dsh-switchman 1.11.5 → 1.11.7

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
@@ -523,8 +523,9 @@ export function ledgerFoldedTokens(messages, ledger) {
523
523
  }
524
524
 
525
525
  /** Ladder census of one ledger: the ACTIVE tier-1 and tier-2 populations
526
- * (id + summary mass estimate + approxTokens) the trigger and the ladder
527
- * ship pass read. Pure; never throws. */
526
+ * (id + summary mass estimate + approxTokens). Test/inspection helper —
527
+ * NOT read by the trigger: production uses index.js ladderRunsOf, which
528
+ * also collects tier-3 runs for T3 planning. Pure; never throws. */
528
529
  export function ladderCensusOf(ledger, countTokens = null) {
529
530
  const estimate = typeof countTokens === "function"
530
531
  ? countTokens
@@ -84,7 +84,7 @@
84
84
  * and every no-op with pending work is logged to decisions.log.
85
85
  */
86
86
 
87
- import { refOf, summaryMessageOf } from "./view.js";
87
+ import { refOf, summaryMessageOf, isFoldSummaryMessage, FOLD_SUMMARY_ID_PREFIX } from "./view.js";
88
88
 
89
89
  /** Ladder-block probe (tier ≥ 2): a distillation whose surface material is
90
90
  * the REPLACEMENT NODES of its consumed blocks, not a raw ref range. */
@@ -545,6 +545,147 @@ export function createSurfaceFoldPass(deps = {}) {
545
545
  }
546
546
  }
547
547
 
