEmptyString.optional(),
+    output: OutputConfigSchema.optional(),
+    defaultWriteTarget: nonEmptyString.optional(),
+    search: SearchConfigSchema.optional(),
+    feedback: FeedbackConfigSchema.optional(),
+    archiveRetentionDays: nonNegativeNumber.optional(),
+    improve: ImproveConfigSchema.optional(),
+    workflow: WorkflowConfigSchema.optional(),
+    setup: SetupConfigSchema.optional(),
+    // D8 — explicit opt-ins for behaviour outside the stability contract. Every
+    // key defaults to OFF; see `src/core/config/experimental.ts` for the readers.
+    experimental: ExperimentalConfigSchema.optional(),
+};
+export const AkmConfigBaseSchema = z.object(AkmConfigShape).passthrough();
 /**
- * Embedding connection config. Both `endpoint` and `model` are optional:
- *   - Remote: provide `endpoint` (http/https URL) + `model`.
- *   - Local-only: omit `endpoint`/`model`; set `localModel` (or fall back to
- *     {@link DEFAULT_LOCAL_MODEL}).
- *
- * Consumers route via `hasRemoteEndpoint()` which checks for an http(s)
- * endpoint — absent fields take the local path naturally, no sentinels needed.
+ * Per-key overrides for the retired `stashDir`/`sources`/`installed` source-
+ * shape rejection below. Keys without an entry fall back to the shared
+ * "retired pre-cutover source shape" message, which already names the
+ * replacement (bundles) and the migration command.
  */
-export const EmbeddingConnectionConfigSchema = z
-    .object({
-    provider: z.string().optional(),
-    endpoint: z.string().optional(),
-    model: z.string().optional(),
-    apiKey: z.string().optional(),
-    dimension: positiveInt.optional(),
-    localModel: z.string().min(1).optional(),
-    maxTokens: positiveInt.optional(),
-    batchSize: positiveInt.optional(),
-    chunkSize: positiveInt.optional(),
-    contextLength: positiveInt.optional(),
-    ollamaOptions: EmbeddingOllamaOptionsSchema.optional(),
-})
-    .passthrough();
-// ── Agent profiles ──────────────────────────────────────────────────────────
-// Derives from the canonical VALID_HARNESS_IDS (#565) so the Zod gate cannot
-// drift from the TS union / parse check / setup detection.
-const AgentPlatformSchema = z.enum(VALID_HARNESS_IDS);
-export const AgentProfileConfigSchema = z
-    .object({
-    platform: AgentPlatformSchema,
-    bin: z.string().min(1).optional(),
-    args: z.array(z.string()).optional(),
-    workspace: z.string().min(1).optional(),
-    model: z.string().min(1).optional(),
-    // Per-call timeout (ms) used when a command does not pass an explicit
-    // timeout (e.g. `akm wiki ingest` without `--timeout-ms`). null = no
-    // timeout. Honored by resolveAgentProfile (integrations/agent/config.ts).
-    timeoutMs: z.union([positiveInt, z.null()]).optional(),
-    // Per-profile model aliases: lowercase alias → exact model string for
-    // this profile's CLI. Highest-precedence tier in resolveModel
-    // (integrations/agent/model-aliases.ts) — beats the config-root
-    // `modelAliases` table and the built-in opus/sonnet/haiku entries.
-    modelAliases: z.record(z.string().min(1), z.string().min(1)).optional(),
-})
-    .passthrough();
-// ── Improve profile / process ──────────────────────────────────────────────
-export const ImproveProcessConfigSchema = z
-    .object({
-    enabled: z.boolean().optional(),
-    mode: z.enum(["llm", "agent", "sdk"]).optional(),
-    profile: z.string().min(1).optional(),
-    timeoutMs: z.union([positiveInt, z.null()]).optional(),
-    allowedTypes: z.array(z.string().min(1)).optional(),
-    // Consolidate process: minimum eligible-memory pool size below which the
-    // consolidation pass skips entirely (emits `pool_below_min_size`). 0 disables
-    // the guard. Only meaningful on the `consolidate` process. Default 500.
-    minPoolSize: z.number().int().min(0).optional(),
-    // Consolidate process: deterministic near-duplicate dedup pre-pass (#617).
-    // A cheap, no-LLM fast path that collapses obvious duplicates (`.derived`
-    // origin pairs + content twins) before the LLM consolidation. Default OFF
-    // — when absent the consolidate pass behaves byte-identically to today.
-    // `cosineThreshold` is a strict floor in [0, 1] (default 0.97) for the
-    // optional embedding-similarity match; exact normalized content-hash
-    // equality always collapses regardless of the threshold. Only meaningful
-    // on the `consolidate` process.
-    dedup: z
-        .object({
-        enabled: z.boolean().optional(),
-        cosineThreshold: z.number().min(0).max(1).optional(),
-        // WS-3a: maximum pool size for the O(n²) cosine-similarity twin compare.
-        // Only the first `cosineCandidateLimit` memories are cosine-compared;
-        // exact-hash matches still run over the full pool. Default 500. Raise
-        // with care — cost is O(n²).
-        cosineCandidateLimit: z.number().int().positive().optional(),
-    })
-        .passthrough()
-        .optional(),
-    // Consolidate process: judged-state cache (#581). When enabled, a memory
-    // whose current content hash equals its cached judged hash is SKIPPED from
-    // the LLM pool (judged-unchanged → no re-judge), letting one run sweep the
-    // whole corpus at O(changed/new) cost instead of narrowing to a recent
-    // time-window slice. Default OFF — when absent the consolidate pass behaves
-    // byte-identically to today (the incrementalSince path is unaffected). Only
-    // meaningful on the `consolidate` process.
-    judgedCache: z.object({ enabled: z.boolean().optional() }).passthrough().optional(),
-    // Distill process: LLM-as-judge lesson quality gate. Default ON (R3);
-    // fail-open — judge failure/timeout/parse errors pass through. Set
-    // `enabled: false` on the distill process to opt out.
-    qualityGate: z.object({ enabled: z.boolean().optional() }).passthrough().optional(),
-    contradictionDetection: z.object({ enabled: z.boolean().optional() }).passthrough().optional(),
-    // Extract process config (only meaningful for extract process)
-    defaultSince: z.string().min(1).optional(),
-    maxTotalChars: positiveInt.optional(),
-    // Extract process: minimum raw session size (pre-filter inputCount) below
-    // which the extract LLM call is skipped (#595/#596). 0 disables the gate.
