@intx/hub-sessions 0.1.2 → 0.3.0

package/dist/workflow-run-kind.js

@@ -0,0 +1,2916 @@
+// KindHandler for the `workflow-run` repo kind.
+//
+// A workflow-run repo holds per-deployment runtime state for one or
+// more in-flight workflow runs. `RepoId.id` is the owning deployment
+// id. The repo's top-level layout is:
+//
+//   - `runs/<runId>/events/<seq>.json` — per-run event log entries.
+//     Each entry is a JSON object whose body carries a `type`
+//     discriminator (the on-disk event vocabulary used by the
+//     workflow-run repo) and a `seq` field that matches the integer
+//     in the filename. Filenames are decimal integers ranging from
+//     `0` upward; the on-disk seq numbering owns the ordering and
+//     the per-blob `seq` field is the redundant cross-check.
+//   - `runs/<runId>/blobs/<sha256-hex>` — content-addressed step
+//     outputs the production `BlobSubstrate` adapter spills here when
+//     a value's JSON-stringified form exceeds the inline-encoding
+//     threshold. The filename is a lowercase 64-character sha256 hex
+//     string; the blob value is opaque bytes. Blobs are append-only
+//     and immutable: any blob present in the prior tree must carry
+//     byte-identical contents in the prospective tree.
+//   - `addresses/<urlEncoded(address)>/inbox/<receivedAt>-<messageId>.json`
+//     — pending inbound mail for the address, FIFO-ordered by the
+//     filename's parsed numeric `receivedAt` prefix (with a
+//     lexicographic messageId tiebreak). The filename keeps the
+//     decimal `<receivedAt>` form unpadded; the substrate sorts by
+//     parsed integer rather than string so the FIFO invariant holds
+//     for non-uniform digit widths (e.g. `99-…` precedes `100-…`).
+//   - `addresses/<urlEncoded(address)>/processing/<receivedAt>-<messageId>.json`
+//     — messages currently being handled. Same filename shape and
+//     JSON envelope as the inbox entry; a `dequeueToProcessing`
+//     commit atomically removes the inbox entry and adds the
+//     processing entry preserving the filename key.
+//   - `addresses/<urlEncoded(address)>/consumed/<messageId>.json` —
+//     dedup index keyed by messageId. A `markConsumed` commit
+//     atomically removes the matching processing entry and writes
+//     this dedup entry. The dedup index is bounded by a per-address
+//     retention watermark (see `watermark.json`): a `markConsumed`
+//     commit prunes consumed entries whose `receivedAt` falls below
+//     the watermark so the index reaches a bounded steady state
+//     instead of growing one entry per message forever.
+//   - `addresses/<urlEncoded(address)>/watermark.json` — the
+//     per-address retention watermark. Carries a single
+//     `receivedAt`-horizon value: the oldest `receivedAt` a consumed
+//     entry may still retain. The watermark only ever advances
+//     (monotonic non-decreasing). `enqueueInbox` rejects any inbound
+//     whose `receivedAt` is strictly below the watermark as
+//     definitively-stale (its dedup entry may have been pruned, so a
+//     duplicate cannot be ruled out -- refuse loudly rather than risk
+//     reprocessing). Above the watermark the `consumed/` index is
+//     authoritative; below it, refuse. The watermark advances only as
+//     the prune advances, both under the single writer, so the two
+//     never diverge. The stale-reject applies ONLY to fresh inbound at
+//     `enqueueInbox`; `replayProcessingToInbox` is intentionally exempt
+//     (a recovered in-flight `processing/` entry is already past dedup
+//     -- see that function's note).
+//   - `.gitignore` — supplied by the asset routes' genesis init body.
+//
+// The control-plane subtree (`control/...`) is not part of this
+// commit's surface and has no v1 use case.
+//
+// Event-log invariants enforced at push:
+//   - Each event body's `seq` matches the integer in its filename.
+//   - Per-run event filenames are unique decimal integers (guaranteed
+//     by the tree shape) and validatePush verifies the body's `seq`
+//     field carries the same number, so the on-disk seq sequence and
+//     the per-blob seq cannot diverge.
+//   - Terminal-phase lock: once a run's events include a `RunCompleted`,
+//     `RunFailed`, or `RunCancelled` entry, no event with a strictly
+//     greater seq may appear for the same run.
+//   - Append-only via prior-tree byte comparison: every event blob
+//     that exists at the same path in the parent commit's tree must
+//     match the prospective blob byte-for-byte. Newly-added event
+//     paths (those absent from the prior tree) are accepted. The
+//     substrate exposes the prior tree via `priorReadBlob` /
+//     `priorListDir` on the validatePush args so the constraint is
+//     owned by this handler rather than relying on caller-layer
+//     discipline.
+//   - A `CancelRequested` event must carry an `origin` in the known
+//     set (`self`, `supervisor-drain`, `supervisor-operator`,
+//     `hub-admin`) and a non-empty `reason`.
+//   - Principal-vs-origin enforcement for `CancelRequested`: a
+//     `hub-admin` origin requires the signing principal to be `hub`;
+//     the other three
+//     origins (`self`, `supervisor-drain`, `supervisor-operator`)
+//     require the signing principal to be `supervisor` — the
+//     supervisor signs on the child's behalf for `self`, and signs
+//     for itself on the drain / operator cases. A principal that does
+//     not match the declared origin produces a rejection naming both
+//     sides so a misconfigured writer surfaces at the boundary
+//     rather than as a downstream mystery.
+//
+// Claim-check subtree invariants enforced at push:
+//   - The `<urlEncoded>` segment under `addresses/` must round-trip
+//     cleanly through `decodeURIComponent` followed by
+//     `encodeURIComponent`. A segment that does not round-trip is
+//     rejected so consumers can rely on a single canonical encoding.
+//   - The only entries permitted under an `addresses/<urlEncoded>/`
+//     subtree are the directories `inbox`, `processing`, and
+//     `consumed`, plus the single `watermark.json` file. Other names
+//     under an address fail the push.
+//   - Inbox and processing filenames must match
+//     `<receivedAt>-<messageId>.json` where `receivedAt` is a decimal
+//     epoch-ms integer. The body's `receivedAt` matches the filename
+//     `receivedAt` and the body's `messageId` matches the filename
+//     `messageId`. The body's `address` field must decode to the
+//     URL-encoded segment.
+//   - Consumed filenames must match `<messageId>.json`. The body's
+//     `messageId` matches the filename `messageId`. The body carries
+//     a `consumedBy` run id and the `receivedAt` of the original
+//     consume for audit.
+//   - Atomicity: a given `<messageId>` appears in at most one
+//     filename across `inbox`, `processing`, and `consumed` combined,
+//     per address per prospective commit. Two inbox entries with the
+//     same `<messageId>` but different `<receivedAt>` are rejected as
+//     a same-state collision; the cross-state check fires when the
+//     same messageId appears in inbox+processing, inbox+consumed, or
+//     processing+consumed.
+//   - `consumed/<messageId>.json` bytes are immutable: a prospective
+//     commit that mutates the bytes of a consumed entry RETAINED from
+//     the prior tree is rejected by the same prior-tree byte-equality
+//     guard used for run events. A retained consumed entry may be
+//     DELETED only as a watermark-consistent retention prune (see the
+//     watermark invariants below); any other deletion is rejected.
+//   - Retention prune (the bounded-`consumed/` contract): the consumed
+//     dedup index may shrink only by a watermark-passed prune. A
+//     consumed entry present in the prior tree may be absent from the
+//     prospective tree only when (a) its `receivedAt` is strictly
+//     below the prospective `watermark.json` value (you may prune only
+//     what the watermark passed) and (b) the watermark did not regress
+//     (`prospective watermark >= prior watermark`). A RETAINED entry is
+//     NOT required to sit at or above the watermark: a message consumed
+//     long after receipt (or one replayed back in-flight after a crash)
+//     may legitimately carry a below-watermark `receivedAt` and survive
+//     until a later commit prunes it. Retaining it gives only EXTRA
+//     dedup -- a re-submission at or above the watermark still hits the
+//     entry, one below it is stale-rejected at enqueue -- so it never
+//     weakens exactly-once.
+//   - Inbox→processing transition: a processing entry that is newly
+//     added (not present in the prior tree) must be backed by a
+//     matching inbox entry in the prior tree at the same
+//     `<receivedAt>-<messageId>.json` key. If the prior tree does
+//     not show that inbox entry the transition is rejected so a
+//     direct write into `processing/` cannot bypass the inbox.
+//   - Processing→consumed transition: a consumed entry that is
+//     newly added (not present in the prior tree) must be backed by
+//     a processing entry in the prior tree at the same address with
+//     the same messageId. The receivedAt and messageId carried in
+//     the prior processing envelope must equal the values carried in
+//     the new consumed envelope so the audit trail is unambiguous.
+//
+// Authz:
+//   - `hub` principal: full access.
+//   - `workflow-process` principal: read/write its own deployment's
+//     event log. The principal carries `{ anchorRunId, runId? }`;
+//     this handler verifies `repoId.id === anchorRunId`.
+//   - `supervisor` principal: read/write its own deployment's event
+//     log. The principal carries `{ anchorRunId }`; this handler
+//     verifies `repoId.id === anchorRunId`.
+//   - `sidecar` principal: read-only (createPack, resolveRef) for
+//     resume.
+//   - `user` principal: gated by bearer-token claims and the route
+//     layer's pre-resolved authz verdict, mirroring the convention
+//     used by the other kinds.
+import fs from "node:fs";
+import git from "isomorphic-git";
+import { type } from "arktype";
+import { getLogger } from "@intx/log";
+import { glob, repoActionToGrantVerb } from "@intx/hub-common";
+import { UserPrincipal, } from "./repo-store/index.js";
+import { WORKFLOW_RUN_EVENTS_FILE, splitCombinedEventLog, encodeCombinedEventLog, } from "./workflow-run-event-log.js";
+const logger = getLogger(["hub-sessions", "workflow-run-kind"]);
+export const WORKFLOW_RUN_GITIGNORE_PATH = ".gitignore";
+export const WORKFLOW_RUN_RUNS_PREFIX = "runs";
+export const WORKFLOW_RUN_EVENTS_DIR = "events";
+export const WORKFLOW_RUN_BLOBS_DIR = "blobs";
+export const WORKFLOW_RUN_GRANTS_FILE = "grants.json";
+export const WORKFLOW_RUN_ADDRESSES_PREFIX = "addresses";
+export const WORKFLOW_RUN_CONTROL_PREFIX = "control";
+export const WORKFLOW_RUN_INBOX_DIR = "inbox";
+export const WORKFLOW_RUN_PROCESSING_DIR = "processing";
+export const WORKFLOW_RUN_CONSUMED_DIR = "consumed";
+/**
+ * Filename of the per-address retention watermark blob, a direct child
+ * of `addresses/<urlEncoded>/` (a file, not a directory). Carries the
+ * monotonic `receivedAt`-horizon below which consumed entries may be
+ * pruned and at-or-below which inbound enqueues are refused as stale.
+ */
+export const WORKFLOW_RUN_WATERMARK_FILE = "watermark.json";
+/**
+ * Default retention horizon for the consumed dedup index, in
+ * milliseconds. The boot edge resolves the operator's
+ * `CONSUMED_RETENTION_MS` config to a concrete value and threads it
+ * into `markConsumed`; this default applies only when no operator
+ * value is supplied. 24 hours is the conservative default: long enough
+ * that a duplicate from a retrying upstream within a day is still
+ * deduped by a retained consumed entry, short enough that `consumed/`
+ * reaches a bounded steady state of one day's message volume.
+ *
+ * INVARIANT (operator-owned): the horizon must be >= the longest window in
+ * which the same `messageId` could legitimately be re-submitted and still must
+ * be caught as a duplicate. The hub now redelivers un-acked inbound mail
+ * (connected-window retry and reconnect-redelivery), so an at-least-once
+ * internal source DOES exist -- but the dedup guarantee against it does not
+ * rest on window arithmetic. It rests on a STRUCTURAL fact: `enqueueInbox` is
+ * only ever called with a freshly stamped `receivedAt` (a redelivery
+ * re-enters `onMailMessage` and re-stamps `Date.now()`, never carrying the
+ * original), and the watermark only ever advances to at most
+ * `consumedAt - retentionHorizonMs`, which is <= now, so a fresh `receivedAt`
+ * always sits a full horizon above the watermark and can never be stale-
+ * refused. A redelivery instead hits the `consumed/`/`processing/`/`inbox/`
+ * dedup index and is deduped there. The sole path that carries an original
+ * (old) `receivedAt` back into the queue is `replayProcessingToInbox`, which
+ * writes straight to `inbox/` and bypasses the stale gate entirely. So
+ * `claim_check_stale_enqueue` is unreachable via redelivery today. If any
+ * redelivery source is ever changed to carry the ORIGINAL `receivedAt` into
+ * `enqueueInbox`, stale becomes reachable, the horizon must then be >= that
+ * source's maximum redelivery window, and `StaleInboxEnqueueError`'s
+ * withhold-not-ack handling becomes load-bearing; a breach surfaces LOUDLY (an
+ * old-`receivedAt` re-submission is refused at enqueue) rather than as silent
+ * double-processing.
+ */
+export const DEFAULT_CONSUMED_RETENTION_MS = 24 * 60 * 60 * 1000;
+/**
+ * Per-agent durable conversation-state subtree (design §3c). A
+ * long-lived single-step agent's multi-turn conversation context is
+ * committed under `agent-state/<agentKey>/...` so it survives child
+ * respawn: on respawn the rebuilt warm agent reads its prior
+ * conversation back from here before the resumed run replays.
+ *
+ * Unlike `runs/` (append-only events, immutable blobs) this subtree is
+ * MUTABLE: each run boundary overwrites the agent's conversation
+ * snapshot with the latest turns. It is therefore exempt from the
+ * append-only / deletion-direction walks `runs/` is subject to; the
+ * only push-time constraint is segment shape (a single round-trip-safe
+ * `<agentKey>` directory layer below the prefix).
+ */
+export const WORKFLOW_RUN_AGENT_STATE_PREFIX = "agent-state";
+/**
+ * Allowed top-level entries in the prospective tree. Anything else
+ * fails the push. `control/` has no v1 use and stays absent.
+ */
+const ALLOWED_TOP_LEVEL = new Set([
+    WORKFLOW_RUN_RUNS_PREFIX,
+    WORKFLOW_RUN_ADDRESSES_PREFIX,
+    WORKFLOW_RUN_AGENT_STATE_PREFIX,
+    WORKFLOW_RUN_GITIGNORE_PATH,
+]);
+const CLAIM_CHECK_SUBDIRS = new Set([
+    WORKFLOW_RUN_INBOX_DIR,
+    WORKFLOW_RUN_PROCESSING_DIR,
+    WORKFLOW_RUN_CONSUMED_DIR,
+]);
+/** Per-event filename shape: a decimal integer followed by `.json`. */
+const EVENT_FILENAME_RE = /^(0|[1-9][0-9]*)\.json$/;
+/**
+ * Parse the seq from a per-event log filename `<seq>.json` under
+ * `runs/<runId>/events/`. Returns the non-negative integer seq, or
+ * `null` when the name is not a legal per-event filename. This is the
+ * one place the filename shape is defined; every reader of the event log
+ * narrows names through it rather than re-encoding the regex. Callers
+ * decide what an illegal name means -- a foreign entry to skip, or a
+ * substrate-invariant violation to surface -- since `validatePush` is
+ * the authority that keeps illegal names from landing in the first place.
+ */
+export function parseEventSeq(filename) {
+    const match = EVENT_FILENAME_RE.exec(filename);
+    if (match === null)
+        return null;
+    const seqStr = match[1];
+    if (seqStr === undefined)
+        return null;
+    return Number.parseInt(seqStr, 10);
+}
+/**
+ * Narrow a per-event filename to its seq, throwing when it is illegal.
+ * A reader that enumerates the committed event log to act on its entries
+ * uses this rather than `parseEventSeq`: `validatePush` is the authority
+ * that keeps an illegal name from ever landing under
+ * `runs/<runId>/events/`, so a name that reaches a reader is corruption,
+ * and silently skipping it would drop an event from processing. `context`
+ * is the repo-root-relative blob path, surfaced in the error so the
+ * offending entry is identifiable.
+ */
+export function requireEventSeq(filename, context) {
+    const seq = parseEventSeq(filename);
+    if (seq === null) {
+        throw new Error(`event_filename_invalid: ${context}`);
+    }
+    return seq;
+}
+/**
+ * Per-blob filename shape for the `runs/<runId>/blobs/` subtree: a
+ * lowercase 64-character sha256 hex string. Pins the regex to the key
+ * the production `BlobSubstrate` adapter computes via `sha256Hex` so a
+ * non-canonical key (uppercase hex, truncated digest, alternate
+ * encoding) fails the push at the boundary rather than landing
+ * silently.