548
+ function nudgeTextOf(message) {
549
+ try {
550
+ const content = message?.content;
551
+ if (typeof content === "string") return content;
552
+ if (!Array.isArray(content)) return "";
553
+ return content.map((part) => typeof part?.text === "string" ? part.text : "").join("");
554
+ } catch {
555
+ return "";
556
+ }
557
+ }
558
+ let nudgeCounter = 0; // unique suffix per speak: colliding explicit ids would collide in ref resolution
559
+ function nudgeMessage(sessionId, text) {
560
+ const token = String(sessionId ?? "").replace(/[^A-Za-z0-9_-]/gu, "_").slice(0, 128) || "session";
561
+ return {
562
+ id: "trim_nudge_" + token + "_" + (++nudgeCounter),
563
+ role: "user",
564
+ content: [{ type: "text", text: String(text ?? "") }],
565
+ source: { kind: "user" },
566
+ };
567
+ }
568
+ /** Surface nudge availability excludes sessions with an open compaction pair. */
569
+ function nudgeSupported(sessionId) {
570
+ try {
571
+ const session = sessionOf(sessionId);
572
+ return surfaceCapable(session) && !lockOpenCached(sessionId, session);
573
+ } catch {
574
+ return false;
575
+ }
576
+ }
577
+ /** Append one advisory nudge as a suffix-only user message. Returns
578
+ * { seq, error }: seq is null when the host rejected the append, and error
579
+ * carries the rejection message so the skip row can name the real cause
580
+ * (a bare append-failed row once hid a v4 turn-discipline rejection). */
581
+ function nudgeAppend(sessionId, text) {
582
+ const session = sessionOf(sessionId);
583
+ if (!surfaceCapable(session)) return { seq: null, error: null };
584
+ try {
585
+ const event = session.append("user/message", nudgeMessage(sessionId, text), { surfaceOp: "append" });
586
+ return { seq: Number.isFinite(event?.seq) ? event.seq : null, error: null };
587
+ } catch (error) {
588
+ return { seq: null, error: String(error?.message ?? error) };
589
+ }
590
+ }
591
+ /** Prune advisory nudge nodes with point-sized compaction events. Only
592
+ * seqs still live on the surface are pruned: a fold may already have
593
+ * consumed one, and re-shadowing a vanished seq fails the v4 validator.
594
+ * Each seq gets its own prune event (start === end) so a non-contiguous
595
+ * set can never cover unlisted live turns — matching the contiguous
596
+ * range/seq pairing of the ladder and T1 passes. The metering prune is
597
+ * paired with its replacement (shadow-price protocol, mirroring the
598
+ * host's compaction-tool-result-pruner): the surfaceOp replace on the
599
+ * replacement is what actually retires the node from surface.nodes —
600
+ * a bare compaction/prune never moves the live surface, so without
601
+ * the pair the same nudge was re-found and re-pruned on EVERY boundary
602
+ * (one duplicate fold-marker event per step, 2026-11-14). */
603
+ /** Retired-nudge stub: id and text intentionally miss both nudgeScan
604
+ * signatures (id prefix "trim_nudge_", text prefix "[SWITCHMAN:TRIM]
605
+ * Advisory nudge") so a retired stub is never re-discovered. */
606
+ function retiredNudgeMessage(sessionId) {
607
+ const token = String(sessionId ?? "").replace(/[^A-Za-z0-9_-]/gu, "_").slice(0, 128) || "session";
608
+ return {
609
+ id: "trim_retired_nudge_" + token + "_" + (++nudgeCounter),
610
+ role: "user",
611
+ content: [{ type: "text", text: "[SWITCHMAN:TRIM] (advisory nudge retired — pressure is back below the fold line)" }],
612
+ source: { kind: "user" },
613
+ };
614
+ }
615
+ function nudgePrune(sessionId, seqs) {
616
+ try {
617
+ const session = sessionOf(sessionId);
618
+ if (!surfaceCapable(session) || !Array.isArray(seqs)) return false;
619
+ const nodes = session.surface.nodes;
620
+ if (!Array.isArray(nodes)) return false;
621
+ const live = [...new Set(seqs.filter((seq) => Number.isFinite(seq)))].filter((seq) => nodes.includes(seq));
622
+ if (live.length === 0) return true; // already consumed by a fold: nothing to undo
623
+ if (live.includes(nodes[0])) {
624
+ let headEvent;
625
+ try {
626
+ headEvent = typeof session.eventAt === "function" ? session.eventAt(nodes[0]) : undefined;
627
+ } catch {
628
+ headEvent = undefined;
629
+ }
630
+ const headMessage = headEvent === undefined ? null : messageOfEvent(session, headEvent);
631
+ // Fail closed: an unreadable head node or the protected system head
632
+ // must never be shadowed ("compaction cannot shadow the protected
633
+ // system head").
634
+ if (headMessage === null || headMessage === undefined || headMessage.role === "system") return false;
635
+ }
636
+ for (const seq of live) {
637
+ const event = session.append("compaction/prune", {
638
+ shadowedRange: { start: seq, end: seq },
639
+ shadowedSeqs: [seq],
640
+ shadowedTokenCount: 1,
641
+ });
642
+ if (!Number.isFinite(event?.seq)) return false;
643
+ // Shadow-price pairing: the replacement — not the metering
644
+ // prune — retires the node from the live surface.
645
+ let retired;
646
+ try {
647
+ retired = session.append("user/message", retiredNudgeMessage(sessionId), {
648
+ surfaceOp: { op: "replace", startSeq: seq, endSeq: seq },
649
+ sourceEventSeqs: [seq],
650
+ });
651
+ } catch {
652
+ return false;
653
+ }
654
+ if (!Number.isFinite(retired?.seq)) return false;
655
+ }
656
+ return true;
657
+ } catch {
658
+ return false;
659
+ }
660
+ }
661
+ /** Find visible advisory nudges from the tail, bounded to recent surface nodes. */
662
+ function nudgeScan(sessionId) {
663
+ try {
664
+ const session = sessionOf(sessionId);
665
+ if (!surfaceCapable(session) || typeof session.eventAt !== "function") return [];
666
+ const nodes = session.surface.nodes;
667
+ const first = Math.max(0, nodes.length - 50);
668
+ const found = [];
669
+ for (let index = nodes.length - 1; index >= first; index--) {
670
+ const seq = nodes[index];
671
+ let event;
672
+ try {
673
+ event = session.eventAt(seq);
674
+ } catch {
675
+ continue;
676
+ }
677
+ const message = event === undefined ? null : messageOfEvent(session, event);
678
+ if (message === null || message === undefined) continue;
679
+ const hasNudgeId = typeof message.id === "string" && message.id.startsWith("trim_nudge_");
680
+ const hasNudgeText = nudgeTextOf(message).startsWith("[SWITCHMAN:TRIM] Advisory nudge");
681
+ if (hasNudgeId || hasNudgeText) found.push(seq);
682
+ }
683
+ return found;
684
+ } catch {
685
+ return [];
686
+ }
687
+ }
688
+
548
689
  /** In-memory idempotence belt over the durable block marker: the same
549
690
  * (session, range, contentGeneration) is never replaced twice even if a
550
691
  * crash lost the ledger persist. Bounded like the other registries. */
@@ -612,6 +753,21 @@ export function createSurfaceFoldPass(deps = {}) {
612
753
  return mark !== null && typeof mark === "object" && typeof mark.reason === "string" && mark.reason !== "" ? mark : null;
613
754
  }
614
755
 
