switchroom 0.18.14 → 0.18.17

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.
Files changed (49) hide show
  1. package/dist/agent-scheduler/index.js +3 -0
  2. package/dist/auth-broker/index.js +473 -49
  3. package/dist/cli/notion-write-pretool.mjs +3 -0
  4. package/dist/cli/switchroom.js +1200 -1067
  5. package/dist/host-control/main.js +56 -51
  6. package/dist/vault/approvals/kernel-server.js +19 -12
  7. package/dist/vault/broker/server.js +675 -668
  8. package/package.json +1 -1
  9. package/profiles/_base/start.sh.hbs +81 -139
  10. package/telegram-plugin/dist/bridge/bridge.js +21 -0
  11. package/telegram-plugin/dist/gateway/gateway.js +531 -259
  12. package/telegram-plugin/dist/server.js +22 -1
  13. package/telegram-plugin/draft-stream.ts +78 -3
  14. package/telegram-plugin/gateway/bridge-dead-watchdog.ts +3 -4
  15. package/telegram-plugin/gateway/effort-command.ts +9 -7
  16. package/telegram-plugin/gateway/gateway.ts +310 -219
  17. package/telegram-plugin/gateway/litellm-local-notice-wiring.ts +200 -0
  18. package/telegram-plugin/gateway/model-command.ts +96 -18
  19. package/telegram-plugin/gateway/pending-session-command.ts +10 -8
  20. package/telegram-plugin/gateway/session-model-file.ts +38 -172
  21. package/telegram-plugin/litellm-local-notice.ts +189 -0
  22. package/telegram-plugin/model-unavailable.ts +214 -0
  23. package/telegram-plugin/quota-watch.ts +16 -4
  24. package/telegram-plugin/runtime-metrics.ts +47 -0
  25. package/telegram-plugin/send-gate-degraded.test.ts +9 -7
  26. package/telegram-plugin/send-gate.ts +34 -4
  27. package/telegram-plugin/session-tail.ts +14 -2
  28. package/telegram-plugin/stream-controller.ts +143 -20
  29. package/telegram-plugin/stream-reply-handler.ts +12 -2
  30. package/telegram-plugin/tests/bot-api.harness.ts +7 -2
  31. package/telegram-plugin/tests/draft-stream.test.ts +110 -1
  32. package/telegram-plugin/tests/effort-command.test.ts +4 -4
  33. package/telegram-plugin/tests/flood-windows-persistence.test.ts +2 -2
  34. package/telegram-plugin/tests/gateway-pending-command-wiring.test.ts +33 -19
  35. package/telegram-plugin/tests/gateway-session-model-relaunch.test.ts +47 -127
  36. package/telegram-plugin/tests/litellm-local-notice.test.ts +417 -0
  37. package/telegram-plugin/tests/model-command.test.ts +84 -1
  38. package/telegram-plugin/tests/model-unavailable.test.ts +187 -0
  39. package/telegram-plugin/tests/operator-events-session-tail.test.ts +55 -0
  40. package/telegram-plugin/tests/quota-watch.test.ts +21 -0
  41. package/telegram-plugin/tests/reaction-gate-routing.test.ts +2 -2
  42. package/telegram-plugin/tests/runtime-metrics.test.ts +24 -0
  43. package/telegram-plugin/tests/session-model-file.test.ts +7 -155
  44. package/telegram-plugin/tests/stream-controller-send-gate.test.ts +521 -0
  45. package/telegram-plugin/tests/stream-reply-handler.test.ts +44 -0
  46. package/telegram-plugin/tests/throttle-tier.test.ts +176 -0
  47. package/telegram-plugin/tests/worker-activity-feed.test.ts +207 -0
  48. package/telegram-plugin/throttle-tier.ts +98 -1
  49. package/telegram-plugin/worker-activity-feed.ts +83 -8
package/package.json CHANGED
@@ -1,7 +1,7 @@
1
1
  {
2
2
  "name": "switchroom",
3
3
  "//version": "NOT the release version — source of truth is the git tag, resolved by scripts/build.mjs:resolveVersion() (see CLAUDE.md > Standard release process). This field is stale by design and only the Layer-4 dev/non-tag fallback for build.mjs + src/cli/resolve-version.ts; do NOT bump it expecting a release to pick it up. npm-pack tarball naming needs a real version — do that as an UNCOMMITTED pack-time bump (see release step 6), never a committed one.",
4
- "version": "0.18.14",
4
+ "version": "0.18.17",
5
5
  "description": "Run Claude Code 24/7 on your Claude Pro/Max subscription over Telegram. Open-source alternative to OpenClaw and NanoClaw — no API keys.",
6
6
  "type": "module",
7
7
  "bin": {
@@ -1130,23 +1130,28 @@ x-litellm-tags: agent:$SWITCHROOM_AGENT_NAME,profile:${SWITCHROOM_AGENT_PROFILE:
1130
1130
  unset sr_ll_key sr_ll_ok sr_ll_unreachable
1131
1131
  fi
1132
1132
 
1133
- # --- Session model resolution (durable .session-model + .relaunch-model-intent) ---
1133
+ # --- Session model resolution (consume-once .session-model carrier) ---
1134
1134
  #
1135
- # Contract: reference/rfcs/session-model-stickiness.md (#3039 revision). A
1136
- # positively-confirmed `/model X` switch persists as
1137
- # `{{agentDir}}/.session-model` (one-line JSON written by the gateway) and is
1138
- # honored on EVERY boot — deploy, watchdog bounce, raw `docker restart`,
1139
- # host reboot, crash. It is cleared only by:
1135
+ # Contract: reference/rfcs/session-model-stickiness.md §0.1 (rev 4, operator
1136
+ # decision 2026-07-12 SESSION-SCOPED, superseding rev 3 keep-by-default). A
1137
+ # `/model` override lasts only for the current session:
1140
1138
  #
1141
- # - explicit user action (`/model default` deletes it live; a fresh
1142
- # (<10 min) explicit "revert" `.relaunch-model-intent` reverts at boot)
1143
- # - invalidation: corrupt/malformed file, or the configured yaml `model:`
1144
- # changed since the switch
1139
+ # - Claude switches apply LIVE in-session (claude's native picker) and write
1140
+ # NO carrier — the explicit `claude --model {{{modelQ}}}` this script execs
1141
+ # reverts them for free on the next boot.