+ */
+const BLOB_FILENAME_RE = /^[0-9a-f]{64}$/;
+/**
+ * Entries the kind handler accepts under `runs/<runId>/`. The `events/`
+ * subtree carries the append-only event log; the `blobs/` subtree carries
+ * opaque, content-addressed step outputs the `BlobSubstrate` adapter spills
+ * there when a value exceeds the inline-encoding threshold; `grants.json`
+ * carries the run's authorization grants, delivered by the hub's
+ * `run.grants` frame ahead of the trigger and read back by the sidecar's
+ * `onRunStart` barrier. The grants file is a run-dir sibling of `events/`,
+ * not part of the event log, so the event-shape and blob-immutability walks
+ * treat it as inert.
+ */
+const RUN_DIR_ALLOWED_CHILDREN = new Set([
+    WORKFLOW_RUN_EVENTS_DIR,
+    WORKFLOW_RUN_BLOBS_DIR,
+    // A terminated run's event log, sealed from the per-event `events/`
+    // files into one combined file by a compaction commit.
+    WORKFLOW_RUN_EVENTS_FILE,
+    WORKFLOW_RUN_GRANTS_FILE,
+]);
+/**
+ * Filename shape for inbox and processing entries:
+ * `<receivedAt>-<messageId>.json`. `receivedAt` is a decimal integer
+ * (epoch ms); `messageId` is captured as the rest of the basename and
+ * is validated separately against the body's `messageId`.
+ */
+const QUEUE_FILENAME_RE = /^(0|[1-9][0-9]*)-(.+)\.json$/;
+/** Filename shape for consumed entries: `<messageId>.json`. */
+const CONSUMED_FILENAME_RE = /^(.+)\.json$/;
+/**
+ * JSON envelope carried by inbox and processing entries. Keys:
+ *   - `messageId`: dedup key for the inbound message.
+ *   - `receivedAt`: epoch-ms timestamp the reactor accepted the
+ *     message; sortable FIFO key prefix.
+ *   - `address`: decoded canonical address (not URL-encoded).
+ *   - `mailAuditRef`: pointer to the raw mail bytes in the mail-audit
+ *     store. For the in-process single-agent path a separate
+ *     `MailAuditStore` holds the authoritative bytes and this ref joins
+ *     onto it.
+ *   - `rawMessage`: base64 of the inbound mail's raw MIME bytes,
+ *     inlined so the workflow-process child can read its step input by
+ *     messageId at `trigger.fired` time. The supervisor is the sole
+ *     mail owner under the unified-execution host (§3a); it has no
+ *     separate durable byte store the child can read, so the bytes ride
+ *     the claim-check envelope itself. Present whenever the supervisor
+ *     enqueued the entry; omitted by callers that only stamp the audit
+ *     ref. The bytes survive the inbox→processing transition verbatim
+ *     (the dequeue copies the entry bytes), so a `trigger.fired` for a
+ *     processing entry can always recover the input.
+ */
+const ClaimCheckEnvelope = type({
+    messageId: "string > 0",
+    receivedAt: "number >= 0",
+    address: "string > 0",
+    mailAuditRef: {
+        store: "string > 0",
+        path: "string > 0",
+    },
+    "rawMessage?": "string > 0",
+    "+": "ignore",
+});
+/**
+ * JSON envelope carried by consumed entries. The consumed entry is the
+ * canonical dedup index keyed by messageId; the envelope preserves
+ * the originating receivedAt for audit and carries the runId that
+ * consumed the message.
+ */
+const ConsumedEnvelope = type({
+    messageId: "string > 0",
+    receivedAt: "number >= 0",
+    address: "string > 0",
+    runId: "string > 0",
+    consumedAt: "number >= 0",
+    mailAuditRef: {
+        store: "string > 0",
+        path: "string > 0",
+    },
+    "rejection?": {
+        code: "string > 0",
+        message: "string > 0",
+    },
+    "+": "ignore",
+});
+/**
+ * JSON envelope carried by the per-address `watermark.json` blob. The
+ * `watermark` is a `receivedAt` horizon (epoch ms): the oldest
+ * `receivedAt` a consumed entry may still retain. It only ever
+ * advances. A retention prune drops consumed entries strictly below
+ * it; `enqueueInbox` refuses any inbound strictly below it.
+ */
+const WatermarkEnvelope = type({
+    watermark: "number >= 0",
+    "+": "ignore",
+});
+/**
+ * Terminal event discriminators mapped to the `workflow_run.status` value
+ * each settles the run into. A run whose log contains an entry with one of
+ * these `type` values must not receive any event with a strictly greater
+ * seq.
+ *
+ * This map is a hand-rolled copy of the runtime's terminal-run vocabulary
+ * (`isTerminalRunPhase` in `@intx/workflow` state-machine `state.ts`,
+ * re-exported from the state-machine index and consumed by `transition.ts`),
+ * duplicated here because `@intx/hub-sessions` must not depend on
+ * `@intx/workflow`. It is the sole authority for that vocabulary and MUST
+ * stay in sync with the canonical runtime definition:
+ * if the runtime adds or removes a terminal run phase, update this map too.
+ * Drift silently reopens the restore-time double-driver collision that
+ * `readOwnedMessageIds` (below) exists to prevent.
+ */
+const TERMINAL_EVENT_STATUS = new Map([
+    ["RunCompleted", "completed"],
+    ["RunFailed", "failed"],
+    ["RunCancelled", "cancelled"],
+]);
+/**
+ * Membership set of terminal event types, derived from
+ * `TERMINAL_EVENT_STATUS` so it always covers exactly the mapped types and
+ * the two cannot drift apart.
+ */
+const TERMINAL_EVENT_TYPES = new Set(TERMINAL_EVENT_STATUS.keys());
+/**
+ * Recognised CancelRequested origins. Mirrors the workflow package's
+ * `CANCEL_ORIGINS` vocabulary; inlined here so the substrate does
+ * not depend on `@intx/workflow`.
+ */
+const CANCEL_REQUESTED_ORIGINS = new Set([
+    "self",
+    "supervisor-drain",
+    "supervisor-operator",
+    "hub-admin",
+]);
+/**
+ * Per-origin signing-principal kind. `hub-admin` is the only origin
+ * a `hub` principal may mint; the other three originate inside the
+ * supervisor's trust boundary (the supervisor signs `self` on behalf
+ * of the workflow-process since the child has no asymmetric keypair,
+ * and signs the `supervisor-drain` / `supervisor-operator` audit-
+ * distinction
+ * cases for itself). Lookup misses fail the push.
+ */
+const CANCEL_ORIGIN_TO_PRINCIPAL_KIND = new Map([
+    ["self", "supervisor"],
+    ["supervisor-drain", "supervisor"],
+    ["supervisor-operator", "supervisor"],
+    ["hub-admin", "hub"],
+]);
+/**
+ * Cross-event shape carried by every blob committed under
+ * `runs/<runId>/events/`. The discriminator field on disk is `type`,
+ * matching the convention used by the substrate's `subscribeKind`
+ * helper and the workflow-host scheduler.
+ */
+const EventEnvelope = type({
+    type: "string",
+    seq: "number >= 0",
+    "+": "ignore",
+});
+/**
+ * Structural validator for the `CancelRequested` payload's
+ * cancellation-specific fields. The kind handler verifies the origin
+ * is a known CancelOrigin and the reason is a non-empty string; the
+ * principal-vs-origin map collapses because every origin is
+ * supervisor-signed in this design.
+ */
+const CancelRequestedFields = type({
+    origin: "string",
+    reason: "string > 0",
+    "+": "ignore",
+});
+const SidecarPrincipal = type({
+    kind: "'sidecar'",
+    agentId: "string",
+});
+const WorkflowProcessPrincipal = type({
+    kind: "'workflow-process'",
+    anchorRunId: "string",
+    "runId?": "string",
+});
+const SupervisorPrincipal = type({
+    kind: "'supervisor'",
+    anchorRunId: "string",
+});
+/**
+ * Resolve the substrate's `changedPathPrefixes` into the set of run ids
+ * the commit could have touched, or `undefined` to validate every run.
+ *
+ * Returns `undefined` (validate-all) when the substrate could not bound
+ * the change set, or when a change prefix reaches into `runs/` without
+ * naming a specific run (`runs` or `runs/` alone). A change prefix that
+ * never touches `runs/` -- e.g. a claim-check write under `addresses/`
+ * -- contributes no run ids; an empty result set means the commit
+ * touched no run, so the per-run walks legitimately validate nothing.
+ */
+function runScopeFromChangedPrefixes(changedPathPrefixes) {
+    if (changedPathPrefixes === undefined)
+        return undefined;
+    const runsPrefix = `${WORKFLOW_RUN_RUNS_PREFIX}/`;
+    const runIds = new Set();
+    for (const prefix of changedPathPrefixes) {
+        if (prefix === WORKFLOW_RUN_RUNS_PREFIX || prefix === runsPrefix) {
+            // The `runs/` subtree changed but the substrate could not name
+            // which run; fall back to validating every run.
+            return undefined;
+        }
+        if (!prefix.startsWith(runsPrefix))
+            continue;
+        const rest = prefix.slice(runsPrefix.length);
+        const slash = rest.indexOf("/");
+        if (slash <= 0)
+            return undefined;
+        runIds.add(rest.slice(0, slash));
+    }
+    return runIds;
+}
+/**
+ * Build the (runId → events[]) map by walking the prospective tree.
+ * The substrate's listDir yields names directly under the given
+ * directory, so the walk is `runs/` → run-id subdirs → `events/` →
+ * event filenames. Filenames outside the `<seq>.json` shape fail the
+ * push. When `scopeRunIds` is supplied, only those runs are walked --
+ * see the substrate's `changedPathPrefixes` contract.
+ */
+async function enumerateEventBlobs(listDir, scopeRunIds) {
+    const runs = new Map();
+    // When the substrate bounds the commit's change set to a specific set
+    // of runs, walk only those `runs/<runId>/` directories instead of
+    // listing every run. An untouched run is carried forward
+    // byte-identical by the substrate's prefix-preserving commit, so its
+    // per-run invariants -- already validated when it was written --
+    // cannot change. `scopeRunIds` may name a run absent from the tree
+    // (e.g. a prior-tree walk for a run the prospective tree dropped);
+    // `listDir` on a missing directory returns `[]`, which the
+    // empty-children guards below handle.
+    const runIds = scopeRunIds === undefined
+        ? await listDir(WORKFLOW_RUN_RUNS_PREFIX)
+        : Array.from(scopeRunIds);
+    for (const runId of runIds) {
+        const runDirPath = `${WORKFLOW_RUN_RUNS_PREFIX}/${runId}`;
+        const runChildren = await listDir(runDirPath);
+        // A scoped run id can name a run that is absent from the tree being
+        // walked -- the substrate's change set is the union of prospective
+        // and prior touched runs, so the prospective walk may receive a run
+        // that exists only in the prior tree (and vice versa). An absent run
+        // directory lists as empty; skip it here so only runs actually
+        // present in this tree are validated. The unscoped walk never
+        // reaches this branch because its run ids come from listing the
+        // present `runs/` directory.
+        if (scopeRunIds !== undefined && runChildren.length === 0)
+            continue;
+        const offender = runChildren.find((c) => !RUN_DIR_ALLOWED_CHILDREN.has(c));
+        if (offender !== undefined) {
+            return {
+                ok: false,
+                reason: `run directory ${runDirPath} contains unexpected entry ${JSON.stringify(offender)}; only "${WORKFLOW_RUN_EVENTS_DIR}", "${WORKFLOW_RUN_BLOBS_DIR}", "${WORKFLOW_RUN_EVENTS_FILE}", and "${WORKFLOW_RUN_GRANTS_FILE}" are allowed`,
+            };
+        }
+        const hasCombined = runChildren.includes(WORKFLOW_RUN_EVENTS_FILE);
+        const hasPerEvent = runChildren.includes(WORKFLOW_RUN_EVENTS_DIR);
+        if (hasCombined && hasPerEvent) {
+            return {
+                ok: false,
+                reason: `run directory ${runDirPath} carries both a combined "${WORKFLOW_RUN_EVENTS_FILE}" and a per-event "${WORKFLOW_RUN_EVENTS_DIR}" subtree`,
+            };
+        }
+        // A sealed (combined) run carries no per-event entries; it is validated
+        // by the combined-form path, not this per-event enumeration.
+        if (hasCombined)
+            continue;
+        if (!hasPerEvent) {
+            // A run dir whose only child is `grants.json` is the legitimate
+            // pre-first-event window: the hub's `run.grants` frame writes the
+            // grants ahead of the trigger, so the grants file lands before the
+            // child emits its first event. Carry it forward untouched -- there is
+            // no event log to enumerate yet. Any other events-less shape (e.g. a
+            // bare `blobs/` with no events) remains rejected below.
+            if (runChildren.length === 1 &&
+                runChildren[0] === WORKFLOW_RUN_GRANTS_FILE) {
+                continue;
+            }
+            return {
+                ok: false,
+                reason: `run directory ${runDirPath} is missing required "${WORKFLOW_RUN_EVENTS_DIR}" subdirectory`,
+            };
+        }
+        const eventsDirPath = `${runDirPath}/${WORKFLOW_RUN_EVENTS_DIR}`;
+        const filenames = await listDir(eventsDirPath);
+        const entries = [];
+        for (const filename of filenames) {
+            const match = EVENT_FILENAME_RE.exec(filename);
+            if (match === null) {
+                return {
+                    ok: false,
+                    reason: `event filename ${eventsDirPath}/${filename} does not match <seq>.json`,
+                };
+            }
+            const seqStr = match[1];
+            if (seqStr === undefined) {
+                return {
+                    ok: false,
+                    reason: `event filename ${eventsDirPath}/${filename} produced no seq capture`,
+                };
+            }
+            entries.push({
+                runId,
+                filename,
+                filenameSeq: Number.parseInt(seqStr, 10),
+                blobPath: `${eventsDirPath}/${filename}`,
+            });
+        }
+        entries.sort((a, b) => a.filenameSeq - b.filenameSeq);
+        runs.set(runId, entries);
+    }
+    return { ok: true, runs };
+}
+/**
+ * Validate the prospective tree's combined-form (sealed) runs and return
+ * the set of run ids that legitimately carry a combined `events.jsonl`.
+ * The deletion-direction guard uses that set to allow a run's per-event
+ * files to disappear when (and only when) they were folded into the
+ * combined file under this same validation.
+ *
+ * Three prior states are accepted:
+ *   - prior already combined  -> the sealed file is immutable; prospective
+ *     bytes must equal prior bytes.
+ *   - prior per-event         -> the compaction transition; the combined
+ *     file must be the byte-for-byte fold of the prior per-event blobs in
+ *     seq order. This is the audit-integrity boundary: a loose check here
+ *     would let compaction silently rewrite history.
+ *   - prior absent            -> a freshly-delivered sealed run (e.g. a
+ *     pack receive); its own structure is validated.
+ */
+async function validateCombinedEventRuns(listDir, readBlob, priorListDir, priorReadBlob, scopeRunIds) {
+    const combinedRunIds = new Set();
+    const runIds = scopeRunIds === undefined
+        ? await listDir(WORKFLOW_RUN_RUNS_PREFIX)
+        : Array.from(scopeRunIds);
+    for (const runId of runIds) {
+        const runDirPath = `${WORKFLOW_RUN_RUNS_PREFIX}/${runId}`;
+        const children = await listDir(runDirPath);
+        if (!children.includes(WORKFLOW_RUN_EVENTS_FILE))
+            continue;
+        const combinedPath = `${runDirPath}/${WORKFLOW_RUN_EVENTS_FILE}`;
+        const combinedBytes = await readBlob(combinedPath);
+        const content = new TextDecoder().decode(combinedBytes);
+        const priorChildren = await priorListDir(runDirPath);
+        if (priorChildren.includes(WORKFLOW_RUN_EVENTS_FILE)) {
+            // Sealed once, immutable thereafter.
+            const immutable = await checkPriorByteEquality(combinedPath, readBlob, priorReadBlob);
+            if (!immutable.ok)
+                return immutable;
+        }
+        else if (priorChildren.includes(WORKFLOW_RUN_EVENTS_DIR)) {
+            const structure = checkCombinedStructure(runId, combinedPath, content);
+            if (!structure.ok)
+                return structure;
+            const fold = await checkCompactionFold(runId, runDirPath, combinedBytes, priorListDir, priorReadBlob);
+            if (!fold.ok)
+                return fold;
+        }
+        else {
+            const structure = checkCombinedStructure(runId, combinedPath, content);
+            if (!structure.ok)
+                return structure;
+        }
+        combinedRunIds.add(runId);
+    }
+    return { ok: true, combinedRunIds };
+}
+/**
+ * The audit-integrity bridge. A compaction commit replaces a run's prior
+ * `events/<seq>.json` files with one combined file; this asserts the
+ * combined file reproduces those prior blobs' bytes verbatim, in seq
+ * order, with nothing added, dropped, reordered, or mutated. It rebuilds
+ * the expected combined bytes from the prior tree through the same encoder
+ * the writer uses, so the two cannot drift, and compares for exact
+ * equality.