756
+ /** Terminal supersede test: EVERY captured event seq has left the live
757
+ * surface, so the fold's material is already off the wire (host
758
+ * compaction consumed the range, or an earlier ship whose marker was
759
+ * later lost). Such a capture can never validate again —
760
+ * the block terminates as surfaceSuperseded instead of living forever
761
+ * in the generation-keyed incomplete-seqs retry loop. */
762
+ function captureFullyShadowed(nodes, block) {
763
+ // An unreadable/empty surface is fail-safe: nothing is proven gone,
764
+ // so never terminalize on it (a valid block would be lost forever).
765
+ if (!Array.isArray(nodes) || nodes.length === 0) return false;
766
+ if (!Array.isArray(block?.surfaceSeqs) || block.surfaceSeqs.length === 0) return false;
767
+ const live = new Set(nodes);
768
+ return block.surfaceSeqs.every((seq) => !live.has(seq));
769
+ }
770
+
615
771
  /** One no-op row (only when the pass had pending work — a vacuous sweep
616
772
  * with nothing to ship stays silent, matching the early-return style of
617
773
  * the fold cycle's own action=none rows). Whole-pass reasons (disabled /
@@ -657,6 +813,8 @@ export function createSurfaceFoldPass(deps = {}) {
657
813
  })();
658
814
  const pending = blocks.filter((block) => {
659
815
  if (block?.surfaceReplaced !== null && block?.surfaceReplaced !== undefined) return false;
816
+ // Terminal supersede (see captureFullyShadowed): never pending again.
817
+ if (block?.surfaceSuperseded !== null && block?.surfaceSuperseded !== undefined) return false;
660
818
  const tomb = tombstoneOf(block);
661
819
  if (tomb === null) return true;
662
820
  // Re-open only when the surface content changed since the skip.
@@ -720,6 +878,21 @@ export function createSurfaceFoldPass(deps = {}) {
720
878
  }
721
879
  const waiting = consumed.find((entry) => entry.surfaceReplaced === null || entry.surfaceReplaced === undefined);
722
880
  if (waiting !== undefined) {
881
+ if (waiting?.surfaceSuperseded !== null && waiting?.surfaceSuperseded !== undefined) {
882
+ // The consumed block's material left the surface entirely: its
883
+ // replacement node can never materialize, so the distillation
884
+ // is terminal too (same proof as the T1 supersede — no
885
+ // generation-keyed retry can revive either block).
886
+ block.surfaceSuperseded = { reason: "ladder-consumed-superseded", at: Date.now() };
887
+ delete block.surfaceSkip;
888
+ try {
889
+ store.persist(ledger);
890
+ } catch {
891
+ /* best effort */
892
+ }
893
+ skip(sessionId, "superseded-ladder-consumed", via, `block=${block.id} consumed=${waiting.id}`, false);
894
+ continue;
895
+ }
723
896
  if (tombstoneOf(waiting) !== null) {
724
897
  // A consumed block is hard-skipped on this generation: its
725
898
  // replacement node can never materialize, so the ladder span
@@ -753,7 +926,7 @@ export function createSurfaceFoldPass(deps = {}) {
753
926
  const idIndexOf = new Map();
754
927
  for (let i = 0; i < nodeMessages.length; i++) {
755
928
  const message = nodeMessages[i];
756
- if (message !== null && typeof message === "object" && typeof message.id === "string" && message.id.startsWith("acp_summary_")) {
929
+ if (isFoldSummaryMessage(message)) {
757
930
  idIndexOf.set(message.id, i);
758
931
  }
759
932
  }
@@ -765,7 +938,7 @@ export function createSurfaceFoldPass(deps = {}) {
765
938
  atSeqs.push(atSeq);
766
939
  continue;
767
940
  }
768
- const relocatedIdx = idIndexOf.get(`acp_summary_${entry.id}`);
941
+ const relocatedIdx = idIndexOf.get(`${FOLD_SUMMARY_ID_PREFIX}${entry.id}`);
769
942
  if (relocatedIdx === undefined) {
770
943
  unresolved = entry;
771
944
  break;
@@ -795,7 +968,7 @@ export function createSurfaceFoldPass(deps = {}) {
795
968
  // picked up a stranger's acp_summary node is not this fold's
796
969
  // material (the length check alone cannot tell them apart).
797
970
  if (spanExact) {
798
- const expectedIds = new Set(consumed.map((entry) => `acp_summary_${entry.id}`));
971