-    // Absent = default 10 (skip only truly empty sessions). Only meaningful
-    // on the `extract` process.
-    minContentChars: z.number().int().min(0).optional(),
-    maxChunkSize: z.number().int().min(1).max(50).optional(),
-    // Consolidate process: narrow candidate pool to memories modified within
-    // this duration window plus their graph neighbours. Only meaningful on
-    // the `consolidate` process. Absent = full-pool sweep.
-    incrementalSince: z.string().optional(),
-    // Consolidate process: hard cap on memories processed per pass.
-    // Reflect/distill: max refs processed (same as profile-level `limit`).
-    limit: positiveInt.optional(),
-    // Consolidate process: graph neighbours per changed memory during
-    // incremental consolidation. Default 5. Only meaningful with incrementalSince.
-    neighborsPerChanged: z.number().int().min(1).optional(),
-    // Distill process: skip distill entirely when reflect produced zero planned refs.
-    requirePlannedRefs: z.boolean().optional(),
-    // proactiveMaintenance process (Layer 2): staleness gate + rotation cooldown
-    // in days (default 30). Only meaningful on `proactiveMaintenance`.
-    dueDays: z.number().int().min(0).optional(),
-    // proactiveMaintenance process: top-N bound per run (default 25). Alias for
-    // `limit`; `maxPerRun` wins when both are set.
-    maxPerRun: positiveInt.optional(),
-    // graphExtraction process (#624 P2): when set, rank eligible files by
-    // utility_scores DESC and process only the top-N per run (incremental
-    // high-signal-first sweep). Unset = process all eligible (current
-    // behavior). Only meaningful on `graphExtraction`.
-    topN: positiveInt.optional(),
-    // graphExtraction process: full-corpus scan. When true, graph extraction
-    // runs on ALL stash files instead of only files touched by actionable refs
-    // in the current run. Used by the `graph-refresh` built-in profile / a
-    // scheduled weekly task. Only meaningful on `graphExtraction`.
-    fullScan: z.boolean().optional(),
-    // #626 — extract process: pre-LLM heuristic triage gate. When enabled, a
-    // deterministic scorer decides BEFORE the extraction LLM call whether a
-    // session carries enough signal to be worth extracting; low-signal sessions
-    // are skipped at zero LLM cost. Default OFF. Only meaningful on `extract`.
-    // `minScore` is the minimum total heuristic score to PASS (default 2).
-    triage: z
-        .object({
-        enabled: z.boolean().optional(),
-        minScore: z.number().min(0).optional(),
-    })
-        .passthrough()
-        .optional(),
-    // MemoryInference process: minimum pending memory count to run the pass.
-    minPendingCount: z.number().int().min(0).optional(),
-    // Extract process: minimum number of new (unseen, in-window) candidate
-    // sessions below which the extract pass skips entirely (emits an
-    // `improve_skipped` event with `reason: "below_min_new_sessions"`). 0
-    // disables the guard. Only meaningful on the `extract` process. Default 0
-    // (disabled) so existing behaviour is preserved; only opted-in profiles set it.
-    minNewSessions: z.number().int().min(0).optional(),
-    // Extract process: cap on NEW sessions processed (LLM-called) per run; the
-    // rest roll to the next run (still unseen). 0 disables. Absent = default 25.
-    maxSessionsPerRun: z.number().int().min(0).optional(),
-    // #561 — index agent sessions as a searchable `session` asset (extract
-    // process). Absent = on-when-an-LLM-is-available (fail-open when offline).
-    // COST: when on, each processed session makes a SECOND LLM call (the session
-    // summary) on top of the extraction call — i.e. ~2 LLM calls/session. Set to
-    // false to halve per-session extract cost at the price of unsearchable
-    // sessions. (Unchanged/skip sessions still cost zero — the content-hash
-    // ledger gates both calls upstream.)
-    indexSessions: z.boolean().optional(),
-    // #561 — minimum session duration in minutes for session indexing. 0
-    // disables the gate. Absent = default 5. Only meaningful on `extract`.
-    minSessionDuration: z.number().min(0).optional(),
-    // Consolidate process: fallback p90 wall-clock time per consolidation chunk
-    // in seconds, used for cold-start budget estimation when no telemetry
-    // history exists. The actual p90 is derived from observed run durations
-    // once sufficient history accumulates; this value is only used on the very
-    // first run. Default 30 s. Only meaningful on the `consolidate` process.
-    p90ChunkSecondsDefault: z.number().finite().positive().optional(),
-    // (WS-3b step 0a `homeostaticDemotion` was removed — R4. The key is
-    // tolerated via passthrough if an old config still carries it; continuous
-    // decay is now part of the always-applied salience recency term.)
-    // WS-3b: Schema-similarity gate (step 0b). At intake, if a new candidate's
-    // body embedding is within epsilon of an existing derived-layer lesson/knowledge
-    // node, mark it schema-consistent and lower its priority. Default ON for
-    // the `extract` process since R3 (fail-open; set `enabled: false` to opt out).
-    // Only meaningful on the `consolidate` and `extract` processes.
-    schemaSimilarity: z
-        .object({
-        enabled: z.boolean().optional(),
-        // Epsilon: cosine similarity threshold above which a candidate is schema-consistent
-        // (default 0.85 — looser than dedup's 0.97 since we want to catch conceptual overlap).
-        epsilon: z.number().min(0).max(1).optional(),
-        // Multiplicative factor applied to candidate confidence when schema-consistent.
-        // Default 0.5 — halves the confidence so schema-consistent candidates are less likely
-        // to pass the quality gate and create redundant stash entries.
-        confidencePenalty: z.number().min(0).max(1).optional(),
-    })
-        .passthrough()
-        .optional(),
-    // WS-3b: Hot-probation intake buffer (step 0c, #604). New system-generated
-    // extractions enter captureMode: hot-probation and spend ONE consolidation
-    // cycle in probation. Dedup + quality second-pass runs before promotion.
-    // Default OFF. Only meaningful on the `extract` process.