+ */
+async function checkCompactionFold(runId, runDirPath, combinedBytes, priorListDir, priorReadBlob) {
+    const priorEventsDir = `${runDirPath}/${WORKFLOW_RUN_EVENTS_DIR}`;
+    const priorEntries = [];
+    for (const filename of await priorListDir(priorEventsDir)) {
+        const match = EVENT_FILENAME_RE.exec(filename);
+        if (match === null || match[1] === undefined) {
+            return {
+                ok: false,
+                reason: `prior event filename ${priorEventsDir}/${filename} does not match <seq>.json; cannot validate compaction of run ${runId}`,
+            };
+        }
+        priorEntries.push({
+            seq: Number.parseInt(match[1], 10),
+            path: `${priorEventsDir}/${filename}`,
+        });
+    }
+    priorEntries.sort((a, b) => a.seq - b.seq);
+    const priorBlobs = [];
+    for (const entry of priorEntries) {
+        const bytes = await priorReadBlob(entry.path);
+        if (bytes === null) {
+            return {
+                ok: false,
+                reason: `prior event ${entry.path} is unreadable; cannot validate compaction of run ${runId}`,
+            };
+        }
+        priorBlobs.push(bytes);
+    }
+    // Byte equality, not decoded-string equality: each event is signed over
+    // its own bytes, so the sealed file must be the verbatim concatenation
+    // of the prior blobs, not merely decode-equivalent to it.
+    const expected = encodeCombinedEventLog(priorBlobs);
+    const sameBytes = combinedBytes.byteLength === expected.byteLength &&
+        combinedBytes.every((b, i) => b === expected[i]);
+    if (!sameBytes) {
+        return {
+            ok: false,
+            reason: `run ${runId} compaction does not fold its prior events verbatim: ${runDirPath}/${WORKFLOW_RUN_EVENTS_FILE} must equal the run's prior events/<seq>.json blobs joined in seq order`,
+        };
+    }
+    return { ok: true };
+}
+/**
+ * Validate a combined event log's own structure: every line a valid event
+ * envelope, contiguous seqs, exactly one terminal event and it is last
+ * (so a sealed run is genuinely terminal). Used for a sealed run with no
+ * prior per-event form to bridge against.
+ */
+function checkCombinedStructure(runId, combinedPath, content) {
+    const lines = splitCombinedEventLog(content);
+    if (lines.length === 0) {
+        return { ok: false, reason: `combined event log ${combinedPath} is empty` };
+    }
+    let baseSeq = null;
+    let terminalSeq = null;
+    for (const [i, line] of lines.entries()) {
+        let body;
+        try {
+            body = JSON.parse(line);
+        }
+        catch {
+            return {
+                ok: false,
+                reason: `combined event log ${combinedPath} line ${String(i)} is not valid JSON`,
+            };
+        }
+        const validated = EventEnvelope(body);
+        if (validated instanceof type.errors) {
+            return {
+                ok: false,
+                reason: `combined event log ${combinedPath} line ${String(i)} envelope invalid: ${validated.summary}`,
+            };
+        }
+        if (baseSeq === null) {
+            baseSeq = validated.seq;
+        }
+        else if (validated.seq !== baseSeq + i) {
+            return {
+                ok: false,
+                reason: `combined event log ${combinedPath} has a sequence gap at line ${String(i)} (expected seq ${String(baseSeq + i)}, got ${String(validated.seq)})`,
+            };
+        }
+        if (terminalSeq !== null) {
+            return {
+                ok: false,
+                reason: `combined event log ${combinedPath} has an event at seq ${String(validated.seq)} after terminal at seq ${String(terminalSeq)}`,
+            };
+        }
+        if (TERMINAL_EVENT_TYPES.has(validated.type)) {
+            terminalSeq = validated.seq;
+        }
+    }
+    if (terminalSeq === null) {
+        return {
+            ok: false,
+            reason: `combined event log ${combinedPath} for run ${runId} has no terminal event; only a terminated run is sealed`,
+        };
+    }
+    return { ok: true };
+}
+/**
+ * Walk every `runs/<runId>/blobs/` directory and validate each blob
+ * filename matches the sha256-hex shape the production `BlobSubstrate`
+ * adapter writes. The `blobs/` subdirectory itself is optional: a run
+ * that has not yet spilled an output to a blob never produces a
+ * `blobs/` directory, and a run with only inline-encoded outputs never
+ * will. Returns the flat list of blob entries so the caller can apply
+ * immutability checks against the prior tree.
+ */
+async function enumerateRunBlobs(listDir, scopeRunIds) {
+    const out = [];
+    // See enumerateEventBlobs: a defined `scopeRunIds` walks only the
+    // commit's touched runs; an untouched run's blobs are carried forward
+    // byte-identical and were validated when written.
+    const runIds = scopeRunIds === undefined
+        ? await listDir(WORKFLOW_RUN_RUNS_PREFIX)
+        : Array.from(scopeRunIds);
+    for (const runId of runIds) {
+        const runDirPath = `${WORKFLOW_RUN_RUNS_PREFIX}/${runId}`;
+        const runChildren = await listDir(runDirPath);
+        if (!runChildren.includes(WORKFLOW_RUN_BLOBS_DIR))
+            continue;
+        const blobsDirPath = `${runDirPath}/${WORKFLOW_RUN_BLOBS_DIR}`;
+        const filenames = await listDir(blobsDirPath);
+        for (const filename of filenames) {
+            if (!BLOB_FILENAME_RE.test(filename)) {
+                return {
+                    ok: false,
+                    reason: `blob filename ${blobsDirPath}/${filename} does not match a lowercase 64-character sha256 hex string`,
+                };
+            }
+            out.push({
+                runId,
+                filename,
+                blobPath: `${blobsDirPath}/${filename}`,
+            });
+        }
+    }
+    return { ok: true, blobs: out };
+}
+/**
+ * Enforce blob immutability via prior-tree byte equality. The blob
+ * value itself is opaque bytes (no JSON envelope, no arktype
+ * validation); the only structural rule beyond filename shape is that
+ * a blob entry present in the prior tree must carry byte-identical
+ * contents in the prospective tree. Mirrors the consumed-entry
+ * discipline in the claim-check subtree.
+ */
+async function checkBlobPriorByteEquality(blobPath, readBlob, priorReadBlob) {
+    const prior = await priorReadBlob(blobPath);
+    if (prior === null)
+        return { ok: true };
+    const prospective = await readBlob(blobPath);
+    if (prior.byteLength !== prospective.byteLength) {
+        return {
+            ok: false,
+            reason: `blob ${blobPath} bytes diverge from the prior tree (lengths ${String(prior.byteLength)} vs ${String(prospective.byteLength)}); blob entries are immutable once written`,
+        };
+    }
+    for (let i = 0; i < prior.byteLength; i++) {
+        if (prior[i] !== prospective[i]) {
+            return {
+                ok: false,
+                reason: `blob ${blobPath} bytes diverge from the prior tree at offset ${String(i)}; blob entries are immutable once written`,
+            };
+        }
+    }
+    return { ok: true };
+}
+async function parseEventBlob(entry, readBlob) {
+    let raw;
+    try {
+        raw = await readBlob(entry.blobPath);
+    }
+    catch (cause) {
+        return {
+            ok: false,
+            reason: `event ${entry.blobPath} could not be read from the tree: ${cause instanceof Error ? cause.message : String(cause)}`,
+        };
+    }
+    let body;
+    try {
+        body = JSON.parse(new TextDecoder().decode(raw));
+    }
+    catch (cause) {
+        return {
+            ok: false,
+            reason: `event ${entry.blobPath} is not valid JSON: ${cause instanceof Error ? cause.message : String(cause)}`,
+        };
+    }
+    const validated = EventEnvelope(body);
+    if (validated instanceof type.errors) {
+        return {
+            ok: false,
+            reason: `event ${entry.blobPath} envelope invalid: ${validated.summary}`,
+        };
+    }
+    if (validated.seq !== entry.filenameSeq) {
+        return {
+            ok: false,
+            reason: `event ${entry.blobPath} body.seq ${String(validated.seq)} does not match filename seq ${String(entry.filenameSeq)}`,
+        };
+    }
+    if (validated.type === "CancelRequested") {
+        const cancelFields = CancelRequestedFields(body);
+        if (cancelFields instanceof type.errors) {
+            return {
+                ok: false,
+                reason: `event ${entry.blobPath} CancelRequested payload invalid: ${cancelFields.summary}`,
+            };
+        }
+        if (!CANCEL_REQUESTED_ORIGINS.has(cancelFields.origin)) {
+            return {
+                ok: false,
+                reason: `event ${entry.blobPath} CancelRequested origin ${JSON.stringify(cancelFields.origin)} is not a recognised CancelOrigin`,
+            };
+        }
+    }
+    return { ok: true, parsed: { entry, body: validated } };
+}
+/**
+ * Compare the prospective bytes of `blobPath` against the bytes at
+ * the same path in the prior tree. Returns `{ ok: true }` when the
+ * blob is newly added (no prior entry) or when the prior and
+ * prospective bytes are byte-identical; returns a rejection otherwise.
+ * Surfaces append-only at the handler scope: the event log invariant
+ * lives here rather than relying on caller-layer discipline at
+ * `writeTreePreservingPrefix`.
+ */
+async function checkPriorByteEquality(blobPath, readBlob, priorReadBlob) {
+    const prior = await priorReadBlob(blobPath);
+    if (prior === null)
+        return { ok: true };
+    const prospective = await readBlob(blobPath);
+    if (prior.byteLength !== prospective.byteLength) {
+        return {
+            ok: false,
+            reason: `event ${blobPath} bytes diverge from the prior tree (lengths ${String(prior.byteLength)} vs ${String(prospective.byteLength)}); event blobs are append-only`,
+        };
+    }
+    for (let i = 0; i < prior.byteLength; i++) {
+        if (prior[i] !== prospective[i]) {
+            return {
+                ok: false,
+                reason: `event ${blobPath} bytes diverge from the prior tree at offset ${String(i)}; event blobs are append-only`,
+            };
+        }
+    }
+    return { ok: true };
+}
+/**
+ * Round-trip an `<urlEncoded(address)>` segment through decode then
+ * encode. A divergence means the segment is not the canonical
+ * encoding of any address, which would leave consumers guessing
+ * which encoding to use when reading the subtree. Surface as a
+ * concrete rejection at push time.
+ */
+function checkAddressSegmentRoundTrip(segment) {
+    let decoded;
+    try {
+        decoded = decodeURIComponent(segment);
+    }
+    catch (cause) {
+        return {
+            ok: false,
+            reason: `address segment ${JSON.stringify(segment)} is not a valid URL-encoded string: ${cause instanceof Error ? cause.message : String(cause)}`,
+        };
+    }
+    const reencoded = encodeURIComponent(decoded);
+    if (reencoded !== segment) {
+        return {
+            ok: false,
+            reason: `address segment ${JSON.stringify(segment)} does not round-trip URL-encoding (re-encoded as ${JSON.stringify(reencoded)})`,
+        };
+    }
+    return { ok: true, decoded };
+}
+/**
+ * FIFO comparator for inbox/processing entries. Sorts by the parsed
+ * numeric `receivedAt` (filename prefix); ties break on the
+ * messageId tail. The numeric compare is the load-bearing piece —
+ * lexicographic compare on `<receivedAt>-…` filenames with
+ * non-uniform digit widths disagrees with chronological order
+ * (e.g. `"100-…"` < `"99-…"` because `'1' < '9'`).
+ */
+function compareQueueEntries(a, b) {
+    const aReceivedAt = a.receivedAtFromFilename;
+    const bReceivedAt = b.receivedAtFromFilename;
+    if (aReceivedAt === null || bReceivedAt === null) {
+        throw new Error("compareQueueEntries: queue entries must carry a parsed receivedAt");
+    }
+    if (aReceivedAt !== bReceivedAt)
+        return aReceivedAt - bReceivedAt;
+    const aId = a.messageIdFromFilename;
+    const bId = b.messageIdFromFilename;
+    if (aId < bId)
+        return -1;
+    if (aId > bId)
+        return 1;
+    return 0;
+}
+async function enumerateClaimCheckBlobs(listDir, resolveConsumedOid) {
+    const perAddress = new Map();
+    const segments = await listDir(WORKFLOW_RUN_ADDRESSES_PREFIX);
+    for (const segment of segments) {
+        const roundTrip = checkAddressSegmentRoundTrip(segment);
+        if (!roundTrip.ok)
+            return roundTrip;
+        const addrDir = `${WORKFLOW_RUN_ADDRESSES_PREFIX}/${segment}`;
+        const children = await listDir(addrDir);
+        for (const child of children) {
+            if (CLAIM_CHECK_SUBDIRS.has(child))
+                continue;
+            if (child === WORKFLOW_RUN_WATERMARK_FILE)
+                continue;
+            return {
+                ok: false,
+                reason: `address directory ${addrDir} contains unexpected entry ${JSON.stringify(child)}; allowed: "${WORKFLOW_RUN_INBOX_DIR}", "${WORKFLOW_RUN_PROCESSING_DIR}", "${WORKFLOW_RUN_CONSUMED_DIR}", "${WORKFLOW_RUN_WATERMARK_FILE}"`,
+            };
+        }
+        const bucket = perAddress.get(segment) ?? {
+            decodedAddress: roundTrip.decoded,
+            inbox: [],
+            processing: [],
+            consumed: [],
+            watermarkPath: null,
+        };
+        if (children.includes(WORKFLOW_RUN_WATERMARK_FILE)) {
+            bucket.watermarkPath = `${addrDir}/${WORKFLOW_RUN_WATERMARK_FILE}`;
+        }
+        for (const subdir of CLAIM_CHECK_SUBDIRS) {
+            if (!children.includes(subdir))
+                continue;
+            const dirPath = `${addrDir}/${subdir}`;
+            const filenames = await listDir(dirPath);
+            for (const filename of filenames) {
+                if (subdir === WORKFLOW_RUN_INBOX_DIR ||
+                    subdir === WORKFLOW_RUN_PROCESSING_DIR) {
+                    const match = QUEUE_FILENAME_RE.exec(filename);
+                    if (match === null) {
+                        return {
+                            ok: false,
+                            reason: `${subdir} filename ${dirPath}/${filename} does not match <receivedAt>-<messageId>.json`,
+                        };
+                    }
+                    const receivedAtStr = match[1];
+                    const messageId = match[2];
+                    if (receivedAtStr === undefined || messageId === undefined) {
+                        return {
+                            ok: false,
+                            reason: `${subdir} filename ${dirPath}/${filename} produced no captures`,
+                        };
+                    }
+                    const entry = {
+                        kind: subdir === WORKFLOW_RUN_INBOX_DIR ? "inbox" : "processing",
+                        addressSegment: segment,
+                        decodedAddress: roundTrip.decoded,
+                        filename,
+                        receivedAtFromFilename: Number.parseInt(receivedAtStr, 10),
+                        messageIdFromFilename: messageId,
+                        blobPath: `${dirPath}/${filename}`,
+                    };
+                    if (subdir === WORKFLOW_RUN_INBOX_DIR)
+                        bucket.inbox.push(entry);
+                    else
+                        bucket.processing.push(entry);
+                }
+                else {
+                    const match = CONSUMED_FILENAME_RE.exec(filename);
+                    if (match === null) {
+                        return {
+                            ok: false,
+                            reason: `${WORKFLOW_RUN_CONSUMED_DIR} filename ${dirPath}/${filename} does not match <messageId>.json`,
+                        };
+                    }
+                    const messageId = match[1];
+                    if (messageId === undefined) {
+                        return {
+                            ok: false,
+                            reason: `${WORKFLOW_RUN_CONSUMED_DIR} filename ${dirPath}/${filename} produced no message-id capture`,
+                        };
+                    }
+                    const consumedBlobPath = `${dirPath}/${filename}`;
+                    const consumedEntry = {
+                        kind: "consumed",
+                        addressSegment: segment,
+                        decodedAddress: roundTrip.decoded,
+                        filename,
+                        receivedAtFromFilename: null,
+                        messageIdFromFilename: messageId,
+                        blobPath: consumedBlobPath,
+                    };
+                    if (resolveConsumedOid !== undefined) {
+                        consumedEntry.oid = await resolveConsumedOid(consumedBlobPath);
+                    }
+                    bucket.consumed.push(consumedEntry);
+                }
+            }
+        }
+        // FIFO ordering: sort by the parsed numeric receivedAt prefix
+        // with a lexicographic messageId tiebreak. String-sorting the
+        // raw filename would put "99-…" after "100-…" because '9' > '1',
+        // breaking the FIFO invariant for non-uniform digit widths.