+ const expectedIds = new Set(consumed.map((entry) => `${FOLD_SUMMARY_ID_PREFIX}${entry.id}`));
799
972
  for (let i = ladderFirstIdx; i <= ladderLastIdx; i++) {
800
973
  const message = nodeMessages[i];
801
974
  if (message === null || message === undefined || typeof message.id !== "string" || !expectedIds.has(message.id)) {
@@ -807,7 +980,7 @@ export function createSurfaceFoldPass(deps = {}) {
807
980
  if (spanExact) {
808
981
  for (let i = ladderFirstIdx; i <= ladderLastIdx; i++) {
809
982
  const message = nodeMessages[i];
810
- if (message === null || message === undefined || typeof message.id !== "string" || !message.id.startsWith("acp_summary_")) {
983
+ if (!isFoldSummaryMessage(message)) {
811
984
  spanExact = false;
812
985
  break;
813
986
  }
@@ -906,6 +1079,23 @@ export function createSurfaceFoldPass(deps = {}) {
906
1079
  const located = locateOnSurface(pairs, block);
907
1080
  if (located === null) {
908
1081
  if (Array.isArray(block.surfaceSeqs)) {
1082
+ if (captureFullyShadowed(nodes, block)) {
1083
+ // Every captured seq has left the live surface: the range was
1084
+ // already shadowed wholesale (host compaction, a prior ship
1085
+ // whose marker was later lost). The material is
1086
+ // off the wire and this capture can never validate again —
1087
+ // terminate the block instead of re-tombstoning it on every
1088
+ // generation bump (the pending filter drops it for good).
1089
+ block.surfaceSuperseded = { reason: "shadowed", at: Date.now() };
1090
+ delete block.surfaceSkip;
1091
+ try {
1092
+ store.persist(ledger);
1093
+ } catch {
1094
+ /* best effort */
1095
+ }
1096
+ skip(sessionId, "superseded-shadowed", via, `block=${block.id}`, false);
1097
+ continue;
1098
+ }
909
1099
  block.surfaceSkip = { reason: "incomplete-seqs", contentGeneration: liveGeneration, at: Date.now() };
910
1100
  skip(sessionId, "incomplete-seqs", via, `block=${block.id}`, false);
911
1101
  } else {
@@ -930,6 +1120,23 @@ export function createSurfaceFoldPass(deps = {}) {
930
1120
  && nodes[located[0]] === storedFirst
931
1121
  && nodes[located[1]] === storedLast;
932
1122
  if (!endpointsMatch) {
1123
+ if (captureFullyShadowed(nodes, block)) {
1124
+ // Every captured seq has left the live surface: the range was
1125
+ // already shadowed wholesale (host compaction, a prior ship
1126
+ // whose marker was later lost). The material is
1127
+ // off the wire and this capture can never validate again —
1128
+ // terminate the block instead of re-tombstoning it on every
1129
+ // generation bump (the pending filter drops it for good).
1130
+ block.surfaceSuperseded = { reason: "shadowed", at: Date.now() };
1131
+ delete block.surfaceSkip;
1132
+ try {
1133
+ store.persist(ledger);
1134
+ } catch {
1135
+ /* best effort */
1136
+ }
1137
+ skip(sessionId, "superseded-shadowed", via, `block=${block.id} relocated-range`, false);
1138
+ continue;
1139
+ }
933
1140
  block.surfaceSkip = { reason: "incomplete-seqs", contentGeneration: liveGeneration, at: Date.now() };
934
1141
  skip(sessionId, "incomplete-seqs", via, `block=${block.id} relocated-range`, false);
935
1142
  continue;
@@ -1127,7 +1334,11 @@ export function createSurfaceFoldPass(deps = {}) {
1127
1334
  const sessionId = sessionIdOfAgent(payload?.agent);
1128
1335
  if (sessionId === null) return;
1129
1336
  const hint = payload?.agent !== null && typeof payload?.agent === "object" ? payload.agent.session : null;
1130
- void sweep(sessionId, "turn-stopping", hint);
1337
+ void sweep(sessionId, "turn-stopping", hint).catch(() => {
1338
+ /* sweep is fire-and-forget on this boundary; a rejection must not
1339
+ surface as an unhandledRejection (the listener's own try only
1340
+ covers synchronous throws — async rejects bypass it). */
1341
+ });
1131
1342
  } catch {
1132
1343
  /* listener must never throw */
1133
1344
  }
@@ -1136,5 +1347,5 @@ export function createSurfaceFoldPass(deps = {}) {
1136
1347
  warn(`[context-trim] agent/turn-stopping hook failed: ${error?.message ?? error}`);
1137
1348
  }
1138
1349
 