-    hotProbation: z
-        .object({
-        enabled: z.boolean().optional(),
-    })
-        .passthrough()
-        .optional(),
-    // WS-3b: Anti-collapse guards (step 8). Prevents the consolidation pipeline
-    // from collapsing too aggressively and losing diversity.
-    //   - maxGeneration: refuse to merge two assets both above this generation (default 2).
-    //   - lexicalDiversityCheck: low n-gram diversity ⇒ raise merge threshold.
-    //   - randomClusterFraction: occasional random (non-similar) cluster in pool (default 0.05).
-    //   - mergeInformationFloor: measure that merges keep provenance + specificity
-    //     (R5 §4.2; ADVISORY in v1 — counted, never refused).
-    //   - minSpecificityRetention: distinct-token retention floor for merges (default 0.6).
-    // Default ON since R5 (opt out via enabled: false). Only meaningful on the
-    // `consolidate` process.
-    antiCollapse: z
-        .object({
-        enabled: z.boolean().optional(),
-        maxGeneration: z.number().int().min(1).optional(),
-        lexicalDiversityCheck: z.boolean().optional(),
-        randomClusterFraction: z.number().min(0).max(1).optional(),
-        mergeInformationFloor: z.boolean().optional(),
-        minSpecificityRetention: z.number().min(0).max(1).optional(),
-    })
-        .passthrough()
-        .optional(),
-    // WS-3b: CLS (Complementary Learning System) interleaving (step 9).
-    // distill/memoryInference prompts include embedding-retrieved existing adjacent
-    // lessons/knowledge to prevent catastrophic interference with prior generalizations.
-    // Default OFF. Only meaningful on `distill` and `memoryInference` processes.
-    cls: z
-        .object({
-        enabled: z.boolean().optional(),
-        // Number of adjacent lessons/knowledge to include in prompts (default 3).
-        adjacentCount: z.number().int().min(1).optional(),
-    })
-        .passthrough()
-        .optional(),
-    // WS-3b: Distill→source fidelity check (step 10). After a distill proposal,
-    // check it against its cited source memories; a contradiction flag forces
-    // human review. Default OFF. Only meaningful on `distill` process.
-    fidelityCheck: z
-        .object({
-        enabled: z.boolean().optional(),
-    })
-        .passthrough()
-        .optional(),
-    // #609 — recombine process: minimum related-memory cluster size before an
-    // LLM generalization call. Default 3. Only meaningful on `recombine`.
-    minClusterSize: z.number().int().min(2).optional(),
-    // #609 — recombine process: hard cap on clusters processed per run (one
-    // bounded LLM call each). Default 5. Only meaningful on `recombine`.
-    maxClustersPerRun: positiveInt.optional(),
-    // #632 — recombine process: max members a cluster may contain before it is
-    // SKIPPED (drops bland over-broad buckets). When set, largest-first ranking
-    // no longer starves tighter clusters. Default UNSET = no cap. Only
-    // meaningful on `recombine`.
-    maxClusterSize: positiveInt.optional(),
-    // #632 — recombine process: tag values that must never form a tag cluster
-    // (generic project-wide tags). Default UNSET/[]. Only meaningful on
-    // `recombine`.
-    excludeTags: z.array(z.string().min(1)).optional(),
-    // #632 — recombine process: entity_norm values that must never form an
-    // entity cluster (user counterpart to the built-in generic-entity filter).
-    // Default UNSET/[]. Only meaningful on `recombine`.
-    excludeEntities: z.array(z.string().min(1)).optional(),
-    // #609 — recombine process: relatedness signal used to form clusters
-    // (tags | graph | both). Clustering is by relatedness, never embedding
-    // similarity. Default "both" (#632). Only meaningful on `recombine`.
-    relatednessSource: z.enum(["tags", "graph", "both"]).optional(),
-    // #609 — recombine process: consecutive re-inductions required before a
-    // hypothesis is promoted to a lesson. Default 2. Only meaningful on
-    // `recombine`.
-    confirmThreshold: z.number().int().min(1).optional(),
-    // #615 — procedural process: minimum number of distinct assets sharing the
-    // same successful normalized ordered-action sequence before it is compiled
-    // into a workflow proposal. Default 3. Only meaningful on `procedural`.
-    minRecurrence: z.number().int().min(2).optional(),
-    // #615 — procedural process: hard cap on workflow proposals emitted per run
-    // (one bounded LLM call each). Default 3. Only meaningful on `procedural`.
-    maxProposalsPerRun: positiveInt.optional(),
-    // #615 — procedural process: asset type a compiled sequence is emitted as.
-    // Reserved; v1 always emits "workflow". Only meaningful on `procedural`.
-    emitAs: z.enum(["workflow", "skill"]).optional(),
-    // #639 — semantic value-floor filter for the `reflect` process. When
-    // enabled, proposals classified as "low-value" by the deterministic noise
-    // gate are deferred. DEFAULT OFF (absent / { enabled: false } = pre-#639
-    // byte-identical behaviour). Only meaningful on the `reflect` process.
-    lowValueFilter: z.object({ enabled: z.boolean().optional() }).passthrough().optional(),
-    // #641 — procedural-aware floor for the `extract` process triage gate.
-    // When true, a session must have markers>=1 OR editCommit>=0.5 to pass, even
-    // if score>=minScore. DEFAULT OFF (absent/false = pre-#641 byte-identical).
-    // Only meaningful on the `extract` process when triage is also enabled.