+        bucket.inbox.sort(compareQueueEntries);
+        bucket.processing.sort(compareQueueEntries);
+        bucket.consumed.sort((a, b) => a.filename < b.filename ? -1 : a.filename > b.filename ? 1 : 0);
+        perAddress.set(segment, bucket);
+    }
+    return { ok: true, perAddress };
+}
+/**
+ * Read + validate the per-address `watermark.json` value from a blob
+ * reader. `null` means the tree has no watermark blob (treated as
+ * watermark 0 -- no entry pruned, nothing refused). The reader may be
+ * the prospective `readBlob` or the `priorReadBlob` (the latter
+ * returns `null` for an absent path, which is the legitimate
+ * never-pruned genesis state).
+ */
+async function readWatermark(watermarkPath, readBlob) {
+    let raw;
+    try {
+        raw = await readBlob(watermarkPath);
+    }
+    catch (cause) {
+        return {
+            ok: false,
+            reason: `watermark ${watermarkPath} could not be read from the tree: ${cause instanceof Error ? cause.message : String(cause)}`,
+        };
+    }
+    if (raw === null)
+        return { ok: true, watermark: 0 };
+    let bodyJson;
+    try {
+        bodyJson = JSON.parse(new TextDecoder().decode(raw));
+    }
+    catch (cause) {
+        return {
+            ok: false,
+            reason: `watermark ${watermarkPath} is not valid JSON: ${cause instanceof Error ? cause.message : String(cause)}`,
+        };
+    }
+    const validated = WatermarkEnvelope(bodyJson);
+    if (validated instanceof type.errors) {
+        return {
+            ok: false,
+            reason: `watermark ${watermarkPath} envelope invalid: ${validated.summary}`,
+        };
+    }
+    return { ok: true, watermark: validated.watermark };
+}
+async function parseConsumedBlob(entry, readBlob) {
+    return parseConsumedBlobFrom(entry, readBlob);
+}
+/**
+ * Read + validate a consumed entry's envelope from a blob reader that
+ * may return `null` for an absent path (the `priorReadBlob` shape).
+ * `null` is treated as a read failure: the caller only passes a path
+ * the prior tree is known to carry, so a `null` is structural damage.
+ */
+async function parseConsumedBlobFrom(entry, readBlob) {
+    let raw;
+    try {
+        raw = await readBlob(entry.blobPath);
+    }
+    catch (cause) {
+        return {
+            ok: false,
+            reason: `consumed ${entry.blobPath} could not be read from the tree: ${cause instanceof Error ? cause.message : String(cause)}`,
+        };
+    }
+    if (raw === null) {
+        return {
+            ok: false,
+            reason: `consumed ${entry.blobPath} was enumerated in the tree but its bytes could not be read`,
+        };
+    }
+    let bodyJson;
+    try {
+        bodyJson = JSON.parse(new TextDecoder().decode(raw));
+    }
+    catch (cause) {
+        return {
+            ok: false,
+            reason: `consumed ${entry.blobPath} is not valid JSON: ${cause instanceof Error ? cause.message : String(cause)}`,
+        };
+    }
+    const validated = ConsumedEnvelope(bodyJson);
+    if (validated instanceof type.errors) {
+        return {
+            ok: false,
+            reason: `consumed ${entry.blobPath} envelope invalid: ${validated.summary}`,
+        };
+    }
+    if (validated.messageId !== entry.messageIdFromFilename) {
+        return {
+            ok: false,
+            reason: `consumed ${entry.blobPath} body.messageId ${JSON.stringify(validated.messageId)} does not match filename messageId ${JSON.stringify(entry.messageIdFromFilename)}`,
+        };
+    }
+    if (validated.address !== entry.decodedAddress) {
+        return {
+            ok: false,
+            reason: `consumed ${entry.blobPath} body.address ${JSON.stringify(validated.address)} does not match decoded address segment ${JSON.stringify(entry.decodedAddress)}`,
+        };
+    }
+    return { ok: true, body: validated };
+}
+async function parseQueueBlob(entry, readBlob) {
+    let raw;
+    try {
+        raw = await readBlob(entry.blobPath);
+    }
+    catch (cause) {
+        return {
+            ok: false,
+            reason: `${entry.kind} ${entry.blobPath} could not be read from the tree: ${cause instanceof Error ? cause.message : String(cause)}`,
+        };
+    }
+    let bodyJson;
+    try {
+        bodyJson = JSON.parse(new TextDecoder().decode(raw));
+    }
+    catch (cause) {
+        return {
+            ok: false,
+            reason: `${entry.kind} ${entry.blobPath} is not valid JSON: ${cause instanceof Error ? cause.message : String(cause)}`,
+        };
+    }
+    const validated = ClaimCheckEnvelope(bodyJson);
+    if (validated instanceof type.errors) {
+        return {
+            ok: false,
+            reason: `${entry.kind} ${entry.blobPath} envelope invalid: ${validated.summary}`,
+        };
+    }
+    if (validated.messageId !== entry.messageIdFromFilename) {
+        return {
+            ok: false,
+            reason: `${entry.kind} ${entry.blobPath} body.messageId ${JSON.stringify(validated.messageId)} does not match filename messageId ${JSON.stringify(entry.messageIdFromFilename)}`,
+        };
+    }
+    if (validated.receivedAt !== entry.receivedAtFromFilename) {
+        return {
+            ok: false,
+            reason: `${entry.kind} ${entry.blobPath} body.receivedAt ${String(validated.receivedAt)} does not match filename receivedAt ${String(entry.receivedAtFromFilename)}`,
+        };
+    }
+    if (validated.address !== entry.decodedAddress) {
+        return {
+            ok: false,
+            reason: `${entry.kind} ${entry.blobPath} body.address ${JSON.stringify(validated.address)} does not match decoded address segment ${JSON.stringify(entry.decodedAddress)}`,
+        };
+    }
+    return { ok: true, body: validated };
+}
+/**
+ * Compute the git blob OID of a consumed entry from a byte reader,
+ * used only when the delta-scoped path lacks a substrate-provided prior
+ * OID listing (e.g. a hand-built test validatePush). `git.hashBlob`
+ * reproduces the same content-addressed OID a `git.readTree` listing
+ * carries, so the delta path's intersection compare is identical
+ * whether the OID came from the tree listing or from hashing the bytes.
+ */
+async function hashConsumedBlobOid(bytes) {
+    const { oid } = await git.hashBlob({ object: bytes });
+    return oid;
+}
+/**
+ * Resolve each consumed entry's git blob OID for the delta-scoped path.
+ * When the substrate supplies a directory OID listing (`listDirOids`) the
+ * OID comes straight from the tree — one `readTree` per consumed
+ * directory, cached — so that side is not re-read blob-by-blob. When the
+ * listing is absent (a hand-built validatePush in a unit test) each OID
+ * falls back to hashing the entry's bytes, which preserves identical
+ * semantics at O(retained) cost. Both the prior and prospective sides use
+ * this; `sideLabel` distinguishes them in the missing-OID error.
+ */
+function makeListingOidResolver(sideLabel, listDirOids, hashFallback) {
+    const dirOidCache = new Map();
+    return async (blobPath) => {
+        if (listDirOids !== undefined) {
+            const slash = blobPath.lastIndexOf("/");
+            const dir = blobPath.slice(0, slash);
+            const name = blobPath.slice(slash + 1);
+            let byName = dirOidCache.get(dir);
+            if (byName === undefined) {
+                byName = new Map();
+                for (const entry of await listDirOids(dir)) {
+                    byName.set(entry.name, entry.oid);
+                }
+                dirOidCache.set(dir, byName);
+            }
+            const oid = byName.get(name);
+            if (oid === undefined) {
+                throw new Error(`delta claim-check: ${sideLabel} tree listing has no OID for enumerated consumed entry ${blobPath}`);
+            }
+            return oid;
+        }
+        return hashFallback(blobPath);
+    };
+}
+function makePriorConsumedOidResolver(priorReadBlob, priorListDirOids) {
+    return makeListingOidResolver("prior", priorListDirOids, async (blobPath) => {
+        const bytes = await priorReadBlob(blobPath);
+        if (bytes === null) {
+            throw new Error(`delta claim-check: consumed entry ${blobPath} was enumerated in the prior tree but its bytes could not be read`);
+        }
+        return hashConsumedBlobOid(bytes);
+    });
+}
+/**
+ * Validate the `addresses/<urlEncoded>/{inbox,processing,consumed}`
+ * subtree as a whole. The walk enforces filename shape, JSON envelope
+ * structure, address round-trip, per-messageId atomicity across the
+ * three queue states, consumed-blob immutability, and the
+ * inbox→processing / processing→consumed transition invariants against
+ * the prior tree.
+ *
+ * The consumed dedup index is validated by its per-commit DELTA against
+ * the prior tree rather than by re-walking the whole retained set:
+ * retained entries (same filename, same blob OID) are skipped as
+ * already-validated-and-immutable, added entries are parsed and
+ * validated, and removed entries are checked against the retention
+ * watermark. `priorListDirOids` and `listDirOids`, when supplied by the
+ * substrate, surface the prior and prospective consumed OIDs straight
+ * from their tree listings so neither side is re-read blob-by-blob.
+ */
+async function validateClaimCheckSubtree(listDir, readBlob, priorReadBlob, priorListDir, priorListDirOids, listDirOids) {
+    // Surface each consumed entry's git blob OID during enumeration
+    // straight from the tree listing on both sides when the substrate
+    // provides it, falling back to hashing the bytes otherwise.
+    const prospectiveConsumedOid = makeListingOidResolver("prospective", listDirOids, async (blobPath) => hashConsumedBlobOid(await readBlob(blobPath)));
+    const priorConsumedOid = makePriorConsumedOidResolver(priorReadBlob, priorListDirOids);
+    const enumerated = await enumerateClaimCheckBlobs(listDir, prospectiveConsumedOid);
+    if (!enumerated.ok)
+        return enumerated;
+    const priorEnumerated = await enumerateClaimCheckBlobs(priorListDir, priorConsumedOid);
+    if (!priorEnumerated.ok) {
+        // The prior tree is the committed state — if its claim-check
+        // shape is already broken, surface it with a distinct rejection
+        // prefix so an operator can tell prior-state damage from a
+        // misconfigured push.
+        return {
+            ok: false,
+            reason: `prior tree's claim-check subtree is structurally invalid: ${priorEnumerated.reason}`,
+        };
+    }
+    const emptyBucket = (decodedAddress) => ({
+        decodedAddress,
+        inbox: [],
+        processing: [],
+        consumed: [],
+        watermarkPath: null,
+    });
+    // Iterate the UNION of prospective and prior address segments so a
+    // prospective tree that wipes an address subtree entirely still
+    // runs the prior-retention checks against that segment's
+    // prior-tree consumed/processing entries.
+    const allSegments = new Set([
+        ...enumerated.perAddress.keys(),
+        ...priorEnumerated.perAddress.keys(),
+    ]);
+    for (const segment of allSegments) {
+        const priorBucket = priorEnumerated.perAddress.get(segment);
+        const prospectiveBucketForSegment = enumerated.perAddress.get(segment);
+        const decodedAddress = prospectiveBucketForSegment?.decodedAddress ??
+            priorBucket?.decodedAddress;
+        if (decodedAddress === undefined) {
+            throw new Error(`validateClaimCheckSubtree: segment ${JSON.stringify(segment)} appeared in the union of prospective and prior segments but neither bucket carries a decoded address`);
+        }
+        const bucket = prospectiveBucketForSegment ?? emptyBucket(decodedAddress);
+        // Per-messageId atomicity: each messageId may appear at most
+        // once across inbox/processing/consumed combined. The check
+        // keys on (messageId, kind, filename) so two inbox entries with
+        // the same messageId at different `receivedAt` values surface
+        // as a same-state collision (the Set-of-kinds shape would
+        // collapse both into a single "inbox" member and miss the
+        // case).
+        const messageIdToLocations = new Map();
+        for (const entry of [...bucket.inbox, ...bucket.processing]) {
+            const parsed = await parseQueueBlob(entry, readBlob);
+            if (!parsed.ok)
+                return parsed;
+            const list = messageIdToLocations.get(entry.messageIdFromFilename) ?? [];
+            list.push({ kind: entry.kind, filename: entry.filename });
+            messageIdToLocations.set(entry.messageIdFromFilename, list);
+        }
+        for (const entry of bucket.consumed) {
+            // Cross-state atomicity needs each consumed messageId in the map;
+            // the messageId is the filename stem, so this needs no blob read.
+            // Retained consumed entries are not re-parsed (their envelope was
+            // validated when first written and their bytes are proven
+            // immutable by the OID compare below); added consumed entries are
+            // parsed and validated by the transition check further down.
+            const list = messageIdToLocations.get(entry.messageIdFromFilename) ?? [];
+            list.push({ kind: entry.kind, filename: entry.filename });
+            messageIdToLocations.set(entry.messageIdFromFilename, list);
+        }
+        for (const [messageId, locations] of messageIdToLocations) {
+            if (locations.length > 1) {
+                const sorted = [...locations].sort((a, b) => {
+                    if (a.kind !== b.kind)
+                        return a.kind < b.kind ? -1 : 1;
+                    if (a.filename !== b.filename)
+                        return a.filename < b.filename ? -1 : 1;
+                    return 0;
+                });
+                const kinds = new Set(sorted.map((l) => l.kind));
+                if (kinds.size > 1) {
+                    return {
+                        ok: false,
+                        reason: `address ${JSON.stringify(bucket.decodedAddress)} message ${JSON.stringify(messageId)} appears in multiple queue states ${JSON.stringify(Array.from(kinds).sort())}; at most one of inbox/processing/consumed is permitted`,
+                    };
+                }
+                const kind = sorted[0]?.kind;
+                if (kind === undefined)
+                    throw new Error("unreachable");
+                return {
+                    ok: false,
+                    reason: `address ${JSON.stringify(bucket.decodedAddress)} message ${JSON.stringify(messageId)} appears at multiple ${kind} positions ${JSON.stringify(sorted.map((l) => l.filename))}; at most one entry per messageId is permitted`,
+                };
+            }
+        }
+        // Consumed entries are immutable. Compare the git blob OID the
+        // enumeration surfaced: a consumed entry present in the prior tree
+        // at the same path must carry the same OID (git trees are
+        // content-addressed, so equal OID proves byte-equality without
+        // reading either blob). A diverging OID is an immutability
+        // violation. Immutability is load-bearing for exactly-once: a
+        // mutated `receivedAt` on a retained consumed entry could fake it
+        // below the watermark, get it pruned, and let a re-submission miss
+        // dedup -- so this compare is not optional.
+        const priorConsumedOidByPath = new Map();
+        for (const e of priorBucket?.consumed ?? []) {
+            if (e.oid === undefined) {
+                throw new Error(`delta claim-check: prior consumed entry ${e.blobPath} was enumerated without an OID`);
+            }
+            priorConsumedOidByPath.set(e.blobPath, e.oid);
+        }
+        for (const entry of bucket.consumed) {
+            const priorOid = priorConsumedOidByPath.get(entry.blobPath);
+            if (priorOid === undefined)
+                continue; // newly added; validated below
+            if (entry.oid === undefined) {
+                throw new Error(`delta claim-check: prospective consumed entry ${entry.blobPath} was enumerated without an OID`);
+            }
+            if (entry.oid !== priorOid) {
+                return {
+                    ok: false,
+                    reason: `consumed ${entry.blobPath} bytes diverge from the prior tree (blob OID ${entry.oid} vs ${priorOid}); consumed entries are immutable once written`,
+                };
+            }
+        }
+        const prospectiveConsumedPaths = new Set(bucket.consumed.map((e) => e.blobPath));
+        const prospectiveProcessingPaths = new Set(bucket.processing.map((e) => e.blobPath));
+        const prospectiveInboxByFilename = new Map();
+        const prospectiveInboxPaths = new Set();
+        for (const e of bucket.inbox) {
+            prospectiveInboxByFilename.set(e.filename, e);
+            prospectiveInboxPaths.add(e.blobPath);
+        }
+        const prospectiveProcessingByFilename = new Map();
+        for (const e of bucket.processing)
+            prospectiveProcessingByFilename.set(e.filename, e);
+        const prospectiveConsumedByMessageId = new Map();
+        for (const e of bucket.consumed)
+            prospectiveConsumedByMessageId.set(e.messageIdFromFilename, e);
+        // Deletion-direction guards: walk every entry the prior tree
+        // carried under `consumed/`, `processing/`, and `inbox/` and reject
+        // any prior path that vanishes from the prospective tree except via
+        // a permitted transition (or, for consumed, a watermark-passed
+        // retention prune). Without this walk a prospective tree that
+        // simply omits a prior entry would slip past the prospective-tree
+        // by-presence checks above.
+        //
+        // Retention-watermark contract for the consumed dedup index. The
+        // watermark is a monotonic `receivedAt` horizon; a `markConsumed`
+        // commit may drop the oldest consumed tail (entries strictly below
+        // the watermark) and the watermark may only advance. Resolve both
+        // the prospective and prior watermark up front so the consumed
+        // deletion check below can bind every drop to the watermark.