1139
- return { sweep, captureFoldSeqs, enabledNow, supported };
1350
+ return { sweep, captureFoldSeqs, enabledNow, supported, nudgeSupported, nudgeAppend, nudgePrune, nudgeScan };
1140
1351
  }
@@ -32,7 +32,7 @@
32
32
  * which keeps the decision unit-testable (golden thresholds).
33
33
  */
34
34
 
35
- import { estimateTokensOf, refPositions, segmentBounds } from "./view.js";
35
+ import { estimateTokensOf, isFoldSummaryMessage, refPositions, segmentBounds } from "./view.js";
36
36
 
37
37
  /** Hard minimum wall-clock interval between side-model folds. */
38
38
  export const FOLD_MIN_INTERVAL_MS = 60_000;
@@ -50,10 +50,16 @@ export const LADDER_MIN_INTERVAL_MS = 120_000;
50
50
  * first segment boundary strictly after the frontier AND after the task
51
51
  * intent (the first user message — bili's protected zone: the session's
52
52
  * opening task statement is never folded) to the last segment end that
53
- * still clears the protected tail. Returns { startIdx, endIdx } or null
54
- * when nothing is foldable (empty history, everything folded, or
55
- * protected tail covers it all). */
56
- export function computeFoldableRange(messages, ledger, protectedTail) {
53
+ * still clears the protected tail. Prior fold summaries (acp_summary_*,
54
+ * which the surface pass relocates right after the intent segment) are
55
+ * NEVER inside the range: they stay live on the surface and consolidate
56
+ * only through ladder T2/T3 distillation. This exclusion is load-bearing —
57
+ * the frontier message is itself shadowed once its fold ships, so without
58
+ * it the fallback start slid onto the last summary and every later fold
59
+ * re-absorbed the previous summary (the rolling-absorb regression).
60
+ * Returns { startIdx, endIdx } or null when nothing is foldable (empty
61
+ * history, everything folded, or protected tail covers it all). */
62
+ export function computeFoldableRange(messages, ledger, protectedTail, options = {}) {
57
63
  if (!Array.isArray(messages) || messages.length === 0) return null;
58
64
  const tail = Number.isInteger(protectedTail) && protectedTail > 0 ? protectedTail : 0;
59
65
  const tailStart = messages.length - tail;
@@ -75,19 +81,30 @@ export function computeFoldableRange(messages, ledger, protectedTail) {
75
81
  // Task-intent protection: the first user message states the task; the
76
82
  // segment containing it is never folded (its summary would become the
77
83
  // only carrier of the task for the rest of the session — the dominant
78
- // early-drift failure this guard prevents).
79
- const firstUserIdx = messages.findIndex((message) => message?.role === "user");
80
- if (firstUserIdx >= 0) {
81
- for (const [start, end] of segmentBounds(messages)) {
82
- if (start <= firstUserIdx && firstUserIdx <= end) {
83
- startFrom = Math.max(startFrom, end + 1);
84
- break;
84
+ // early-drift failure this guard prevents). Opt-out: trimProtectTaskIntent
85
+ // =false lets the intent segment fold like any other raw content; the
86
+ // system-head clamp above is unaffected — it applies only when the head is actually a system message.
87
+ if (options?.protectTaskIntent !== false) {
88
+ const firstUserIdx = messages.findIndex((message) => message?.role === "user");
89
+ if (firstUserIdx >= 0) {
90
+ for (const [start, end] of segmentBounds(messages)) {
91
+ if (start <= firstUserIdx && firstUserIdx <= end) {
92
+ startFrom = Math.max(startFrom, end + 1);
93
+ break;
94
+ }
85
95
  }
86
96
  }
87
97
  }
88
98
  let best = null;
89
99
  for (const [start, end] of segmentBounds(messages)) {
90
100
  if (start < startFrom) continue; // behind the frontier / intent
101
+ if (isFoldSummaryMessage(messages[start])) {
102
+ // A summary is its own segment (plain user message, no tool pair).
103
+ // Ahead of the first raw segment: skip the summary run. After raw
104
+ // segments have begun: stop — never bridge a span across one.
105
+ if (best !== null) break;
106
+ continue;
107
+ }
91
108
  if (end >= tailStart) break; // segment would cut into the protected tail
92
109
  if (best === null) best = { startIdx: start, endIdx: end };
93
110
  else best.endIdx = end; // extend the contiguous span
@@ -250,18 +250,33 @@ export function renderSummaryText(block) {
250
250
  return `[SWITCHMAN:TRIM] ${block.summary}`;
251
251
  }
252
252
 