1142
+ # - sr-* / sr→Claude switches and a queued /model persisted at shutdown need
1143
+ # a relaunch to take effect, so the gateway writes `{{agentDir}}/.session-model`
1144
+ # (one-line JSON) IMMEDIATELY before that relaunch. This block APPLIES the
1145
+ # carrier on the single boot that reads it and then DELETES it (consume-once).
1146
+ # Any SUBSEQUENT restart — deploy, `/restart`, `/new`, watchdog recovery,
1147
+ # crash, raw `docker restart`, host reboot — finds no carrier and boots the
1148
+ # configured default. That deletion is the marker distinguishing the
1149
+ # model-apply relaunch from every later restart.
1145
1150
  #
1146
- # Every clearing path writes `.session-model-alert`, which the gateway
1147
- # relays to the operator chat once at boot — an override is never dropped
1148
- # silently. Default (no intent / stale / corrupt intent) is KEEP (operator
1149
- # decision 2026-07-11, superseding the earlier revert-by-default).
1151
+ # `/model default` deletes the carrier live. Invalidation paths (corrupt file,
1152
+ # configured yaml `model:` changed) also delete + write `.session-model-alert`,
1153
+ # which the gateway relays to the operator chat once at boot. The normal apply
1154
+ # path is silent here — the gateway already acked the `/model` in chat.
1150
1155
  #
1151
1156
  # NB {{{modelQ}}} is already shell-single-quoted by the scaffold (it renders as
1152
1157
  # a quoted token, e.g. 'claude-sonnet-5'), so it is assigned BARE here — never
@@ -1154,63 +1159,42 @@ fi
1154
1159
  # the value and break `claude --model`.
1155
1160
  _EFFECTIVE_MODEL={{{modelQ}}}
1156
1161
  # Record the RESOLVED configured default (raw, unquoted) every boot, before
1157
- # override resolution. The gateway copies this into the override's
1162
+ # override resolution. The gateway copies this into the carrier's
1158
1163
  # `configuredDefaultAtWrite`, so both sides of the invalidation compare below
1159
1164
  # come from the same resolver. Overwrite, not consumed.
1160
1165
  printf '%s\n' "$_EFFECTIVE_MODEL" > "{{agentDir}}/.configured-default-model" 2>/dev/null || true
1161
1166
 
1162
- # Migration shim (one release, RFC §7): a leftover one-shot
1163
- # `.session-model-override` carrier means an OLD gateway wrote it immediately
1164
- # before this very bounce it is the newest user intent and WINS over any
1165
- # `.session-model` on disk. Convert (overwrite) + consume, and apply THIS
1166
- # boot (the carrier's presence is itself the keep signal).
1167
- _sm_keep=""
1167
+ # Rev-4 hygiene (#3184 review LOW-2): remove the files retired with the
1168
+ # keep/revert intent subsystem and the crashloop self-heal. Nothing reads them
1169
+ # anymore; without this rm they would linger in the bind-mounted state dir
1170
+ # forever after rollout. Idempotent a cheap no-op once clean.
1171
+ rm -f "{{agentDir}}/.relaunch-model-intent" "{{agentDir}}/.session-model-boot-attempts" "{{agentDir}}/.session-model-kept-notified"
1172
+
1173
+ # Migration shim (one release): a leftover one-shot `.session-model-override`
1174
+ # carrier from a pre-consume-once gateway means an OLD gateway wrote it
1175
+ # immediately before this very bounce — convert it to `.session-model` so the
1176
+ # block below applies + consumes it this boot (same one-shot semantics).