+        let prospectiveWatermark = 0;
+        if (bucket.watermarkPath !== null) {
+            const wm = await readWatermark(bucket.watermarkPath, (p) => readBlob(p));
+            if (!wm.ok)
+                return wm;
+            prospectiveWatermark = wm.watermark;
+        }
+        let priorWatermark = 0;
+        if (priorBucket?.watermarkPath != null) {
+            const wm = await readWatermark(priorBucket.watermarkPath, priorReadBlob);
+            if (!wm.ok)
+                return wm;
+            priorWatermark = wm.watermark;
+        }
+        if (prospectiveWatermark < priorWatermark) {
+            return {
+                ok: false,
+                reason: `address ${JSON.stringify(decodedAddress)} retention watermark regressed from ${String(priorWatermark)} to ${String(prospectiveWatermark)}; the watermark is monotonically non-decreasing`,
+            };
+        }
+        if (priorBucket !== undefined) {
+            // The consumed dedup index may shrink only by a watermark-passed
+            // prune: a consumed entry dropped from the prior tree must have a
+            // receivedAt strictly below the prospective watermark (you may
+            // prune only what the watermark passed). Combined with the
+            // already-verified watermark monotonicity, this is the whole of
+            // the exactly-once retention contract: pruning is bound to the
+            // watermark and the watermark only advances.
+            //
+            // The suffix relation (dropped entries older than every retained
+            // entry) is deliberately NOT enforced. A RETAINED entry is NOT
+            // required to sit at or above the watermark: a message consumed
+            // long after receipt (or replayed back in-flight) may
+            // legitimately carry a below-watermark receivedAt and survive
+            // until a later commit prunes it. Holding it gives EXTRA dedup (a
+            // re-submission at or above the watermark still hits the retained
+            // entry; one below is stale-rejected at enqueue), so a hole left
+            // by an out-of-order prune weakens nothing.
+            //
+            // Only the dropped entries are read. A retained entry (present in
+            // both trees) is proven byte-identical by the OID compare above,
+            // so its receivedAt is unchanged and need not be read. The
+            // receivedAt lives in the body; read it from the prior tree
+            // (retained bytes are immutable, so prior and prospective agree).
+            for (const e of priorBucket.consumed) {
+                if (prospectiveConsumedPaths.has(e.blobPath))
+                    continue;
+                const priorParsed = await parseConsumedBlobFrom(e, priorReadBlob);
+                if (!priorParsed.ok)
+                    return priorParsed;
+                const receivedAt = priorParsed.body.receivedAt;
+                if (receivedAt >= prospectiveWatermark) {
+                    return {
+                        ok: false,
+                        reason: `consumed ${e.blobPath} present in the prior tree is missing from the prospective tree but its receivedAt ${String(receivedAt)} is not below the retention watermark ${String(prospectiveWatermark)}; consumed entries may be pruned only once the watermark has passed them`,
+                    };
+                }
+            }
+            for (const e of priorBucket.processing) {
+                if (prospectiveProcessingPaths.has(e.blobPath))
+                    continue;
+                // A processing entry may legitimately disappear in two
+                // shapes: (1) markConsumed wrote a matching consumed entry
+                // keyed by the same messageId, or (2) replayProcessingToInbox
+                // moved the entry back to inbox preserving the
+                // `<receivedAt>-<messageId>.json` filename. Anything else is
+                // an in-flight loss.
+                const consumedMatch = prospectiveConsumedByMessageId.get(e.messageIdFromFilename);
+                const inboxMatch = prospectiveInboxByFilename.get(e.filename);
+                if (consumedMatch !== undefined || inboxMatch !== undefined)
+                    continue;
+                return {
+                    ok: false,
+                    reason: `processing ${e.blobPath} present in the prior tree is missing from the prospective tree without a matching consumed or inbox transition; in-flight processing entries cannot be silently dropped`,
+                };
+            }
+            for (const e of priorBucket.inbox) {
+                if (prospectiveInboxPaths.has(e.blobPath))
+                    continue;
+                // A prior inbox entry may legitimately disappear when it
+                // transitions to processing (same `<receivedAt>-<messageId>`
+                // filename) or directly to consumed (matching messageId).
+                // Anything else is an inbound-mail loss — the FIFO claim-check
+                // contract requires the entry to reappear somewhere.
+                const processingMatch = prospectiveProcessingByFilename.get(e.filename);
+                const consumedMatch = prospectiveConsumedByMessageId.get(e.messageIdFromFilename);
+                if (processingMatch !== undefined || consumedMatch !== undefined)
+                    continue;
+                return {
+                    ok: false,
+                    reason: `inbox ${e.blobPath} present in the prior tree is missing from the prospective tree without a matching processing or consumed transition; pending inbox entries cannot be silently dropped`,
+                };
+            }
+        }
+        const priorInboxByFilename = new Map();
+        const priorProcessingByMessageId = new Map();
+        if (priorBucket !== undefined) {
+            for (const e of priorBucket.inbox)
+                priorInboxByFilename.set(e.filename, e);
+            for (const e of priorBucket.processing)
+                priorProcessingByMessageId.set(e.messageIdFromFilename, e);
+        }
+        const priorProcessingPaths = new Set((priorBucket?.processing ?? []).map((e) => e.blobPath));
+        const priorConsumedPaths = new Set((priorBucket?.consumed ?? []).map((e) => e.blobPath));
+        // Newly-added processing entries must match an inbox entry that
+        // existed in the prior tree at the same `<receivedAt>-<messageId>`
+        // filename. This makes inbox→processing the only legal way to
+        // grow processing/.
+        for (const entry of bucket.processing) {
+            if (priorProcessingPaths.has(entry.blobPath))
+                continue;
+            const priorInbox = priorInboxByFilename.get(entry.filename);
+            if (priorInbox === undefined) {
+                return {
+                    ok: false,
+                    reason: `processing ${entry.blobPath} is newly added but the prior tree has no matching inbox entry ${JSON.stringify(`${WORKFLOW_RUN_ADDRESSES_PREFIX}/${segment}/${WORKFLOW_RUN_INBOX_DIR}/${entry.filename}`)}; processing entries must originate from a prior-tree inbox entry`,
+                };
+            }
+        }
+        // Newly-added consumed entries must match a processing entry that
+        // existed in the prior tree at the same address+messageId, and
+        // the receivedAt carried in the consumed envelope must equal the
+        // receivedAt the processing entry's filename carried.
+        for (const entry of bucket.consumed) {
+            if (priorConsumedPaths.has(entry.blobPath))
+                continue;
+            const priorProcessing = priorProcessingByMessageId.get(entry.messageIdFromFilename);
+            if (priorProcessing === undefined) {
+                return {
+                    ok: false,
+                    reason: `consumed ${entry.blobPath} is newly added but the prior tree has no matching processing entry for messageId ${JSON.stringify(entry.messageIdFromFilename)}; consumed entries must originate from a prior-tree processing entry`,
+                };
+            }
+            const parsed = await parseConsumedBlob(entry, readBlob);
+            if (!parsed.ok)
+                return parsed;
+            const consumedBody = parsed.body;
+            if (consumedBody.receivedAt !== priorProcessing.receivedAtFromFilename) {
+                return {
+                    ok: false,
+                    reason: `consumed ${entry.blobPath} body.receivedAt ${String(consumedBody.receivedAt)} does not match the prior processing entry's receivedAt ${String(priorProcessing.receivedAtFromFilename)} for messageId ${JSON.stringify(entry.messageIdFromFilename)}`,
+                };
+            }
+        }
+    }
+    return { ok: true };
+}
+/**
+ * Enforce the Q3 principal-vs-origin map for a parsed
+ * `CancelRequested` event. The principal kind is matched against the
+ * origin's required-signer kind; a mismatch rejects with both the
+ * declared origin and the actual principal kind in the message so a
+ * misconfigured writer surfaces concretely at the push boundary.
+ */
+function checkCancelOriginPrincipal(blobPath, origin, principal) {
+    const required = CANCEL_ORIGIN_TO_PRINCIPAL_KIND.get(origin);
+    if (required === undefined) {
+        return {
+            ok: false,
+            reason: `event ${blobPath} CancelRequested origin ${JSON.stringify(origin)} has no principal-kind binding`,
+        };
+    }
+    if (principal.kind !== required) {
+        return {
+            ok: false,
+            reason: `event ${blobPath} CancelRequested origin ${JSON.stringify(origin)} requires principal.kind=${JSON.stringify(required)} but the push was signed by principal.kind=${JSON.stringify(principal.kind)}`,
+        };
+    }
+    return { ok: true };
+}
+/**
+ * Path-scoping for the `workflow-process` principal. A workflow-process
+ * proxies writes for the workflow-run repo's `runs/<runId>/` subtree
+ * only; the supervisor owns the `addresses/...` claim-check subtree.
+ * If the principal carries a `runId`, every prospective `runs/<X>/`
+ * subtree must use `X === principal.runId`. A workflow-process that
+ * touches the `addresses/...` subtree is rejected outright so the
+ * single-writer contract on inbox/processing/consumed holds at the
+ * substrate boundary.
+ *
+ * The check only fires for `workflow-process` principals; `hub` and
+ * `supervisor` have broader write authority by design.
+ */
+async function enforceWorkflowProcessPathScope(principal, topLevelTreePaths, listDir) {
+    if (principal.kind !== "workflow-process")
+        return { ok: true };
+    const parsed = WorkflowProcessPrincipal(principal);
+    if (parsed instanceof type.errors) {
+        // `workflowRunAuthorize` already rejects malformed
+        // `workflow-process` principals at `gateAccess`, so this branch is
+        // unreachable when the substrate is wired against the real
+        // authorize callback. Fail closed so a future wiring that supplies
+        // a permissive authorize (e.g. test substrates using `allowAll`)
+        // cannot silently bypass the path-scope enforcement below.
+        return {
+            ok: false,
+            reason: `workflow-process principal is malformed: ${parsed.summary}`,
+        };
+    }
+    if (topLevelTreePaths.includes(WORKFLOW_RUN_ADDRESSES_PREFIX)) {
+        return {
+            ok: false,
+            reason: `workflow-process principal may not write under ${WORKFLOW_RUN_ADDRESSES_PREFIX}/; the supervisor owns the claim-check subtree`,
+        };
+    }
+    if (parsed.runId !== undefined &&
+        topLevelTreePaths.includes(WORKFLOW_RUN_RUNS_PREFIX)) {
+        const runIds = await listDir(WORKFLOW_RUN_RUNS_PREFIX);
+        for (const runId of runIds) {
+            if (runId !== parsed.runId) {
+                return {
+                    ok: false,
+                    reason: `workflow-process principal scoped to runId ${JSON.stringify(parsed.runId)} may not write under ${WORKFLOW_RUN_RUNS_PREFIX}/${runId}/`,
+                };
+            }
+        }
+    }
+    return { ok: true };
+}
+/**
+ * Validate the `agent-state/` subtree shape (design §3c). The subtree
+ * holds one MUTABLE per-agent conversation snapshot directory per agent
+ * below the prefix; each entry directly under `agent-state/` must be a
+ * `<agentKey>/` DIRECTORY (not a dangling blob), and each `<agentKey>`
+ * segment must round-trip URL-encoding so a reader can recover the
+ * agent's identity from the path. The conversation blobs inside a
+ * `<agentKey>/` directory are opaque to the substrate (the warm agent's
+ * ContextStore owns their shape), so no file-level shape is enforced
+ * here.
+ *
+ * A blob written DIRECTLY at `agent-state/<name>` (with no `<agentKey>/`
+ * layer) is rejected: it would not be keyed by an agent and would not be
+ * recoverable by any reader walking the per-agent layout.
+ */
+async function validateAgentStateSubtree(topLevelTreePaths, listDir) {
+    if (!topLevelTreePaths.includes(WORKFLOW_RUN_AGENT_STATE_PREFIX)) {
+        return { ok: true };
+    }
+    const segments = await listDir(WORKFLOW_RUN_AGENT_STATE_PREFIX);
+    for (const segment of segments) {
+        const roundTrip = checkAddressSegmentRoundTrip(segment);
+        if (!roundTrip.ok) {
+            return {
+                ok: false,
+                reason: `agent-state segment ${JSON.stringify(segment)} does not round-trip URL-encoding; ${roundTrip.reason}`,
+            };
+        }
+        // Reject a blob dangling directly at `agent-state/<segment>`: every
+        // entry under the prefix must be a `<agentKey>/` directory carrying
+        // the agent's snapshot files. A directory has children under
+        // `agent-state/<segment>/`; a direct blob has none.
+        const children = await listDir(`${WORKFLOW_RUN_AGENT_STATE_PREFIX}/${segment}`);
+        if (children.length === 0) {
+            return {
+                ok: false,
+                reason: `agent-state entry ${JSON.stringify(segment)} is a blob directly under ${WORKFLOW_RUN_AGENT_STATE_PREFIX}/; entries must be a <agentKey>/ directory carrying the agent's snapshot files`,
+            };
+        }
+    }
+    return { ok: true };
+}
+export const workflowRunKindHandler = {
+    kind: "workflow-run",
+    directoryPrefix: "workflow-runs",
+    async validatePush({ repoId, ref, principal, topLevelTreePaths, readBlob, listDir, listDirOids, priorReadBlob, priorListDir, priorListDirOids, changedPathPrefixes, }) {
+        // Bound the per-run event/blob walks to the runs this commit could
+        // have touched. The substrate guarantees a prefix-preserving commit
+        // mutates only paths under `changedPathPrefixes`; every run outside
+        // them is carried forward byte-identical, so its per-run invariants
+        // (seq-contiguity, terminal-lock, append-only, blob-immutability)
+        // cannot change and were already validated when the run was last
+        // written. `scopeRunIds` is the set of run ids under a
+        // `runs/<runId>/` change prefix. It stays `undefined` -- validate
+        // every run -- whenever the substrate could not bound the change set
+        // (`changedPathPrefixes` is undefined) OR a change prefix touches the
+        // `runs/` subtree at a coarser-than-per-run granularity (a bare
+        // `runs/` prefix, which cannot identify which run changed), so the
+        // scoping never narrows below what the substrate can prove.
+        const scopeRunIds = runScopeFromChangedPrefixes(changedPathPrefixes);
+        for (const entry of topLevelTreePaths) {
+            if (entry.startsWith(`${WORKFLOW_RUN_CONTROL_PREFIX}/`) ||
+                entry === WORKFLOW_RUN_CONTROL_PREFIX) {
+                return {
+                    ok: false,
+                    reason: `top-level entry ${JSON.stringify(entry)} is under the unsupported ${WORKFLOW_RUN_CONTROL_PREFIX}/ subtree`,
+                };
+            }
+            if (!ALLOWED_TOP_LEVEL.has(entry)) {
+                return {
+                    ok: false,
+                    reason: `unexpected top-level entry ${JSON.stringify(entry)}; allowed: "${WORKFLOW_RUN_RUNS_PREFIX}", "${WORKFLOW_RUN_ADDRESSES_PREFIX}", "${WORKFLOW_RUN_AGENT_STATE_PREFIX}", "${WORKFLOW_RUN_GITIGNORE_PATH}"`,
+                };
+            }
+        }
+        const scopingCheck = await enforceWorkflowProcessPathScope(principal, topLevelTreePaths, listDir);
+        if (!scopingCheck.ok) {
+            logger.debug `workflow-run validatePush rejected ${repoId.kind}/${repoId.id} on ${ref}: ${scopingCheck.reason}`;
+            return scopingCheck;
+        }
+        const agentStateCheck = await validateAgentStateSubtree(topLevelTreePaths, listDir);
+        if (!agentStateCheck.ok) {
+            logger.debug `workflow-run validatePush rejected ${repoId.kind}/${repoId.id} on ${ref}: ${agentStateCheck.reason}`;
+            return agentStateCheck;
+        }
+        const priorTopLevels = await priorListDir("");
+        const addressesPresent = topLevelTreePaths.includes(WORKFLOW_RUN_ADDRESSES_PREFIX) ||
+            priorTopLevels.includes(WORKFLOW_RUN_ADDRESSES_PREFIX);
+        if (addressesPresent) {
+            // Enter claim-check validation when the prospective OR prior
+            // tree carries an `addresses/` subtree. A prospective tree that
+            // omits `addresses/` while the prior tree had consumed or
+            // processing entries must still go through the subtree walk so
+            // those prior entries' deletion-direction invariants fire.
+            const claimCheck = await validateClaimCheckSubtree(listDir, readBlob, priorReadBlob, priorListDir, priorListDirOids, listDirOids);
+            if (!claimCheck.ok) {
+                logger.debug `workflow-run validatePush rejected ${repoId.kind}/${repoId.id} on ${ref}: ${claimCheck.reason}`;
+                return claimCheck;
+            }
+        }
+        const runsPresent = topLevelTreePaths.includes(WORKFLOW_RUN_RUNS_PREFIX) ||
+            priorTopLevels.includes(WORKFLOW_RUN_RUNS_PREFIX);
+        if (!runsPresent) {
+            // A workflow-run repo without any `runs/` directory in either
+            // the prior or the prospective tree is a genesis state for the
+            // events subtree — `.gitignore`-only or claim-check-only trees
+            // are accepted so the asset routes' init can land before any
+            // run has produced an event.