-    proceduralAwareFloor: z.boolean().optional(),
-    // Triage process config (only meaningful for the `triage` process)
-    applyMode: z.enum(["queue", "promote"]).optional(),
-    policy: z.string().min(1).optional(),
-    maxAcceptsPerRun: positiveInt.optional(),
-    maxDiffLines: positiveInt.optional(),
-    rejectEmpty: z.boolean().optional(),
-    judgment: z
-        .object({
-        mode: z.enum(["llm", "agent", "sdk"]).optional(),
-        profile: z.string().min(1).optional(),
-        timeoutMs: z.union([positiveInt, z.null()]).optional(),
-    })
-        .passthrough()
-        .optional(),
-})
-    .passthrough();
-const ImproveProfileProcessesSchema = z
-    .object({
-    reflect: ImproveProcessConfigSchema.optional(),
-    distill: ImproveProcessConfigSchema.optional(),
-    consolidate: ImproveProcessConfigSchema.optional(),
-    memoryInference: ImproveProcessConfigSchema.optional(),
-    graphExtraction: ImproveProcessConfigSchema.optional(),
-    extract: ImproveProcessConfigSchema.optional(),
-    validation: ImproveProcessConfigSchema.optional(),
-    triage: ImproveProcessConfigSchema.optional(),
-    proactiveMaintenance: ImproveProcessConfigSchema.optional(),
-    recombine: ImproveProcessConfigSchema.optional(),
-    procedural: ImproveProcessConfigSchema.optional(),
-})
-    .passthrough()
-    .superRefine((val, ctx) => {
-    // 0.8.0 removed the duplicated `feedbackDistillation` process key — it was
-    // a thin wrapper around `processes.distill.enabled`. Keep the migration
-    // hint so a stale config gets an actionable message rather than silently
-    // doing nothing. All OTHER unknown process keys are tolerated (passthrough)
-    // — see the unknown-key policy in this file's header. Hard-rejecting them
-    // turned benign cross-version skew into INVALID_CONFIG_FILE failures.
-    if ("feedbackDistillation" in val) {
-        ctx.addIssue({
-            code: z.ZodIssueCode.custom,
-            message: "feedbackDistillation was removed in 0.8.0 — use processes.distill.enabled instead. " +
-                "It now controls both the orchestration gate and the LLM-call gate.",
-        });
+const RETIRED_SOURCE_SHAPE_KEY_MESSAGES = {
+    stashDir: "stashDir is retired in 0.9; the stash path now comes from `bundles`. Run `akm-migrate apply` to convert a pre-0.9 config, or see `akm config path --all` / `akm info`.",
+};
+export const AkmConfigSchema = AkmConfigBaseSchema.superRefine((config, ctx) => {
+    const raw = config;
+    for (const key of ["profiles", "llm", "agent", "features", "stashes"]) {
+        if (key in raw) {
+            ctx.addIssue({
+                code: z.ZodIssueCode.custom,
+                path: [key],
+                message: `${key} is retired in 0.9; configure engines and improve.strategies instead`,
+            });
+        }
     }
-});
-export const ImproveProfileConfigSchema = z
-    .object({
-    description: z.string().min(1).optional(),
-    processes: ImproveProfileProcessesSchema.optional(),
-    autoAccept: nonNegativeNumber.optional(),
-    limit: positiveInt.optional(),
-    // #616 — bounded multi-cycle phasing. Number of prep->loop->post-loop
-    // cycles per run. positiveInt forbids 0/negative. DEFAULT 1 => byte-identical
-    // single-pass behavior.
-    maxCycles: positiveInt.optional(),
-    // #614 — symmetric valence weighting in the eligibility sort. When true,
-    // the attention term becomes |valence| MAGNITUDE so BOTH strong positive
-    // and strong negative feedback drive attention (utility stays dominant) and
-    // strong-signed assets are routed to a fix/reinforce lane. DEFAULT OFF —
-    // false/absent preserves the legacy negative-only ranking byte-for-byte.
-    symmetricValence: z.boolean().optional(),
-    sync: z
-        .object({
-        enabled: z.boolean().optional(),
-        push: z.boolean().optional(),
-        message: z.string().min(1).optional(),
-    })
-        .passthrough()
-        .optional(),
-})
-    .passthrough();
-// ── Profiles / defaults ────────────────────────────────────────────────────
-export const ProfilesSchema = z
-    .object({
-    llm: z.record(z.string(), LlmProfileConfigSchema).optional(),
-    agent: z.record(z.string(), AgentProfileConfigSchema).optional(),
-    improve: z.record(z.string(), ImproveProfileConfigSchema).optional(),
-})
-    .passthrough();
-export const DefaultsSchema = z
-    .object({
-    llm: z.string().min(1).optional(),
-    agent: z.string().min(1).optional(),
-    improve: z.string().min(1).optional(),
-})
-    .passthrough();
-// ── Sources / registries / installed ────────────────────────────────────────
-const SourceConfigEntryOptionsSchema = z
-    .object({
-    /**
-     * @deprecated 0.9.0 (issue #507). Retired per-asset push-on-commit. Kept so
-     * old configs still parse; its intent maps onto the batch push gate and
-     * encountering it emits a one-time deprecation warning.
-     */
-    pushOnCommit: z.boolean().optional(),
-})
-    .passthrough();
-export const SourceConfigEntrySchema = z
-    .object({
-    type: nonEmptyString,
-    path: z.string().min(1).optional(),
-    url: z.string().min(1).optional(),
-    name: z.string().min(1).optional(),
-    enabled: z.boolean().optional(),
-    writable: z.boolean().optional(),
-    primary: z.boolean().optional(),
-    options: SourceConfigEntryOptionsSchema.optional(),
-    wikiName: z.string().min(1).optional(),
-})
-    .passthrough()
-    .superRefine((entry, ctx) => {
-    if (entry.writable === true && (entry.type === "website" || entry.type === "npm")) {
+    // `bindings` (spec §10.1) is Tier B — never emitted, never accepted. The
+    // top-level schema is `.passthrough()`, so without this it would round-trip
+    // silently; reject it loudly so a stray/hand-written bindings block is caught.