253
+ /** Wire-id prefix of every fold-summary synthetic message. The single
254
+ * definition: summaryMessageOf builds ids from it, isFoldSummaryMessage
255
+ * recognizes them, and ladder relocation re-derives the same ids. */
256
+ export const FOLD_SUMMARY_ID_PREFIX = "acp_summary_";
257
+
253
258
  /** The ONE synthetic message replacing a folded range (brief: synthetic
254
259
  * user message; id carries the acp_summary_ prefix downstream tooling
255
260
  * skips). Pure — same block, same bytes. The ladder replaces messages BY
256
261
  * this id (id-suffix = block id), which is why ids must stay stable. */
257
262
  export function summaryMessageOf(block) {
258
263
  return {
259
- id: `acp_summary_${block.id}`,
264
+ id: `${FOLD_SUMMARY_ID_PREFIX}${block.id}`,
260
265
  role: "user",
261
266
  content: [{ type: "text", text: renderSummaryText(block) }],
262
267
  };
263
268
  }
264
269
 
270
+ /** True when one wire message is a prior fold's synthetic summary
271
+ * (acp_summary_<blockId>). Recognition rides the id prefix only — never
272
+ * the source kind (the v4 validator rejects compact-checkpoint sources
273
+ * outside a compaction/start..end pair). Range selection and the advisory
274
+ * fold tool use this to KEEP summaries out of new raw folds: summaries
275
+ * stay live on the surface and consolidate only through ladder T2/T3. */
276
+ export function isFoldSummaryMessage(message) {
277
+ return typeof message?.id === "string" && message.id.startsWith(FOLD_SUMMARY_ID_PREFIX);
278
+ }
279
+
265
280
  /** Locate a block's range in a positions map at/after afterIdx. Strict:
266
281
  * both endpoint refs must resolve and the located span must equal the
267
282
  * block's stored message count (guards against re-sent histories that no
@@ -310,7 +325,7 @@ function locateBlock(positions, block, afterIdx) {
310
325
  * -> index, for every synthetic message. One index per id (a block renders
311
326
  * at most once — single consumption). */
312
327
  function summaryPositionsOf(messages) {
313
- const prefix = "acp_summary_";
328
+ const prefix = FOLD_SUMMARY_ID_PREFIX;
314
329
  const positions = new Map();
315
330
  messages.forEach((message, index) => {
316
331
  const id = message?.id;
@@ -136,8 +136,8 @@ function isRoot(ctx, agent) {
136
136
  function readThresholds(config) {
137
137
  return {
138
138
  soft: readNumber(config.wmSoftTokens, 80000),
139
- hard: readNumber(config.wmHardTokens, 95000),
140
- force: readNumber(config.wmForceTokens, 150000),
139
+ hard: readNumber(config.wmHardTokens, 100000),
140
+ force: readNumber(config.wmForceTokens, 200000),
141
141
  };
142
142
  }
143
143
 