+            return { ok: true };
+        }
+        const enumerated = await enumerateEventBlobs(listDir, scopeRunIds);
+        if (!enumerated.ok) {
+            logger.debug `workflow-run validatePush rejected ${repoId.kind}/${repoId.id} on ${ref}: ${enumerated.reason}`;
+            return { ok: false, reason: enumerated.reason };
+        }
+        const newlyTerminalRuns = [];
+        for (const [runId, entries] of enumerated.runs) {
+            if (entries.length === 0) {
+                return {
+                    ok: false,
+                    reason: `run ${runId} has an empty events directory`,
+                };
+            }
+            // Sequence contiguity: per-run events must run contiguously
+            // through the tip from whatever seq the first entry uses. Without
+            // this, a downstream consumer that iterates the log by seq would
+            // skip past a gap silently. `entries` is sorted by filenameSeq
+            // above. The first seq is not pinned to 0 because the runtime
+            // body's emptyState carries `lastSeq = 0` and emits its first
+            // event at `seq = lastSeq + 1 = 1`, while the supervisor's
+            // self-signed CancelRequested path lands seq=0 against an empty
+            // events tree.
+            const firstEntry = entries[0];
+            if (firstEntry === undefined)
+                throw new Error("unreachable");
+            const baseSeq = firstEntry.filenameSeq;
+            for (let i = 0; i < entries.length; i++) {
+                const e = entries[i];
+                if (e === undefined)
+                    throw new Error("unreachable");
+                const expectedSeq = baseSeq + i;
+                if (e.filenameSeq !== expectedSeq) {
+                    const expectedPath = `${WORKFLOW_RUN_RUNS_PREFIX}/${runId}/${WORKFLOW_RUN_EVENTS_DIR}/${String(expectedSeq)}.json`;
+                    return {
+                        ok: false,
+                        reason: `run ${runId} events have a sequence gap: ${expectedPath} is missing (next observed is ${e.blobPath})`,
+                    };
+                }
+            }
+            let terminalSeq = null;
+            let terminalType = null;
+            for (const entry of entries) {
+                const priorCheck = await checkPriorByteEquality(entry.blobPath, readBlob, priorReadBlob);
+                if (!priorCheck.ok) {
+                    logger.debug `workflow-run validatePush rejected ${repoId.kind}/${repoId.id} on ${ref}: ${priorCheck.reason}`;
+                    return priorCheck;
+                }
+                const parsed = await parseEventBlob(entry, readBlob);
+                if (!parsed.ok) {
+                    logger.debug `workflow-run validatePush rejected ${repoId.kind}/${repoId.id} on ${ref}: ${parsed.reason}`;
+                    return { ok: false, reason: parsed.reason };
+                }
+                if (parsed.parsed.body.type === "CancelRequested") {
+                    const origin = parsed.parsed.body.origin;
+                    if (typeof origin !== "string") {
+                        return {
+                            ok: false,
+                            reason: `event ${entry.blobPath} CancelRequested origin must be a string`,
+                        };
+                    }
+                    // Enforce the cancel-origin principal only for a NEWLY-ADDED blob.
+                    // A CancelRequested's origin-vs-signer rule is a write-time
+                    // authorization: it belongs to the commit that authors the event. A
+                    // later commit that merely carries the event forward -- e.g. the
+                    // run's own workflow-process cascade write of CancelPropagated /
+                    // RunCancelled, which re-lists the whole events prefix -- must not be
+                    // rejected because the carried-forward cancel was authored under a
+                    // different (supervisor) signer. Re-checking it protects nothing: the
+                    // byte-equality check above already proves a carried-forward blob is
+                    // unchanged, and the deletion-direction check proves it cannot be
+                    // dropped. A tampered (byte-diverged) blob never reaches here --
+                    // checkPriorByteEquality rejects it first. Mirrors the newly-terminal
+                    // gate below, which likewise acts only on a blob absent from the
+                    // prior tree.
+                    if ((await priorReadBlob(entry.blobPath)) === null) {
+                        const principalCheck = checkCancelOriginPrincipal(entry.blobPath, origin, principal);
+                        if (!principalCheck.ok) {
+                            logger.debug `workflow-run validatePush rejected ${repoId.kind}/${repoId.id} on ${ref}: ${principalCheck.reason}`;
+                            return principalCheck;
+                        }
+                    }
+                }
+                if (terminalSeq !== null) {
+                    return {
+                        ok: false,
+                        reason: `run ${runId} has event at seq ${String(entry.filenameSeq)} after terminal ${terminalType} at seq ${String(terminalSeq)}`,
+                    };
+                }
+                if (TERMINAL_EVENT_TYPES.has(parsed.parsed.body.type)) {
+                    terminalSeq = entry.filenameSeq;
+                    terminalType = parsed.parsed.body.type;
+                    // Surface the run as newly terminal only when this commit is
+                    // the one that ADDS the terminal event -- i.e. the terminal
+                    // blob is absent from the prior tree. A commit that carries an
+                    // already-terminal run forward unchanged (a later compaction
+                    // commit folding the per-event files into one) finds the
+                    // terminal blob already present in the prior tree and emits no
+                    // signal, so a downstream consumer keyed on the signal does
+                    // not double-fire.
+                    if ((await priorReadBlob(entry.blobPath)) === null) {
+                        const status = TERMINAL_EVENT_STATUS.get(parsed.parsed.body.type);
+                        if (status === undefined) {
+                            throw new Error(`terminal event type ${parsed.parsed.body.type} has no workflow_run.status mapping`);
+                        }
+                        const terminalBytes = await readBlob(entry.blobPath);
+                        newlyTerminalRuns.push({
+                            runId,
+                            status,
+                            terminalEventJson: new TextDecoder().decode(terminalBytes),
+                        });
+                    }
+                }
+            }
+        }
+        const combinedRuns = await validateCombinedEventRuns(listDir, readBlob, priorListDir, priorReadBlob, scopeRunIds);
+        if (!combinedRuns.ok) {
+            logger.debug `workflow-run validatePush rejected ${repoId.kind}/${repoId.id} on ${ref}: ${combinedRuns.reason}`;
+            return { ok: false, reason: combinedRuns.reason };
+        }
+        const blobsEnumerated = await enumerateRunBlobs(listDir, scopeRunIds);
+        if (!blobsEnumerated.ok) {
+            logger.debug `workflow-run validatePush rejected ${repoId.kind}/${repoId.id} on ${ref}: ${blobsEnumerated.reason}`;
+            return { ok: false, reason: blobsEnumerated.reason };
+        }
+        for (const blob of blobsEnumerated.blobs) {
+            const immutability = await checkBlobPriorByteEquality(blob.blobPath, readBlob, priorReadBlob);
+            if (!immutability.ok) {
+                logger.debug `workflow-run validatePush rejected ${repoId.kind}/${repoId.id} on ${ref}: ${immutability.reason}`;
+                return immutability;
+            }
+        }
+        // Append-only / immutability extended to the deletion direction
+        // for the runs subtree. The prospective-tree walks above only
+        // see paths PRESENT in the prospective tree; a prospective tree
+        // that omits a prior `runs/<runId>/events/<seq>.json` or
+        // `runs/<runId>/blobs/<sha>` slips past those iterations
+        // entirely. Enumerate the prior tree's runs subtree under the
+        // same shapes and reject any prior path that does not reappear.
+        const priorEnumerated = await enumerateEventBlobs(priorListDir, scopeRunIds);
+        if (!priorEnumerated.ok) {
+            return {
+                ok: false,
+                reason: `prior tree's runs subtree is structurally invalid: ${priorEnumerated.reason}`,
+            };
+        }
+        const prospectiveEventPaths = new Set();
+        for (const entries of enumerated.runs.values()) {
+            for (const e of entries)
+                prospectiveEventPaths.add(e.blobPath);
+        }
+        for (const entries of priorEnumerated.runs.values()) {
+            for (const e of entries) {
+                if (prospectiveEventPaths.has(e.blobPath))
+                    continue;
+                // A run sealed into its combined events.jsonl by this commit
+                // legitimately drops its per-event files; the fold was validated
+                // byte-for-byte against these same prior blobs above.
+                if (combinedRuns.combinedRunIds.has(e.runId))
+                    continue;
+                return {
+                    ok: false,
+                    reason: `event ${e.blobPath} present in the prior tree is missing from the prospective tree; event blobs are append-only`,
+                };
+            }
+        }
+        const priorBlobsEnumerated = await enumerateRunBlobs(priorListDir, scopeRunIds);
+        if (!priorBlobsEnumerated.ok) {
+            return {
+                ok: false,
+                reason: `prior tree's blobs subtree is structurally invalid: ${priorBlobsEnumerated.reason}`,
+            };
+        }
+        const prospectiveBlobPaths = new Set(blobsEnumerated.blobs.map((b) => b.blobPath));
+        for (const b of priorBlobsEnumerated.blobs) {
+            if (prospectiveBlobPaths.has(b.blobPath))
+                continue;
+            return {
+                ok: false,
+                reason: `blob ${b.blobPath} present in the prior tree is missing from the prospective tree; blob entries are immutable once written`,
+            };
+        }
+        return { ok: true, newlyTerminalRuns };
+    },
+    onRefUpdated() {
+        // No cached index today. Consumers read events through the
+        // substrate's subscribe / blob-read API.
+    },
+};
+export const workflowRunAuthorize = (principal, repoId, ref, action) => {
+    if (repoId.kind !== "workflow-run") {
+        return {
+            allowed: false,
+            reason: `workflow-run authorize received non-workflow-run repo ${repoId.kind}/${repoId.id}`,
+        };
+    }
+    if (principal.kind === "hub") {
+        return { allowed: true };
+    }
+    if (principal.kind === "workflow-process") {
+        const parsed = WorkflowProcessPrincipal(principal);
+        if (parsed instanceof type.errors) {
+            return {
+                allowed: false,
+                reason: `workflow-process principal is malformed: ${parsed.summary}`,
+            };
+        }
+        if (parsed.anchorRunId !== repoId.id) {
+            return {
+                allowed: false,
+                reason: `workflow-process deployment ${parsed.anchorRunId} cannot access workflow-run ${repoId.id}`,
+            };
+        }
+        switch (action) {
+            case "init":
+            case "writeTree":
+            case "receivePack":
+            case "createPack":
+            case "resolveRef":
+                return { allowed: true };
+            default: {
+                const _exhaustive = action;
+                return {
+                    allowed: false,
+                    reason: `unhandled action: ${String(_exhaustive)}`,
+                };
+            }
+        }
+    }
+    if (principal.kind === "supervisor") {
+        const parsed = SupervisorPrincipal(principal);
+        if (parsed instanceof type.errors) {
+            return {
+                allowed: false,
+                reason: `supervisor principal is malformed: ${parsed.summary}`,
+            };
+        }
+        if (parsed.anchorRunId !== repoId.id) {
+            return {
+                allowed: false,
+                reason: `supervisor deployment ${parsed.anchorRunId} cannot access workflow-run ${repoId.id}`,
+            };
+        }
+        switch (action) {
+            case "init":
+            case "writeTree":
+            case "receivePack":
+            case "createPack":
+            case "resolveRef":
+                return { allowed: true };
+            default: {
+                const _exhaustive = action;
+                return {
+                    allowed: false,
+                    reason: `unhandled action: ${String(_exhaustive)}`,
+                };
+            }
+        }
+    }
+    if (principal.kind === "sidecar") {
+        const parsed = SidecarPrincipal(principal);
+        if (parsed instanceof type.errors) {
+            return {
+                allowed: false,
+                reason: `sidecar principal is malformed: ${parsed.summary}`,
+            };
+        }
+        switch (action) {
+            case "createPack":
+            case "resolveRef":
+                return { allowed: true };
+            case "init":
+            case "writeTree":
+            case "receivePack":
+                return {
+                    allowed: false,
+                    reason: `sidecars may only read workflow-run repos, not ${action}`,
+                };
+            default: {
+                const _exhaustive = action;
+                return {
+                    allowed: false,
+                    reason: `unhandled action: ${String(_exhaustive)}`,
+                };
+            }
+        }
+    }
+    if (principal.kind === "user") {
+        // The route layer has already pre-resolved the grant verdict and
+        // attached it as `authz`. The substrate does NOT re-query the
+        // grant store here; it (a) checks the bearer-token's claims
+        // bound the requested (ref, action) and have not expired, and
+        // (b) sanity-checks that the pre-resolved verdict targets this
+        // exact resource and grant verb. Both gates must pass before the
+        // verdict's `effect` is honoured.
+        const parsed = UserPrincipal(principal);
+        if (parsed instanceof type.errors) {
+            return {
+                allowed: false,
+                reason: `user principal is malformed: ${parsed.summary}`,
+            };
+        }
+        if (!parsed.tokenClaims.actions.includes(action)) {
+            return {
+                allowed: false,
+                reason: `token does not grant action ${action}`,
+            };
+        }
+        // `ref === "*"` is the substrate's sentinel for the bulk read
+        // performed by `listRefs`. Per-ref filtering is the advertise-refs
+        // layer's responsibility, so the bulk read is gated on action and
+        // expiry alone.
+        if (ref !== "*" && !glob.match(parsed.tokenClaims.refPattern, ref)) {
+            return {
+                allowed: false,
+                reason: `token refPattern ${parsed.tokenClaims.refPattern} does not match ${ref}`,
+            };
+        }
+        if (Date.now() >= parsed.tokenClaims.expiresAt) {
+            return {
+                allowed: false,
+                reason: `token expired at ${parsed.tokenClaims.expiresAt}`,
+            };
+        }
+        const expectedResource = `workflow-run:${repoId.id}`;
+        if (parsed.authz.resource !== expectedResource) {
+            return {
+                allowed: false,
+                reason: `authz verdict resource ${parsed.authz.resource} does not match ${expectedResource}`,
+            };
+        }
+        const expectedGrantVerb = repoActionToGrantVerb(action);
+        if (parsed.authz.grantVerb !== expectedGrantVerb) {
+            return {
+                allowed: false,
+                reason: `authz verdict grantVerb ${parsed.authz.grantVerb} does not match ${expectedGrantVerb}`,
+            };
+        }
+        if (parsed.authz.effect === "allow") {
+            return { allowed: true };
+        }
+        return {
+            allowed: false,
+            reason: `authz verdict denied for ${expectedResource} ${expectedGrantVerb}`,
+        };
+    }
+    // Fail closed on any kind not handled above. The tenant-level
+    // `workflow` principal kind (`@intx/types` principalKinds) is a
+    // grant owner, not a workflow-run repo bearer, and never carries a
+    // workflow-run push here -- so it is intentionally left denied.
+    return {
+        allowed: false,
+        reason: `unknown principal kind: ${principal.kind}`,
+    };
+};
+// ---------------------------------------------------------------------
+// Claim-check API.
+//
+// Four operations layer on top of `RepoStore.writeTreeDelta` to give
+// the workflow runtime a FIFO claim-check queue per address:
+//
+//   enqueueInbox          — append a new inbox entry for an inbound
+//                           message.
+//   dequeueToProcessing   — pick the lexicographically-first inbox
+//                           entry and atomically move it to
+//                           processing.
+//   markConsumed          — atomically remove the processing entry
+//                           and write the canonical
+//                           consumed/<messageId>.json dedup index
+//                           entry.
+//   replayProcessingToInbox — recovery path that moves every
+//                           processing entry back to inbox preserving
+//                           its `<receivedAt>-<messageId>` filename
+//                           key so FIFO ordering survives a crash.
+//
+// All four route through `writeTreeDelta`, scoped to the per-address
+// subtree via `changedPathPrefixes`. The substrate serializes concurrent
+// claim-check operations on the per-repo lock and invokes each
+// operation's `computeDelta` callback with a `prior` view of the
+// committed tree. The callback reads only what it needs directly --
+// `prior.listDirOids` for a directory's names and OIDs, and
+// `prior.readBlobByOid` for a specific entry's bytes -- and returns a
+// TARGETED delta (the `puts` and `deletes` for the paths that change),
+// not the full subtree. The substrate applies that delta atomically over
+// the prior tree, carrying every untouched entry forward by OID and
+// landing the whole delta in a single commit, which is the atomic-commit
+// guarantee these operations require.
+function claimCheckCommitRef() {
+    // Every claim-check operation targets the same canonical ref used by
+    // the workflow-run kind handler's event log so subscribers see a
+    // single coherent commit stream.
+    return "refs/heads/events";
+}
+function addressSegmentFor(address) {
+    // The substrate boundary is the only place URL-encoding happens.
+    // `validatePush` rejects non-round-trip segments; mirroring the same
+    // encoder here is the only legitimate way to produce one.