+    if ("bindings" in raw) {
         ctx.addIssue({
             code: z.ZodIssueCode.custom,
-            message: `writable: true is only supported on filesystem and git sources (got "${entry.type}"` +
-                (entry.name ? ` on source "${entry.name}"` : "") +
-                ").",
+            path: ["bindings"],
+            message: "bindings is not supported in 0.9.0 (Tier B); it is neither emitted nor accepted",
         });
     }
-});
-export const RegistryConfigEntrySchema = z
-    .object({
-    url: httpUrl,
-    name: z.string().min(1).optional(),
-    enabled: z.boolean().optional(),
-    provider: z.string().min(1).optional(),
-    options: z.record(z.unknown()).optional(),
-})
-    .passthrough();
-const KitSourceSchema = z.enum(["filesystem", "git", "npm", "github", "website", "local"]);
-export const InstalledStashEntrySchema = z
-    .object({
-    id: nonEmptyString,
-    source: KitSourceSchema,
-    ref: nonEmptyString,
-    artifactUrl: nonEmptyString,
-    stashRoot: nonEmptyString,
-    cacheDir: nonEmptyString,
-    installedAt: nonEmptyString,
-    writable: z.boolean().optional(),
-    resolvedVersion: z.string().min(1).optional(),
-    resolvedRevision: z.string().min(1).optional(),
-    wikiName: z.string().min(1).optional(),
-})
-    .passthrough()
-    .superRefine((entry, ctx) => {
-    if (entry.writable === true && entry.source !== "git" && entry.source !== "filesystem") {
+    // 0.9.0 config-shape cutover (spec §10.1): the retired `stashDir`/`sources`/
+    // `installed` trio is HARD-REJECTED at load whenever present — `bundles` +
+    // `defaultBundle` fully supersede it. A pre-cutover config never loads through
+    // this validated path; the migrator ({@link migrateConfigSourcesToBundles})
+    // normalizes old→bundles BEFORE validation, and `inspectConfig` classifies an
+    // old-shape-alone config "old" (migration-eligible) via that same normalize.
+    for (const key of ["stashDir", "sources", "installed"]) {
+        if (key in raw && raw[key] !== undefined) {
+            ctx.addIssue({
+                code: z.ZodIssueCode.custom,
+                path: [key],
+                message: RETIRED_SOURCE_SHAPE_KEY_MESSAGES[key] ??
+                    `${key} is the retired pre-cutover source shape; run \`akm-migrate apply\` to convert it to bundles`,
+            });
+        }
+    }
+    if ("writable" in raw) {
         ctx.addIssue({
             code: z.ZodIssueCode.custom,
-            message: `writable: true is only supported on filesystem and git sources (got "${entry.source}" on installed entry "${entry.id}").`,
+            path: ["writable"],
+            message: "top-level writable is not supported; configure bundles.<id>.writable instead",
         });
     }
-});
-// ── Output ──────────────────────────────────────────────────────────────────
-export const OutputConfigSchema = z
-    .object({
-    format: z.enum(["json", "yaml", "text"]).optional(),
-    detail: z.enum(["brief", "normal", "full"]).optional(),
-})
-    .passthrough();
-// ── Search ──────────────────────────────────────────────────────────────────
-const SearchGraphBoostSchema = z
-    .object({
-    directBoostPerEntity: nonNegativeNumber.optional(),
-    directBoostCap: nonNegativeNumber.optional(),
-    hopBoostPerEntity: nonNegativeNumber.optional(),
-    hopBoostCap: nonNegativeNumber.optional(),
-    /** Hard-capped at 3; values > 3 hard-error so users see the typo. */
-    maxHops: positiveInt.max(3).optional(),
-    confidenceMode: z.enum(["off", "blend", "multiply"]).default("blend").optional(),
-    /** Range [0, 1]; values > 1 hard-error (no silent clamp). */
-    confidenceWeight: z.number().finite().min(0).max(1).default(0.2).optional(),
-})
-    .passthrough();
-export const SearchConfigSchema = z
-    .object({
-    minScore: nonNegativeNumber.optional(),
-    defaultExcludeTypes: z.array(nonEmptyString).optional(),
-    graphBoost: SearchGraphBoostSchema.optional(),
-})
-    .passthrough();
-// ── Feedback ────────────────────────────────────────────────────────────────
-export const FeedbackConfigSchema = z
-    .object({
-    requireReason: z.boolean().optional(),
-    allowedFailureModes: z.array(nonEmptyString).optional(),
-})
-    .passthrough();
-// ── Improve top-level (utility decay, event retention) ─────────────────────
-const ImproveUtilityDecaySchema = z
-    .object({
-    halfLifeDays: z.number().finite().min(0.1).optional(),
-    feedbackStabilityBoost: z.number().finite().min(1).optional(),
-})
-    .passthrough();
-// #612 / WS-4 — auto-accept gate calibration + bounded, opt-in per-phase
-// threshold auto-tune. DEFAULT OFF: when absent (or `autoTune: false`) no
-// tuning occurs, so the gate behaves byte-identically to today.
-// WS-4 adds: per-phase persistence (state.db) + auto-tune ceiling default 85.
-const ImproveCalibrationSchema = z
-    .object({
-    /** Master switch for the bounded threshold auto-tune. Default false (parity). */
-    autoTune: z.boolean().optional(),
-    /** Lower bound (0-100) the tuned threshold may never drop below. */
-    minThreshold: z.number().int().min(0).max(100).optional(),
-    /**
-     * Upper bound (0-100) the tuned threshold may never rise above.
-     * WS-4 default: 85 (prevents gate converging to pure exploitation).
-     */
-    maxThreshold: z.number().int().min(0).max(100).optional(),
-    /** Maximum adjustment magnitude (points) applied in one tune step. */
-    maxStep: positiveInt.optional(),
-    /** Minimum acted-on sample count required before any adjustment. */
-    minSamples: nonNegativeNumber.optional(),
-    /** Target realized accept rate in [0, 1]. Default 0.9. */
-    targetAcceptRate: z.number().finite().min(0).max(1).optional(),
-})
-    .passthrough();
-// WS-4 — exploration budget: a fixed fraction of proposals accepted per run
-// regardless of confidence. DEFAULT OFF.
-const ImproveExplorationSchema = z
-    .object({
-    /**
-     * Enable the exploration budget lane. Default false (parity).
-     * When true, a fraction of proposals are accepted regardless of confidence.
-     */
-    enabled: z.boolean().optional(),
-    /**
-     * Fraction of proposals per run to accept as exploration [0, 1].
-     * Default 0.05 (5%). Clamped to [0, 1] at read time.
-     */
-    budgetFraction: z.number().finite().min(0).max(1).optional(),
-})
-    .passthrough();
-const ImproveSalienceSchema = z
-    .object({
-    /**
-     * WS-2 Part-V gate: enable the outcome-weight term in the salience projection.
-     * Default false (parity — WS-1 weights w_e=0.30, w_r=0.70 until Part-V confirms
-     * no regression). Set to true after running scripts/akm-eval + health report.
-     */
-    outcomeWeightEnabled: z.boolean().optional(),
-    /**
-     * Minimum encoding salience score [0, 1] for a zero-feedback asset to be
-     * admitted to the high-salience improve lane (#608).