@@ -154,8 +154,9 @@ export function measureNow(ctx, agent) {
154
154
  // longer receives — high bias), while the bare view never counts the
155
155
  // system/tool sections (low bias). max(view, wire) fixes both and puts
156
156
  // the watermark tiers (soft/hard/force + read budget) on the SAME
157
- // boundary as the fold engine — wmHardTokens 95k ≈ trim pressure
158
- // 97.5k (65% of the 150k window) — holds in units, not just in
157
+ // boundary as the fold engine — wmHardTokens 100k = trimAdvisoryPercent
158
+ // 50% × the 200k cap, and wmForceTokens 200k = the cap itself — holds
159
+ // in units, not just in
159
160
  // numbers. Strictly additive +
160
161
  // fail-open: any trim failure falls back to the stored measure.
161
162
  try {
@@ -207,7 +208,11 @@ const PAUSED_LINE =
207
208
  function renderRootWatermark(level, status) {
208
209
  if (level === "ok") return "";
209
210
  if (level === "soft") return "[SWITCHMAN:WATERMARK] soft tier: prefer delegating lightweight work (subagent/workflow) over reading more context yourself.";
210
- if (level === "hard") return "[SWITCHMAN:WATERMARK] hard tier: stop broad reading (read/glob/grep are budget-limited this turn) and start wrapping up.";
211
+ if (level === "hard")
212
+ return (
213
+ "[SWITCHMAN:WATERMARK] hard tier: stop broad reading (read/glob/grep are budget-limited this turn) and start wrapping up. " +
214
+ "If a fold or handover was committed within the last turn, this tier may reflect pre-fold pressure — the next turn carries the reduced history; do not re-verify with broad reads."
215
+ );
211
216
  const lines = ["[SWITCHMAN:WATERMARK] force tier: MANDATORY wrap-up now — produce the final answer from what you already hold."];
212
217
  const phase = status?.phase;
213
218
  if (phase === "backup") {
@@ -276,7 +281,11 @@ function subagentWindowTokens(ctx, state, sessionId) {
276
281
  async function primeSubagentWindow(ctx, state, route, key) {
277
282
  try {
278
283
  const llm = ctx.get?.("llm", false);
279
- const info = await llm?.resolveModelInfo?.({ provider: route.provider, model: route.model });
284
+ // Host contract: resolveModelInfo(provider, model) takes TWO strings. The
285
+ // object shape threw NO_ADAPTER and subagent window scaling silently fell
286
+ // back to static tiers (found 2026-11-14; same contract every other
287
+ // call site already follows).
288
+ const info = await llm?.resolveModelInfo?.(route.provider, route.model);
280
289
  const windowTokens = info?.context?.contextWindow;
281
290
  const value = Number.isFinite(windowTokens) && windowTokens > 0 ? windowTokens : null;
282
291
  state.subWindow.set(key, { value, at: Date.now() });
@@ -900,7 +909,7 @@ function applyContextWatch(ctx, config) {
900
909
  if (sessionId !== null && ACCOUNT_TOOLS.has(exec?.name ?? "")) {
901
910
  const entry = accountRead(state, sessionId, resultTokens(result));
902
911
  if (GATED_TOOLS.has(exec.name) && !state.paused.has(sessionId)) {
903
- const budget = readNumber(config.wmReadBudgetTokens, 3000);
912
+ const budget = readNumber(config.wmReadBudgetTokens, 3500);
904
913
  const ceiling = budget * CEILING_MULTIPLE;
905
914
  const used = measureNow(ctx, exec.agent);
906
915
  const th = readThresholds(config);
@@ -946,7 +955,7 @@ function applyContextWatch(ctx, config) {
946
955
  if (!GATED_TOOLS.has(exec?.name ?? "")) return next();
947
956
  const sessionId = sessionIdOf(exec?.agent);
948
957
  if (sessionId === null || state.paused.has(sessionId)) return next();
949
- const budget = readNumber(config.wmReadBudgetTokens, 3000);
958
+ const budget = readNumber(config.wmReadBudgetTokens, 3500);
950
959
  if (turnUsed(state, sessionId) < budget * CEILING_MULTIPLE) return next();
951
960
  const used = measureNow(ctx, exec.agent);
952
961
  if (used === null) return next();
package/host/intent.js CHANGED
@@ -144,9 +144,9 @@ function routeOf(value) {
144
144
  * handover off the user's raw window is respected as-is — they opted out
145
145
  * of the handover pipeline, so the 0.8 line may sit wherever they put it. */
146
146
  export function effectiveContextWindow(config) {
147
- const user = Math.max(1024, Math.round(readNumber(config.intentContextWindow, 200000)));
147
+ const user = Math.max(1024, Math.round(readNumber(config.intentContextWindow, 250000)));
148
148
  if (!readBool(config.wmAutoHandover)) return user;
149
- const floor = Math.ceil(readNumber(config.wmForceTokens, 150000) / 0.8);
149
+ const floor = Math.ceil(readNumber(config.wmForceTokens, 200000) / 0.8);
150
150
  return Math.max(user, floor);
151
151
  }
152
152
 
@@ -553,6 +553,38 @@ async function classifyOnce(ctx, config, route, system, input, turnSignal) {
553
553
  }
554
554
  }
555
555
 