+    return encodeURIComponent(address);
+}
+function addressPrefix(addressSegment) {
+    return `${WORKFLOW_RUN_ADDRESSES_PREFIX}/${addressSegment}/`;
+}
+function inboxPath(addressSegment, key) {
+    return `${WORKFLOW_RUN_ADDRESSES_PREFIX}/${addressSegment}/${WORKFLOW_RUN_INBOX_DIR}/${key}.json`;
+}
+function processingPath(addressSegment, key) {
+    return `${WORKFLOW_RUN_ADDRESSES_PREFIX}/${addressSegment}/${WORKFLOW_RUN_PROCESSING_DIR}/${key}.json`;
+}
+function consumedPath(addressSegment, messageId) {
+    return `${WORKFLOW_RUN_ADDRESSES_PREFIX}/${addressSegment}/${WORKFLOW_RUN_CONSUMED_DIR}/${messageId}.json`;
+}
+function watermarkPath(addressSegment) {
+    return `${WORKFLOW_RUN_ADDRESSES_PREFIX}/${addressSegment}/${WORKFLOW_RUN_WATERMARK_FILE}`;
+}
+function filenameKey(receivedAt, messageId) {
+    return `${String(receivedAt)}-${messageId}`;
+}
+/**
+ * Read one address's claim-check listing from the parent commit: the
+ * filenames and blob OIDs directly under
+ * `addresses/<addressSegment>/{inbox,processing,consumed}/` (NOT their
+ * bytes), plus the retention watermark. The bytes of the single entry a
+ * leg actually moves are read separately by OID via
+ * `prior.readBlobByOid`, so the unbounded consumed/ dedup index is
+ * enumerated (one `listDirOids` per bucket, names and OIDs only) but
+ * never read blob-by-blob. Every read goes through the store's
+ * cache-backed `prior` closures under the write lock. An empty listing
+ * covers the repo/ref/address-absent first-write states -- all
+ * legitimate for a brand-new operation.
+ */
+async function readAddressListing(prior, addressSegment) {
+    const listing = {
+        inbox: [],
+        processing: [],
+        consumed: [],
+        watermark: 0,
+    };
+    const addrDir = `${WORKFLOW_RUN_ADDRESSES_PREFIX}/${addressSegment}`;
+    for (const child of await prior.listDirOids(addrDir)) {
+        if (child.name === WORKFLOW_RUN_WATERMARK_FILE) {
+            const blob = await prior.readBlobByOid(child.oid);
+            listing.watermark = parseWatermark(blob, watermarkPath(addressSegment));
+            continue;
+        }
+        const bucket = child.name === WORKFLOW_RUN_INBOX_DIR
+            ? listing.inbox
+            : child.name === WORKFLOW_RUN_PROCESSING_DIR
+                ? listing.processing
+                : child.name === WORKFLOW_RUN_CONSUMED_DIR
+                    ? listing.consumed
+                    : null;
+        if (bucket === null)
+            continue;
+        for (const entry of await prior.listDirOids(`${addrDir}/${child.name}`)) {
+            bucket.push({ name: entry.name, oid: entry.oid });
+        }
+    }
+    return listing;
+}
+function utf8(s) {
+    return new TextEncoder().encode(s);
+}
+function decodeQueueEnvelopeOrThrow(bytes, blobPath) {
+    let parsed;
+    try {
+        parsed = JSON.parse(new TextDecoder().decode(bytes));
+    }
+    catch (cause) {
+        throw new Error(`claim_check_corrupt_json: ${blobPath}`, { cause });
+    }
+    const validated = ClaimCheckEnvelope(parsed);
+    if (validated instanceof type.errors) {
+        throw new Error(`claim_check_envelope_invalid: ${blobPath}: ${validated.summary}`);
+    }
+    return validated;
+}
+function decodeConsumedReceivedAtOrThrow(bytes, blobPath) {
+    let parsed;
+    try {
+        parsed = JSON.parse(new TextDecoder().decode(bytes));
+    }
+    catch (cause) {
+        throw new Error(`claim_check_corrupt_json: ${blobPath}`, { cause });
+    }
+    const validated = ConsumedEnvelope(parsed);
+    if (validated instanceof type.errors) {
+        throw new Error(`claim_check_consumed_invalid: ${blobPath}: ${validated.summary}`);
+    }
+    return validated.receivedAt;
+}
+/**
+ * Decode the per-address retention watermark from its blob bytes. The
+ * caller treats an absent watermark blob as 0 (the address has never
+ * pruned; nothing refused).
+ */
+function parseWatermark(bytes, watermarkFull) {
+    let parsed;
+    try {
+        parsed = JSON.parse(new TextDecoder().decode(bytes));
+    }
+    catch (cause) {
+        throw new Error(`claim_check_corrupt_json: ${watermarkFull}`, { cause });
+    }
+    const validated = WatermarkEnvelope(parsed);
+    if (validated instanceof type.errors) {
+        throw new Error(`claim_check_watermark_invalid: ${watermarkFull}: ${validated.summary}`);
+    }
+    return validated.watermark;
+}
+/**
+ * Internal signal thrown from the `enqueueInbox` merge callback when the
+ * messageId is already present in a queue state. Caught at the `enqueueInbox`
+ * boundary and turned into an `already-present` outcome; never escapes. It
+ * carries the specific `reason` so the boundary maps it without re-deriving.
+ */
+class InboxEntryAlreadyPresent extends Error {
+    reason;
+    constructor(reason, message) {
+        super(message);
+        this.reason = reason;
+        this.name = "InboxEntryAlreadyPresent";
+    }
+}
+/**
+ * Thrown by `enqueueInbox` when the inbound's `receivedAt` is strictly below
+ * the address's retention watermark. This is refusal under UNCERTAINTY, not
+ * proof of prior receipt: the consumed dedup entry that would rule out a
+ * duplicate may have been pruned, so the substrate can no longer tell a
+ * duplicate from a never-processed message and refuses rather than risk
+ * reprocessing. A caller gating a durable-receipt ack MUST NOT acknowledge on
+ * this -- acking an "I cannot tell" would terminally drop a message that was
+ * never written. It is its own type (not a generic Error) so that a caller,
+ * and monitoring, can surface it as a distinct loud signal rather than
+ * blending it into ordinary I/O-failure noise.
+ *
+ * Structurally unreachable on the mail-inbound path today: `enqueueInbox` is
+ * only ever called with a freshly stamped `receivedAt` (a redelivery
+ * re-stamps `Date.now()` rather than carrying the original), and the watermark
+ * only ever advances to at most `consumedAt - retentionHorizonMs <= now`, so a
+ * fresh `receivedAt` sits a full horizon above it. The sole path that carries
+ * an original (old) `receivedAt` back into the queue is
+ * `replayProcessingToInbox`, which writes straight to `inbox/` and bypasses
+ * this gate entirely. If any redelivery source is ever changed to carry the
+ * original `receivedAt` into `enqueueInbox`, this becomes reachable and its
+ * withhold-not-ack handling becomes load-bearing.
+ */
+export class StaleInboxEnqueueError extends Error {
+    constructor(message) {
+        super(message);
+        this.name = "StaleInboxEnqueueError";
+    }
+}
+/**
+ * Append a new inbox entry for `address`. The merge callback reads
+ * the address subtree under the per-repo lock, augments the inbox
+ * with the new entry, and returns the full set of address files. The
+ * substrate replaces the address subtree wholesale.
+ *
+ * Rejects if a same-messageId entry already exists in any queue
+ * state at the address — including a prior inbox entry at a
+ * different `receivedAt`. The caller is expected to consult the
+ * dedup index (consumed/) before calling, but enforcing the
+ * invariant here also catches the concurrent-enqueue race that the
+ * per-repo lock alone cannot surface.
+ */
+export async function enqueueInbox(store, principal, repoId, args) {
+    const addressSegment = addressSegmentFor(args.address);
+    const ref = claimCheckCommitRef();
+    const inboxKey = filenameKey(args.receivedAt, args.messageId);
+    const envelope = {
+        messageId: args.messageId,
+        receivedAt: args.receivedAt,
+        address: args.address,
+        mailAuditRef: args.mailAuditRef,
+        ...(args.rawMessage !== undefined ? { rawMessage: args.rawMessage } : {}),
+    };
+    const newInboxPath = inboxPath(addressSegment, inboxKey);
+    const inboxFname = `${inboxKey}.json`;
+    const consumedFname = `${args.messageId}.json`;
+    const messageIdSuffix = `-${args.messageId}.json`;
+    // The already-present cases throw `InboxEntryAlreadyPresent` from the merge
+    // callback and are caught here into an `already-present` outcome; the
+    // stale-refusal throws `StaleInboxEnqueueError`, and a substrate/I/O failure
+    // throws a generic error -- both of those propagate. The return/throw split
+    // is deliberate and load-bearing: it is the ack/withhold boundary for a
+    // caller gating a durable-receipt ack (return = safe to ack, throw =
+    // withhold), so `stale` sits with I/O on the throw side, NOT with the
+    // already-present cases (see `StaleInboxEnqueueError`).
+    let commitSha;
+    try {
+        ({ commitSha } = await store.writeTreeDelta(principal, repoId, ref, {
+            changedPathPrefixes: new Set([addressPrefix(addressSegment)]),
+            message: `enqueue inbox ${args.address} ${args.messageId}`,
+            computeDelta: async (_parentCommitSha, prior) => {
+                const listing = await readAddressListing(prior, addressSegment);
+                // Refuse a definitively-stale enqueue: a message whose receivedAt
+                // is strictly below the retention watermark could have had its
+                // consumed/ dedup entry pruned, so a duplicate can no longer be
+                // ruled out. Reject it LOUDLY rather than risk reprocessing. This
+                // is the second half of the exactly-once guarantee: above the
+                // watermark the consumed/ index is authoritative; below it, refuse.
+                if (args.receivedAt < listing.watermark) {
+                    throw new StaleInboxEnqueueError(`claim_check_stale_enqueue: address ${args.address} message ${args.messageId} receivedAt ${String(args.receivedAt)} is below the retention watermark ${String(listing.watermark)}; its dedup entry may have been pruned, so it is refused as definitively-stale`);
+                }
+                if (listing.inbox.some((e) => e.name === inboxFname)) {
+                    throw new InboxEntryAlreadyPresent("duplicate", `claim_check_duplicate_inbox: ${newInboxPath} already exists`);
+                }
+                // consumed/ is keyed by messageId alone, so this is an exact
+                // filename lookup against the dedup index.
+                if (listing.consumed.some((e) => e.name === consumedFname)) {
+                    throw new InboxEntryAlreadyPresent("consumed", `claim_check_already_consumed: address ${args.address} message ${args.messageId} is already in the consumed dedup index`);
+                }
+                if (listing.processing.some((e) => e.name.endsWith(messageIdSuffix))) {
+                    throw new InboxEntryAlreadyPresent("processing", `claim_check_already_processing: address ${args.address} message ${args.messageId} is currently in processing`);
+                }
+                // Reject a second inbox entry for the same messageId at a
+                // different receivedAt. The validatePush atomicity check also
+                // catches this on the commit path, but surfacing it here gives the
+                // caller a precise error and keeps the bad tree off the substrate.
+                const inboxDup = listing.inbox.find((e) => e.name.endsWith(messageIdSuffix));
+                if (inboxDup !== undefined) {
+                    throw new InboxEntryAlreadyPresent("already_inbox", `claim_check_already_inbox: address ${args.address} message ${args.messageId} is already in the inbox at ${inboxPath(addressSegment, inboxDup.name.slice(0, -".json".length))}`);
+                }
+                return {
+                    puts: { [newInboxPath]: utf8(JSON.stringify(envelope)) },
+                    deletes: [],
+                };
+            },
+        }));
+    }
+    catch (err) {
+        if (err instanceof InboxEntryAlreadyPresent) {
+            return { outcome: "already-present", reason: err.reason };
+        }
+        throw err;
+    }
+    return { outcome: "enqueued", commitSha, inboxKey, envelope };
+}
+/**
+ * Move the FIFO-first inbox entry for `address` to processing.
+ * Returns `null` when the inbox is empty so the caller can
+ * distinguish "nothing to do" from "operation failed".
+ *
+ * FIFO is keyed on the parsed numeric `receivedAt` prefix of the
+ * inbox filename, with a lexicographic messageId tiebreak. The
+ * substrate does NOT rely on uniform digit widths — sorting raw
+ * filenames would put `"100-…"` ahead of `"99-…"` since `'1' < '9'`,
+ * which violates the FIFO invariant.
+ */
+export async function dequeueToProcessing(store, principal, repoId, address) {
+    const addressSegment = addressSegmentFor(address);
+    const ref = claimCheckCommitRef();
+    let dequeued = null;
+    const { commitSha } = await store.writeTreeDelta(principal, repoId, ref, {
+        changedPathPrefixes: new Set([addressPrefix(addressSegment)]),
+        message: `dequeue ${address}`,
+        computeDelta: async (_parentCommitSha, prior) => {
+            const listing = await readAddressListing(prior, addressSegment);
+            const inboxDir = `${addressPrefix(addressSegment)}${WORKFLOW_RUN_INBOX_DIR}/`;
+            const candidates = [];
+            for (const entry of listing.inbox) {
+                const m = QUEUE_FILENAME_RE.exec(entry.name);
+                if (m === null || m[1] === undefined || m[2] === undefined) {
+                    throw new Error(`claim_check_invalid_inbox_filename: ${inboxDir}${entry.name}`);
+                }
+                candidates.push({
+                    entry,
+                    receivedAt: Number.parseInt(m[1], 10),
+                    messageId: m[2],
+                });
+            }
+            candidates.sort((a, b) => {
+                if (a.receivedAt !== b.receivedAt)
+                    return a.receivedAt - b.receivedAt;
+                if (a.messageId < b.messageId)
+                    return -1;
+                if (a.messageId > b.messageId)
+                    return 1;
+                return 0;
+            });
+            const first = candidates[0];
+            if (first === undefined) {
+                // Empty inbox: nothing to move. The commit is a no-op rewrite of
+                // the same tree; the caller reads `dequeued === null`.
+                dequeued = null;
+                return { puts: {}, deletes: [] };
+            }
+            const firstPath = `${inboxDir}${first.entry.name}`;
+            const key = first.entry.name.slice(0, -".json".length);
+            const bytes = await prior.readBlobByOid(first.entry.oid);
+            const envelope = decodeQueueEnvelopeOrThrow(bytes, firstPath);
+            dequeued = { key, envelope };
+            return {
+                puts: { [processingPath(addressSegment, key)]: bytes },
+                deletes: [firstPath],
+            };
+        },
+    });
+    if (dequeued === null)
+        return null;
+    const captured = dequeued;
+    return { commitSha, key: captured.key, envelope: captured.envelope };
+}
+/**
+ * Read the processing-queue entry for `messageId` at `address` without
+ * mutating the tree. Returns the decoded claim-check envelope (carrying
+ * `mailAuditRef` and, when the enqueuer inlined them, the base64
+ * `rawMessage` bytes) or `null` when no processing entry exists for the
+ * messageId.
+ *
+ * This is the read half of mailbox ownership (§3a): the supervisor's
+ * dispatch loop moves an inbox entry to processing and forwards a
+ * `trigger.fired{messageId}` to the workflow-process child; the child
+ * calls this to recover the inbound message bytes that become its step
+ * input.
+ *
+ * The read is a flat working-tree read of
+ * `addresses/<seg>/processing/`. The substrate materializes each
+ * claim-check commit's touched paths into the repo's working tree (the
+ * delta write removes each deleted path and writes each put after
+ * validation passes), so a read issued after `dequeueToProcessing`
+ * committed -- which is exactly when the supervisor forwards
+ * `trigger.fired` -- observes the processing entry. Reading the working tree (rather than walking the
+ * committed git tree) matches the workflow-process child's sibling
+ * read of `runs/<runId>/events/`. Because the
+ * read issues no commit it cannot race the supervisor's `markConsumed`
+ * write; it returns a point-in-time snapshot of the directory.
+ */
+export async function readProcessingEntry(store, _principal, repoId, address, messageId) {
+    const addressSegment = addressSegmentFor(address);
+    const repoDir = store.getRepoDir(repoId);
+    const processingDir = `${repoDir}/${WORKFLOW_RUN_ADDRESSES_PREFIX}/${addressSegment}/${WORKFLOW_RUN_PROCESSING_DIR}`;
+    const suffix = `-${messageId}.json`;
+    let filenames;
+    try {
+        filenames = await fs.promises.readdir(processingDir);
+    }
+    catch (cause) {
+        // A missing processing directory is the legitimate "no entry yet"
+        // state; any other failure surfaces.