-     * Default 0.75. Set to 1.0 to disable the lane entirely.
-     */
-    salienceThreshold: z.number().min(0).max(1).optional(),
-    /**
-     * Per-run additive replay budget (#610). Up to this many top-salience refs are
-     * revisited even with no reactive signal and regardless of cooldown. Additive
-     * on top of --limit. Default 0 = no replay.
-     */
-    replayBudget: z.number().int().min(0).optional(),
-})
-    .passthrough();
-// R5 — longitudinal collapse/churn detector (observe-only in v1; deterministic,
-// fail-open, runs only on cycles where consolidate/recombine did work).
-// Default ON; opt out via `improve.collapseDetector.enabled: false`.
-// See docs/design/improve-collapse-churn-detector-design.md.
-const ImproveCollapseDetectorSchema = z
-    .object({
-    enabled: z.boolean().optional(),
-    // Canary set size minted on first run (owner-approved 30–50 range; default 40).
-    canaryCount: z.number().int().min(3).max(200).optional(),
-    // Top-K cutoff for canary recall/nDCG (default 10).
-    k: z.number().int().min(1).max(100).optional(),
-    // Trend window in qualifying cycles (default 5).
-    windowCycles: z.number().int().min(2).max(50).optional(),
-    // Absolute mean-recall drop vs window median that fires collapse (default 0.15).
-    recallDropThreshold: z.number().min(0).max(1).optional(),
-    // distinct-content-ratio decline over the window that fires collapse (default 0.05).
-    entropyDropThreshold: z.number().min(0).max(1).optional(),
-    // Accepted-action volume over the window below which churn never fires (default 25).
-    churnMinAcceptedActions: z.number().int().min(1).optional(),
-    // improve_cycle_metrics retention (default 365 days, owner-approved).
-    retentionDays: z.number().int().min(1).optional(),
-})
-    .passthrough();
-export const ImproveConfigSchema = z
-    .object({
-    utilityDecay: ImproveUtilityDecaySchema.optional(),
-    eventRetentionDays: nonNegativeNumber.optional(),
-    calibration: ImproveCalibrationSchema.optional(),
-    exploration: ImproveExplorationSchema.optional(),
-    salience: ImproveSalienceSchema.optional(),
-    collapseDetector: ImproveCollapseDetectorSchema.optional(),
-})
-    .passthrough();
-// ── Index / per-pass ────────────────────────────────────────────────────────
-const GRAPH_EXTRACTION_INCLUDE_TYPES_ALLOWED = [
-    "memory",
-    "knowledge",
-    "skill",
-    "command",
-    "agent",
-    "workflow",
-    "lesson",
-    "task",
-    "wiki",
-    "fact",
-];
-const INDEX_PASS_PROVIDER_KEYS = new Set([
-    "endpoint",
-    "model",
-    "provider",
-    "apiKey",
-    "baseUrl",
-    "temperature",
-    "maxTokens",
-    "capabilities",
-]);
-const INDEX_PASS_KNOWN_KEYS = new Set([
-    "llm",
-    "graphExtractionBatchSize",
-    "graphExtractionIncludeTypes",
-    "memoryInferenceBatchSize",
-    "lazyGraphExtraction",
-]);
-/**
- * Per-pass `index.<pass>` entry. Uses preprocess + manual validation so we can
- * emit the legacy parser's targeted error messages ("Duplicate LLM provider
- * configuration", "Unknown key `index.<pass>.<key>`", "expected a boolean")
- * instead of Zod's generic `Unrecognized key` / `Expected boolean, received
- * string` strings — keeps `akm` startup errors actionable.
- */
-export const IndexPassConfigSchema = z.preprocess((raw, ctx) => {
-    if (typeof raw !== "object" || raw === null || Array.isArray(raw)) {
-        return raw; // let z.object below produce the type error
-    }
-    const obj = raw;
-    for (const key of Object.keys(obj)) {
-        if (INDEX_PASS_PROVIDER_KEYS.has(key)) {
+    // `defaultBundle`, when present, must name a configured bundle.
+    if (config.defaultBundle !== undefined) {
+        if (config.bundles === undefined) {
             ctx.addIssue({
                 code: z.ZodIssueCode.custom,
-                message: `Duplicate LLM provider configuration: \`${[...(ctx.path ?? []), key].join(".")}\` is not allowed. ` +
-                    "Configure provider/model/endpoint under `profiles.llm` only; per-pass entries support `{ llm: false }` opt-out.",
+                path: ["defaultBundle"],
+                message: "defaultBundle requires a bundles map",
             });
-            return raw;
         }
-        if (!INDEX_PASS_KNOWN_KEYS.has(key)) {
+        else if (!(config.defaultBundle in config.bundles)) {
             ctx.addIssue({
                 code: z.ZodIssueCode.custom,
-                message: `Unknown key \`${[...(ctx.path ?? []), key].join(".")}\`. Per-pass entries support \`llm\` ` +
-                    "(boolean opt-out), `graphExtractionBatchSize`, `graphExtractionIncludeTypes`, " +
-                    "`memoryInferenceBatchSize`, and `lazyGraphExtraction`.",
+                path: ["defaultBundle"],
+                message: `defaultBundle "${config.defaultBundle}" does not name a configured bundle`,
             });
-            return raw;
         }
     }
-    if ("llm" in obj && typeof obj.llm !== "boolean") {
+    for (const key of ["llm", "agent", "improve"]) {
+        if (config.defaults && key in config.defaults) {
+            ctx.addIssue({
+                code: z.ZodIssueCode.custom,
+                path: ["defaults", key],
+                message: `defaults.${key} is retired in 0.9`,
+            });
+        }
+    }
+    for (const [name, engine] of Object.entries(config.engines ?? {})) {
+        if (engine.kind === "agent" && engine.llmEngine) {
+            const fallback = config.engines?.[engine.llmEngine];
+            if (!fallback || fallback.kind !== "llm") {
+                ctx.addIssue({
+                    code: z.ZodIssueCode.custom,
+                    path: ["engines", name, "llmEngine"],
+                    message: "llmEngine must name an LLM engine",
+                });
+            }
+        }
+    }
+    const defaultEngine = config.defaults?.engine;
+    if (defaultEngine && !config.engines?.[defaultEngine]) {
         ctx.addIssue({
             code: z.ZodIssueCode.custom,
-            message: `Invalid \`${[...(ctx.path ?? []), "llm"].join(".")}\`: expected a boolean (true to use the default LLM profile, false to opt out). Got ${typeof obj.llm}.`,
+            path: ["defaults", "engine"],
+            message: "engine does not name a configured engine",
         });
-        return raw;
     }
-    return raw;
-}, z
-    .object({
-    llm: z.boolean().optional(),
-    graphExtractionBatchSize: positiveInt.optional(),
-    graphExtractionIncludeTypes: z.array(z.enum(GRAPH_EXTRACTION_INCLUDE_TYPES_ALLOWED)).nonempty().optional(),
-    memoryInferenceBatchSize: positiveInt.optional(),
-    lazyGraphExtraction: z.boolean().optional(),
-})
-    .passthrough());
-const MetadataEnhanceSchema = z.object({ enabled: z.boolean().optional() }).passthrough();
-/**
- * RETIRED (meta-review 10-Q3): the staleness-detect pass was deleted; nothing
- * reads this section anymore. The key stays TOLERATED here so configs that
- * still carry `index.stalenessDetection` (written by 0.8.x migrations) do not
- * fail validation — deleting the key would route it into the per-pass
- * catchall, which rejects its `enabled`/`thresholdDays` fields.