556
+ /** Whether a classifier endpoint URL may receive conversation excerpts.
557
+ * The systemone channel is meant for a LOCAL/LAN decision endpoint
558
+ * (ollama); an arbitrary URL would turn the setting into an
559
+ * exfiltration channel — any settings writer could aim the per-turn
560
+ * excerpt (recent conversation text) at a public server. Allowed:
561
+ * http/https URLs whose host is loopback (localhost, 127/8, ::1) or
562
+ * private-network (RFC1918 10/8, 172.16/12, 192.168/16, link-local
563
+ * 169.254/16, fe80::/10, unique-local fc00::/7). Everything else is
564
+ * rejected before any byte leaves the process; the classifier then
565
+ * fails open to the fallback chain. Exported for tests. */
566
+ export function isAllowedClassifierEndpoint(endpoint) {
567
+ if (typeof endpoint !== "string") return false;
568
+ let url;
569
+ try {
570
+ url = new URL(endpoint.trim());
571
+ } catch {
572
+ return false;
573
+ }
574
+ if (url.protocol !== "http:" && url.protocol !== "https:") return false;
575
+ const host = url.hostname.toLowerCase().replace(/^\[|\]$/gu, "");
576
+ if (host === "localhost" || host.endsWith(".localhost") || host.endsWith(".local")) return true;
577
+ const parts = host.split(".");
578
+ if (parts.length === 4 && parts.every((p) => /^\d{1,3}$/u.test(p))) {
579
+ const octets = parts.map(Number);
580
+ if (octets.some((n) => n > 255)) return false;
581
+ const [a, b] = octets;
582
+ return a === 127 || a === 10 || (a === 172 && b >= 16 && b <= 31) || (a === 192 && b === 168) || (a === 169 && b === 254);
583
+ }
584
+ // IPv6 (brackets stripped): loopback, link-local fe80::/10, unique-local fc00::/7.
585
+ return host === "::1" || host.startsWith("fe80:") || /^[fF][cCdD][0-9a-fA-F]{2}:/.test(host);
586
+ }
587
+
556
588
  /**
557
589
  * Classify via an ollama /v1/systemone decision endpoint (direct HTTP —
558
590
  * no llm-service provider registration needed; the model name comes from
@@ -567,6 +599,8 @@ async function classifyOnce(ctx, config, route, system, input, turnSignal) {
567
599
  export async function classifySystemone(ctx, config, candidates, current, excerpt, turnSignal) {
568
600
  const endpoint = readField(config.intentClassifierEndpoint).trim();
569
601
  if (endpoint === "") throw new Error("systemone endpoint unset (intentClassifierEndpoint)");
602
+ if (!isAllowedClassifierEndpoint(endpoint))
603
+ throw new Error(`systemone endpoint refused (loopback/private URLs only): ${endpoint}`);
570
604
  const model = readField(config.intentClassifierSystemoneModel).trim();
571
605
  if (model === "") throw new Error("systemone model unset (intentClassifierSystemoneModel)");
572
606
  if (candidates.length === 0) throw new Error("no candidates");
@@ -603,6 +637,9 @@ export async function classifySystemone(ctx, config, candidates, current, excerp
603
637
  const signal = turnSignal instanceof AbortSignal ? AbortSignal.any([controller.signal, turnSignal]) : controller.signal;
604
638
  try {
605
639
  signal.throwIfAborted();
640
+ // Auditable egress record: the classifier ships a conversation excerpt
641
+ // off-process — every send is logged with the destination host.
642
+ ctx?.logger?.info?.(`dsh-switchman: intent systemone classifier POST ${new URL(endpoint).host} (excerpt ~${excerpt.length} chars)`);
606
643
  const response = await fetch(endpoint, {
607
644
  method: "POST",
608
645
  headers: { "Content-Type": "application/json" },
@@ -766,7 +803,7 @@ export async function intentStateOf(ctx, config, sessionId = null) {
766
803
  }
767
804
  }
768
805
  }
769
- const forceTokens = readNumber(config.wmForceTokens, 150000);
806
+ const forceTokens = readNumber(config.wmForceTokens, 200000);
770
807
  return {
771
808
  enabled,
772
809
  autoSelected: selection?.provider === AUTO_PROVIDER,
@@ -774,7 +811,7 @@ export async function intentStateOf(ctx, config, sessionId = null) {
774
811
  // catalog route must not mark the feature unconfigured.
775
812
  configured:
776
813
  (kind === "systemone"
777
- ? readField(config.intentClassifierEndpoint).trim() !== "" &&
814
+ ? isAllowedClassifierEndpoint(readField(config.intentClassifierEndpoint).trim()) &&
778
815
  readField(config.intentClassifierSystemoneModel).trim() !== ""
779
816
  : classifier !== null) &&
780
817
  candidates.length > 0,
package/host/lang.js CHANGED
@@ -55,10 +55,12 @@ const PROJECT_FILE = "lang.json";
55
55
  /** How long a cached settings-describe locale stays fresh (ms). */
56
56
  const LOCALE_CACHE_MS = 10_000;
57
57
 
58
- /** Normalize one language value: trim and bound the length; anything else → null. */
58
+ /** Normalize one language value: trim, collapse internal whitespace (newlines
59
+ * would break the single-line [SWITCHMAN:LANG] anchor and could smuggle extra
60
+ * protocol lines into the system prompt), bound the length; anything else → null. */
59
61
  function normalizeLangValue(value) {
60
62
  if (typeof value !== "string") return null;
61
- const trimmed = value.trim();
63
+ const trimmed = value.trim().replace(/\s+/gu, " ");
62
64
  if (trimmed.length === 0 || trimmed.length > MAX_LANG_LEN) return null;
63
65
  return trimmed;
64
66
  }