+        if (cause instanceof Error &&
+            cause.code === "ENOENT") {
+            return null;
+        }
+        throw cause;
+    }
+    for (const filename of filenames) {
+        if (!filename.endsWith(suffix))
+            continue;
+        const blobPath = `${processingDir}/${filename}`;
+        const bytes = await fs.promises.readFile(blobPath);
+        const envelope = decodeQueueEnvelopeOrThrow(new Uint8Array(bytes), blobPath);
+        return { envelope };
+    }
+    return null;
+}
+/**
+ * Atomically remove the processing entry for `messageId` at `address`,
+ * write the canonical `consumed/<messageId>.json` dedup index entry,
+ * advance the per-address retention watermark, and prune consumed
+ * entries the watermark has passed. The caller is expected to have
+ * called `dequeueToProcessing` for this messageId; calling
+ * `markConsumed` without a matching processing entry throws.
+ *
+ * The consumed envelope preserves the original `receivedAt` and
+ * `mailAuditRef` from the processing entry so the dedup index doubles
+ * as an audit record.
+ *
+ * Retention (the bounded-`consumed/` contract): the watermark advances
+ * to `max(priorWatermark, min(consumedAt - retentionHorizonMs,
+ * thisEntry.receivedAt))` -- monotonic, and never past the entry being
+ * written so the new entry is always retained. Every consumed entry
+ * whose `receivedAt` is strictly below the new watermark is dropped
+ * (the oldest age-ordered tail). `consumed/` therefore reaches a
+ * bounded steady state of roughly one horizon's worth of entries
+ * instead of growing one entry per message forever.
+ */
+export async function markConsumed(store, principal, repoId, args) {
+    const addressSegment = addressSegmentFor(args.address);
+    const ref = claimCheckCommitRef();
+    const retentionHorizonMs = args.retentionHorizonMs ?? DEFAULT_CONSUMED_RETENTION_MS;
+    let consumedEnvelope = null;
+    let advancedWatermark = 0;
+    const prunedMessageIds = [];
+    const { commitSha } = await store.writeTreeDelta(principal, repoId, ref, {
+        changedPathPrefixes: new Set([addressPrefix(addressSegment)]),
+        message: `consume ${args.address} ${args.messageId}`,
+        computeDelta: async (_parentCommitSha, prior) => {
+            const listing = await readAddressListing(prior, addressSegment);
+            const consumedFull = consumedPath(addressSegment, args.messageId);
+            const consumedFname = `${args.messageId}.json`;
+            if (listing.consumed.some((e) => e.name === consumedFname)) {
+                throw new Error(`claim_check_already_consumed: ${consumedFull} already in the dedup index`);
+            }
+            const processingDir = `${addressPrefix(addressSegment)}${WORKFLOW_RUN_PROCESSING_DIR}/`;
+            const processingEntry = listing.processing.find((e) => e.name.endsWith(`-${args.messageId}.json`));
+            if (processingEntry === undefined) {
+                throw new Error(`claim_check_processing_not_found: address ${args.address} message ${args.messageId} has no processing entry`);
+            }
+            const processingFull = `${processingDir}${processingEntry.name}`;
+            const processingBytes = await prior.readBlobByOid(processingEntry.oid);
+            const processingEnvelope = decodeQueueEnvelopeOrThrow(processingBytes, processingFull);
+            const envelope = {
+                messageId: args.messageId,
+                receivedAt: processingEnvelope.receivedAt,
+                address: args.address,
+                runId: args.runId,
+                consumedAt: args.consumedAt,
+                mailAuditRef: processingEnvelope.mailAuditRef,
+                ...(args.rejection !== undefined ? { rejection: args.rejection } : {}),
+            };
+            consumedEnvelope = envelope;
+            // The watermark may only advance, and never past the entry this
+            // commit writes (so the new entry is always retained -- a message
+            // consumed long after receipt may legitimately sit below
+            // `consumedAt - horizon`, and it is pruned on a later commit once
+            // the watermark passes ITS receivedAt).
+            const horizonBoundary = args.consumedAt - retentionHorizonMs;
+            const newWatermark = Math.max(listing.watermark, Math.min(horizonBoundary, envelope.receivedAt));
+            advancedWatermark = newWatermark;
+            // Prune the oldest consumed tail: read each retained consumed
+            // entry's receivedAt and drop any that has fallen strictly below
+            // the new watermark. This is the one leg that must scan the
+            // consumed index — its filenames carry only the messageId, so the
+            // receivedAt lives in the bytes — and is the residual the
+            // consumed-shard lever removes. The new entry (added via puts) is
+            // never below the watermark by construction.
+            const consumedDir = `${addressPrefix(addressSegment)}${WORKFLOW_RUN_CONSUMED_DIR}/`;
+            const deletes = [processingFull];
+            for (const entry of listing.consumed) {
+                const blobPath = `${consumedDir}${entry.name}`;
+                const bytes = await prior.readBlobByOid(entry.oid);
+                const consumedReceivedAt = decodeConsumedReceivedAtOrThrow(bytes, blobPath);
+                if (consumedReceivedAt < newWatermark) {
+                    prunedMessageIds.push(entry.name.slice(0, -".json".length));
+                    deletes.push(blobPath);
+                }
+            }
+            return {
+                puts: {
+                    [consumedFull]: utf8(JSON.stringify(envelope)),
+                    [watermarkPath(addressSegment)]: utf8(JSON.stringify({ watermark: newWatermark })),
+                },
+                deletes,
+            };
+        },
+    });
+    if (consumedEnvelope === null)
+        throw new Error("unreachable");
+    const captured = consumedEnvelope;
+    return {
+        commitSha,
+        envelope: captured,
+        watermark: advancedWatermark,
+        prunedMessageIds,
+    };
+}
+/**
+ * Read the run event logs under `runs/` and return the set of
+ * `consumedMessageId`s belonging to NON-terminal runs -- the messages a
+ * live run still owns. The caller (the supervisor's spawn-time replay)
+ * feeds this into `replayProcessingToInbox`'s `ownedMessageIds` so a
+ * parked run's message is not re-admitted to inbox and dispatched a
+ * second time while the run is recovered by re-driving its durable log.
+ * Without this, the re-drive AND the re-triggered fresh run both re-park
+ * the same awaitSignal gate on the same runId, and the two concurrent
+ * runtime bodies race to a corrupt terminal.
+ *
+ * Reads the substrate's working tree via `getRepoDir`, mirroring the
+ * child's `discoverInFlightRuns`. The working tree tracks the run-event
+ * ref (`refs/heads/main`); the claim-check ref (`refs/heads/events`)
+ * cannot see it, which is why this lives at the caller rather than inside
+ * `replayProcessingToInbox`'s single-ref delta. A run whose log is sealed
+ * (combined `events.json`, only permitted for a terminated run) or
+ * carries a terminal event is excluded; an absent `runs/` directory
+ * yields an empty set.
+ */
+export async function readOwnedMessageIds(store, repoId) {
+    const fs = await import("node:fs/promises");
+    const path = await import("node:path");
+    const repoDir = store.getRepoDir(repoId);
+    const runsDir = path.join(repoDir, WORKFLOW_RUN_RUNS_PREFIX);
+    let runIds;
+    try {
+        runIds = await fs.readdir(runsDir);
+    }
+    catch (cause) {
+        if (cause instanceof Error && "code" in cause && cause.code === "ENOENT") {
+            return new Set();
+        }
+        throw cause;
+    }
+    const owned = new Set();
+    for (const runId of runIds) {
+        const runDir = path.join(runsDir, runId);
+        // A sealed run (combined events file) is terminal by the handler's
+        // own invariant -- only a terminated run is sealed -- so it owns
+        // nothing. Its presence also means the per-event directory is absent.
+        let sealed = false;
+        try {
+            await fs.access(path.join(runDir, WORKFLOW_RUN_EVENTS_FILE));
+            sealed = true;
+        }
+        catch {
+            sealed = false;
+        }
+        if (sealed)
+            continue;
+        const eventsDir = path.join(runDir, WORKFLOW_RUN_EVENTS_DIR);
+        let files;
+        try {
+            files = await fs.readdir(eventsDir);
+        }
+        catch {
+            continue;
+        }
+        let terminal = false;
+        let consumedMessageId;
+        for (const file of files) {
+            if (!file.endsWith(".json"))
+                continue;
+            let parsed;
+            try {
+                parsed = JSON.parse(await fs.readFile(path.join(eventsDir, file), "utf8"));
+            }
+            catch {
+                continue;
+            }
+            if (typeof parsed !== "object" ||
+                parsed === null ||
+                !("type" in parsed)) {
+                continue;
+            }
+            const type = parsed.type;
+            if (typeof type !== "string")
+                continue;
+            if (TERMINAL_EVENT_TYPES.has(type)) {
+                terminal = true;
+                break;
+            }
+            if (type === "RunStarted" && "consumedMessageId" in parsed) {
+                const mid = parsed.consumedMessageId;
+                if (typeof mid === "string")
+                    consumedMessageId = mid;
+            }
+        }
+        if (terminal)
+            continue;
+        if (consumedMessageId !== undefined)
+            owned.add(consumedMessageId);
+    }
+    return owned;
+}
+/** Read one run's lifecycle from a committed workflow-run tree. */
+export async function readCommittedWorkflowRunLifecycle(reads, runId) {
+    if (reads === null)
+        return "absent";
+    const runPath = `${WORKFLOW_RUN_RUNS_PREFIX}/${runId}`;
+    const runChildren = await reads.listDir(runPath);
+    if (runChildren.some((entry) => entry.type === "blob" && entry.name === WORKFLOW_RUN_EVENTS_FILE)) {
+        return "terminal";
+    }
+    const eventsPath = `${runPath}/${WORKFLOW_RUN_EVENTS_DIR}`;
+    const eventEntries = (await reads.listDir(eventsPath)).filter((entry) => entry.type === "blob" && parseEventSeq(entry.name) !== null);
+    const latest = eventEntries.reduce((candidate, entry) => {
+        if (candidate === undefined)
+            return entry;
+        const candidateSeq = parseEventSeq(candidate.name);
+        const entrySeq = parseEventSeq(entry.name);
+        return entrySeq !== null &&
+            candidateSeq !== null &&
+            entrySeq > candidateSeq
+            ? entry
+            : candidate;
+    }, undefined);
+    if (latest !== undefined) {
+        const eventPath = `${eventsPath}/${latest.name}`;
+        let parsed;
+        try {
+            parsed = JSON.parse(new TextDecoder().decode(await reads.readBlobByOid(latest.oid)));
+        }
+        catch (cause) {
+            throw new Error(`workflow_run_event_unreadable: ${eventPath}`, { cause });
+        }
+        if (typeof parsed === "object" &&
+            parsed !== null &&
+            "type" in parsed &&
+            typeof parsed.type === "string" &&
+            TERMINAL_EVENT_TYPES.has(parsed.type)) {
+            return "terminal";
+        }
+    }
+    return eventEntries.length === 0 ? "absent" : "live";
+}
+/**
+ * Read the durable lifecycle of one run from the workflow-run working tree.
+ * `grants.json` alone is still an absent run: grants are staged before the
+ * first trigger, while the first event is the durable proof that the run was
+ * fired. A sealed event log is terminal by the kind handler's compaction
+ * invariant.
+ *
+ * The supervisor uses this when in-memory cohort membership is empty. That
+ * happens both for a genuinely new deployment and briefly during recovery,
+ * so treating both states as "fire" would start a second driver for a live
+ * log or reuse a terminal run.
+ */
+export async function readWorkflowRunLifecycle(store, repoId, runId) {
+    const fs = await import("node:fs/promises");
+    const path = await import("node:path");
+    const runDir = path.join(store.getRepoDir(repoId), WORKFLOW_RUN_RUNS_PREFIX, runId);
+    try {
+        await fs.access(path.join(runDir, WORKFLOW_RUN_EVENTS_FILE));
+        return "terminal";
+    }
+    catch (cause) {
+        if (!(cause instanceof Error) ||
+            !("code" in cause) ||
+            cause.code !== "ENOENT") {
+            throw cause;
+        }
+    }
+    const eventsDir = path.join(runDir, WORKFLOW_RUN_EVENTS_DIR);
+    let files;
+    try {
+        files = await fs.readdir(eventsDir);
+    }
+    catch (cause) {
+        if (cause instanceof Error && "code" in cause && cause.code === "ENOENT") {
+            return "absent";
+        }
+        throw cause;
+    }
+    const eventFiles = files.filter((file) => parseEventSeq(file) !== null);
+    const latest = eventFiles.reduce((candidate, file) => {
+        if (candidate === undefined)
+            return file;
+        const candidateSeq = parseEventSeq(candidate);
+        const fileSeq = parseEventSeq(file);
+        return fileSeq !== null && candidateSeq !== null && fileSeq > candidateSeq
+            ? file
+            : candidate;
+    }, undefined);
+    if (latest !== undefined) {
+        const eventPath = path.join(eventsDir, latest);
+        let parsed;
+        try {
+            parsed = JSON.parse(await fs.readFile(eventPath, "utf8"));
+        }
+        catch (cause) {
+            throw new Error(`workflow_run_event_unreadable: ${eventPath}`, { cause });
+        }
+        if (typeof parsed === "object" &&
+            parsed !== null &&
+            "type" in parsed &&
+            typeof parsed.type === "string" &&
+            TERMINAL_EVENT_TYPES.has(parsed.type)) {
+            return "terminal";
+        }
+    }
+    return eventFiles.length === 0 ? "absent" : "live";
+}
+/**
+ * Recovery path: move every processing entry at `address` back to
+ * inbox preserving the original `<receivedAt>-<messageId>` filename
+ * key so FIFO ordering survives a workflow-process crash. Returns
+ * the set of keys that were moved; when nothing was in processing
+ * the returned `replayedKeys` is empty (and the commit is a no-op
+ * rewrite of the same tree).
+ *
+ * The replay is atomic across all processing entries — a partial
+ * replay that left some entries in processing would corrupt the
+ * FIFO discipline (the next dequeue would pull the wrong entry).
+ *
+ * Watermark carve-out (load-bearing — do NOT "tighten" this): the
+ * replay deliberately does NOT apply the `receivedAt < watermark`
+ * stale-reject that `enqueueInbox` applies. A `processing/` entry was
+ * already dequeued past the dedup index, so re-admitting it to
+ * `inbox/` even when its `receivedAt` has fallen below an advanced
+ * watermark is correct — the message is a legitimately in-flight one
+ * recovered after a crash, not a fresh inbound that could be a
+ * duplicate. Applying the stale-reject here would silently LOSE that
+ * message. The watermark only ever gates fresh inbound at the enqueue
+ * boundary; the recovery replay is exempt by design.
+ */
+export async function replayProcessingToInbox(store, principal, repoId, address, opts = {}) {
+    const addressSegment = addressSegmentFor(address);
+    const ref = claimCheckCommitRef();
+    const ownedMessageIds = opts.ownedMessageIds ?? new Set();
+    const replayedKeys = [];
+    const { commitSha } = await store.writeTreeDelta(principal, repoId, ref, {
+        changedPathPrefixes: new Set([addressPrefix(addressSegment)]),
+        message: `replay processing ${address}`,
+        computeDelta: async (_parentCommitSha, prior) => {
+            const listing = await readAddressListing(prior, addressSegment);
+            const processingDir = `${addressPrefix(addressSegment)}${WORKFLOW_RUN_PROCESSING_DIR}/`;
+            const inboxDir = `${addressPrefix(addressSegment)}${WORKFLOW_RUN_INBOX_DIR}/`;
+            const inboxNames = new Set(listing.inbox.map((e) => e.name));
+            const puts = {};
+            const deletes = [];
+            for (const entry of listing.processing) {
+                const bytes = await prior.readBlobByOid(entry.oid);
+                // A processing entry whose run is still live (non-terminal durable
+                // log) is owned by the recovery re-drive of that same run. Re-
+                // admitting the message to inbox would dispatch a SECOND run for
+                // it, colliding with the re-drive on the shared runId. Leave such
+                // an entry in processing untouched; the run's eventual
+                // `markConsumed` clears it. Only genuinely orphaned entries (no
+                // run, or a terminal run) are replayed. The run logs live on a
+                // different ref, so the caller precomputes the owned set.
+                const envelope = decodeQueueEnvelopeOrThrow(bytes, `${processingDir}${entry.name}`);
+                if (ownedMessageIds.has(envelope.messageId)) {
+                    continue;
+                }
+                const inboxFull = `${inboxDir}${entry.name}`;
+                if (inboxNames.has(entry.name)) {
+                    throw new Error(`claim_check_replay_collision: ${inboxFull} already exists; cannot replay processing entry`);
+                }
+                // Re-admit the in-flight entry WITHOUT the watermark stale-reject
+                // enqueueInbox applies: it was already past dedup, so a
+                // below-watermark receivedAt is no reason to refuse it. Applying
+                // the stale-check here would lose a legitimately in-flight
+                // message after a crash. Do not tighten this.
+                puts[inboxFull] = bytes;
+                deletes.push(`${processingDir}${entry.name}`);
+                replayedKeys.push(entry.name.slice(0, -".json".length));
+            }
+            return { puts, deletes };
+        },
+    });
+    return { commitSha, replayedKeys };
+}