- */
-const StalenessDetectionSchema = z
-    .object({
-    enabled: z.boolean().optional(),
-    thresholdDays: positiveInt.optional(),
-})
-    .passthrough();
-/**
- * Index config is a union of reserved feature sections and per-pass entries.
- * Passthrough so per-pass entries (keyed by arbitrary pass names like `graph`,
- * `enrichment`) can live next to the reserved keys.
- *
- * The outer preprocess emits the legacy parser's actionable error messages
- * for the two most common type-shape mistakes:
- *   - An array at the `index` block.
- *   - A non-object at `index.<passName>`.
- * Inner field validation (graphExtractionIncludeTypes enum, llm boolean,
- * provider-key rejection) is delegated to {@link IndexPassConfigSchema}.
- */
-export const IndexConfigSchema = z.preprocess((raw, ctx) => {
-    if (raw === undefined || raw === null)
-        return raw;
-    if (Array.isArray(raw)) {
+    const defaultLlm = config.defaults?.llmEngine;
+    if (defaultLlm && config.engines?.[defaultLlm]?.kind !== "llm") {
         ctx.addIssue({
             code: z.ZodIssueCode.custom,
-            message: 'Invalid `index` config: expected an object keyed by pass name (e.g. `{ "enrichment": { "llm": false } }`).',
+            path: ["defaults", "llmEngine"],
+            message: "llmEngine must name an LLM engine",
         });
-        return raw;
     }
-    if (typeof raw !== "object")
-        return raw;
-    for (const [passName, value] of Object.entries(raw)) {
-        if (typeof value !== "object" || value === null || Array.isArray(value)) {
-            ctx.addIssue({
-                code: z.ZodIssueCode.custom,
-                message: `Invalid \`index.${passName}\` config: expected an object like \`{ "llm": false }\`.`,
-            });
-            return raw;
-        }
-        if (passName !== "metadataEnhance" &&
-            passName !== "stalenessDetection" &&
-            Array.isArray(value.graphExtractionIncludeTypes)) {
-            const arr = value.graphExtractionIncludeTypes;
-            const invalid = [];
-            for (const t of arr) {
-                if (typeof t === "string" &&
-                    !GRAPH_EXTRACTION_INCLUDE_TYPES_ALLOWED.includes(t.toLowerCase())) {
-                    invalid.push(t);
-                }
+    const defaultStrategy = config.defaults?.improveStrategy;
+    if (defaultStrategy &&
+        !BUILTIN_IMPROVE_STRATEGY_NAMES.includes(defaultStrategy) &&
+        !config.improve?.strategies?.[defaultStrategy]) {
+        ctx.addIssue({
+            code: z.ZodIssueCode.custom,
+            path: ["defaults", "improveStrategy"],
+            message: "improveStrategy does not name a built-in or configured strategy",
+        });
+    }
+    for (const [strategyName, strategy] of Object.entries(config.improve?.strategies ?? {})) {
+        const strategyEngine = strategy.engine;
+        if (strategyEngine) {
+            const engine = config.engines?.[strategyEngine];
+            if (!engine || engine.kind !== "llm") {
+                ctx.addIssue({
+                    code: z.ZodIssueCode.custom,
+                    path: ["improve", "strategies", strategyName, "engine"],
+                    message: engine ? "strategy engine must be an LLM engine" : "engine does not name a configured engine",
+                });
             }
-            if (invalid.length > 0) {
+        }
+        for (const [processName, process] of Object.entries(strategy.processes ?? {})) {
+            const processConfig = process;
+            const capability = IMPROVE_PROCESS_ENGINE_CAPABILITIES[processName];
+            if (processConfig.engine && capability === null) {
                 ctx.addIssue({
                     code: z.ZodIssueCode.custom,
-                    message: `Invalid \`index.${passName}.graphExtractionIncludeTypes\`: unsupported type(s): ${invalid.join(", ")}.`,
+                    path: ["improve", "strategies", strategyName, "processes", processName, "engine"],
+                    message: `${processName} does not dispatch an engine`,
                 });
-                return raw;
+            }
+            else {
+                const processEngine = processConfig.engine ?? strategyEngine;
+                if (processEngine && capability === "llm") {
+                    const engine = config.engines?.[processEngine];
+                    if (!engine || engine.kind !== "llm") {
+                        ctx.addIssue({
+                            code: z.ZodIssueCode.custom,
+                            path: ["improve", "strategies", strategyName, "processes", processName, "engine"],
+                            message: engine ? `${processName} requires an LLM engine` : "engine does not name a configured engine",
+                        });
+                    }
+                }
+                else if (processConfig.engine && capability === "runner" && !config.engines?.[processConfig.engine]) {
+                    ctx.addIssue({
+                        code: z.ZodIssueCode.custom,
+                        path: ["improve", "strategies", strategyName, "processes", processName, "engine"],
+                        message: "engine does not name a configured engine",
+                    });
+                }
+            }
+            const judgmentEngine = processConfig.judgment?.engine;
+            if (judgmentEngine) {
+                const engine = config.engines?.[judgmentEngine];
+                if (!engine) {
+                    ctx.addIssue({
+                        code: z.ZodIssueCode.custom,
+                        path: ["improve", "strategies", strategyName, "processes", processName, "judgment", "engine"],
+                        message: "engine does not name a configured engine",
+                    });
+                }
             }
         }
     }
-    return raw;
-}, z
-    .object({
-    metadataEnhance: MetadataEnhanceSchema.optional(),
-    stalenessDetection: StalenessDetectionSchema.optional(),
-})
-    .catchall(IndexPassConfigSchema));
-// ── Setup-derived recommendations ──────────────────────────────────────────
-/**
- * Cron-style schedule hints derived by `akm setup --reset-recommended`.
- *
- * These record the *recommended* cadence for the improve and index background
- * tasks. They are advisory metadata persisted into config so the value
- * survives a re-run; actual task scheduling lives in the tasks subsystem.
- */
-export const SetupTaskSchedulesSchema = z
-    .object({
-    improve: z.string().min(1).optional(),
-    index: z.string().min(1).optional(),
-})
-    .passthrough();
-export const SetupConfigSchema = z
-    .object({
-    taskSchedules: SetupTaskSchedulesSchema.optional(),
-})
-    .passthrough();
-// ── Top-level AkmConfig ────────────────────────────────────────────────────
-/**
- * Base object schema used both as the top-level shape and as the source of
- * truth for {@link listTopLevelConfigKeys}. {@link AkmConfigSchema} wraps this
- * with cross-field refinements (`.superRefine()`).
- *
- * All fields validate loudly — typos and shape errors throw at load time. The
- * legacy parser's warn-and-drop tolerance was a frequent source of silent
- * configuration loss; the migration module ({@link migrateConfigShape}) handles
- * one-time 0.7→0.8 input transforms before the schema sees the value.
- */
-export const AkmConfigShape = {
-    configVersion: z.union([z.string().min(1), z.number()]).optional(),
-    profiles: ProfilesSchema.optional(),
-    defaults: DefaultsSchema.optional(),
-    // Global model-alias tiers: alias → platform → exact model string, with a
-    // reserved `"*"` platform key as fallback. Lets workflows/callers name a
-    // semantic tier ("fast", "deep") that resolves per-harness at dispatch
-    // time. Values are literal model strings, never other aliases (one
-    // resolution level). Platform keys match the platform string a command
-    // builder resolves against ("claude", "opencode", "opencode-sdk", or a
-    // custom profile's name for the default builder) — unknown keys are inert.
-    // Precedence: profile modelAliases > this table > built-in aliases.
-    modelAliases: z.record(z.string().min(1), z.record(z.string().min(1), z.string().min(1))).optional(),
-    stashDir: nonEmptyString.optional(),
-    semanticSearchMode: z.enum(["off", "auto"]).default("auto"),
-    embedding: EmbeddingConnectionConfigSchema.optional(),
-    index: IndexConfigSchema.optional(),
-    // The `installed[]` shape is OWNED by the registry (`InstalledStashEntry`):
-    // its `source` is the 4-value `InstallKind` produced by the registry ref
-    // parser, and installed entries never carry the extra passthrough keys. The
-    // schema still validates entries at runtime, but its OUTPUT type is pinned to
-    // the domain type so config consumers get the registry `InstalledStashEntry`
-    // (not a looser schema-local mirror) — the single-source-of-truth boundary.
-    installed: z.array(InstalledStashEntrySchema).optional(),
-    registries: z.array(RegistryConfigEntrySchema).optional(),
-    sources: z.array(SourceConfigEntrySchema).optional(),
-    output: OutputConfigSchema.optional(),
-    writable: z.boolean().optional(),
-    defaultWriteTarget: nonEmptyString.optional(),
-    search: SearchConfigSchema.optional(),
-    feedback: FeedbackConfigSchema.optional(),
-    archiveRetentionDays: nonNegativeNumber.optional(),
-    improve: ImproveConfigSchema.optional(),
-    setup: SetupConfigSchema.optional(),
-};
-export const AkmConfigBaseSchema = z.object(AkmConfigShape).passthrough();
-export const AkmConfigSchema = AkmConfigBaseSchema.superRefine((config, ctx) => {
-    // #464.a: defaultWriteTarget must name a configured source when sources
-    // are present. With no sources configured, error out instead of silently
-    // accepting (no implicit "first writable" fallback — see locked decision 3).
+    // #464.a: defaultWriteTarget must name a configured source. 0.9.0 (spec
+    // §10.1): sources are `bundles` keys, so it must name a bundle. With no
+    // bundles configured, error out instead of silently accepting (no implicit
+    // "first writable" fallback — see locked decision 3).
     if (config.defaultWriteTarget !== undefined) {
-        const knownNames = (config.sources ?? [])
-            .map((s) => s.name)
-            .filter((n) => typeof n === "string" && n.length > 0);
+        const knownNames = Object.keys(config.bundles ?? {});
         if (knownNames.length === 0) {
             ctx.addIssue({
                 code: z.ZodIssueCode.custom,
                 path: ["defaultWriteTarget"],
-                message: `defaultWriteTarget "${config.defaultWriteTarget}" cannot be resolved: no sources configured. ` +
-                    "Add at least one entry to `sources` with a matching `name` first.",
+                message: `defaultWriteTarget "${config.defaultWriteTarget}" cannot be resolved: no bundles configured. ` +
+                    "Add at least one entry to `bundles` first.",
             });
         }
         else if (!knownNames.includes(config.defaultWriteTarget)) {
             ctx.addIssue({
                 code: z.ZodIssueCode.custom,
                 path: ["defaultWriteTarget"],
-                message: `defaultWriteTarget "${config.defaultWriteTarget}" does not match any configured source name: ${knownNames.map((n) => `"${n}"`).join(", ")}.`,
+                message: `defaultWriteTarget "${config.defaultWriteTarget}" does not match any configured bundle: ${knownNames.map((n) => `"${n}"`).join(", ")}.`,
             });
         }
     }