1168
1177
  if [ -f "{{agentDir}}/.session-model-override" ]; then
1169
1178
  _mig="$(tr -d '[:space:]' < "{{agentDir}}/.session-model-override" 2>/dev/null || true)"
1170
1179
  rm -f "{{agentDir}}/.session-model-override"
1171
1180
  if printf '%s' "$_mig" | grep -Eq '^[A-Za-z0-9][]A-Za-z0-9._[/-]{0,99}$'; then
1172
1181
  printf '{"model":"%s","configuredDefaultAtWrite":"%s","ts":%s}\n' "$_mig" "$_EFFECTIVE_MODEL" "$(( $(date +%s) * 1000 ))" > "{{agentDir}}/.session-model" 2>/dev/null || true
1173
- _sm_keep="1"
1174
- echo "session-model: migrated legacy one-shot carrier '$_mig' to durable .session-model (applying this boot)" >&2
1182
+ echo "session-model: migrated legacy one-shot carrier '$_mig' to .session-model (applying + consuming this boot)" >&2
1175
1183
  else
1176
1184
  echo "session-model: ignoring malformed legacy .session-model-override (failed shape gate)" >&2
1177
1185
  fi
1178
1186
  unset _mig
1179
1187
  fi
1180
1188
 
1181
- # One-shot intent: consume unconditionally, honor only a fresh "keep".
1182
- # Freshness clock is the EMBEDDED ts (ms), never file mtime — same clock the
1183
- # gateway writes with.
1184
- _sm_reason=""
1185
- _sm_revert=""
1186
- if [ -f "{{agentDir}}/.relaunch-model-intent" ]; then
1187
- _int_raw="$(cat "{{agentDir}}/.relaunch-model-intent" 2>/dev/null || true)"
1188
- rm -f "{{agentDir}}/.relaunch-model-intent"
1189
- _int="$(printf '%s' "$_int_raw" | sed -n 's/.*"intent"[[:space:]]*:[[:space:]]*"\([a-z]*\)".*/\1/p')"
1190
- _int_ts="$(printf '%s' "$_int_raw" | sed -n 's/.*"ts"[[:space:]]*:[[:space:]]*\([0-9]\{1,\}\).*/\1/p')"
1191
- _sm_reason="$(printf '%s' "$_int_raw" | sed -n 's/.*"reason"[[:space:]]*:[[:space:]]*"\([^"]*\)".*/\1/p')"
1192
- # #3039: boot default is KEEP. Only a FRESH explicit "revert" intent
1193
- # (stamped by an explicit user/gateway revert path) clears the override.
1194
- # A stale or corrupt intent counts as no intent → keep.
1195
- if [ "$_int" = "revert" ] && [ -n "$_int_ts" ] && [ $(( $(date +%s) * 1000 - _int_ts )) -lt 600000 ]; then
1196
- _sm_revert="1"
1197
- fi
1198
- unset _int_raw _int _int_ts
1199
- fi
1200
-
1201
1189
  if [ -f "{{agentDir}}/.session-model" ]; then
1202
1190
  _smf="$(cat "{{agentDir}}/.session-model" 2>/dev/null || true)"
1203
1191
  _sm_model="$(printf '%s' "$_smf" | sed -n 's/.*"model"[[:space:]]*:[[:space:]]*"\([^"]*\)".*/\1/p')"
1204
1192
  _sm_cfg="$(printf '%s' "$_smf" | sed -n 's/.*"configuredDefaultAtWrite"[[:space:]]*:[[:space:]]*"\([^"]*\)".*/\1/p')"
1205
- _sm_ts="$(printf '%s' "$_smf" | sed -n 's/.*"ts"[[:space:]]*:[[:space:]]*\([0-9]\{1,\}\).*/\1/p')"
1206
- if [ -n "$_sm_revert" ]; then
1207
- # Explicit fresh revert intent the ONLY boot path that clears a valid
1208
- # override by design (#3039). Everything else keeps.
1209
- rm -f "{{agentDir}}/.session-model" "{{agentDir}}/.session-model-kept-notified"
1210
- _sm_why="${_sm_reason:-explicit revert intent}"
1211
- echo "session-model: reverting to configured default '$_EFFECTIVE_MODEL' — $_sm_why (session override '$_sm_model' cleared)" >&2
1212
- printf 'Session model override `%s` was cleared as requested (%s) — this relaunch booted the configured default `%s`. Re-issue /model %s to switch back.\n' "$_sm_model" "$_sm_why" "$_EFFECTIVE_MODEL" "$_sm_model" > "{{agentDir}}/.session-model-alert" 2>/dev/null || true
1213
- unset _sm_why
1193
+ # CONSUME-ONCE: delete the carrier now, before any apply/exec. The carrier
1194
+ # governs exactly this one boot; a bad token can therefore crash at most one
1195
+ # boot (the next boot has no carrier and reverts), so no crashloop self-heal
1196
+ # is needed. Every branch below has already removed the file.
1197
+ rm -f "{{agentDir}}/.session-model"
1214
1198
  # Shape gate — kept BYTE-IDENTICAL with MODEL_ARG_RE in
1215
1199
  # telegram-plugin/gateway/model-command.ts. `/` is allowed for
1216
1200
  # OpenRouter-style `sr-vendor/model` ids; it is not a shell metachar inside
@@ -1219,116 +1203,69 @@ if [ -f "{{agentDir}}/.session-model" ]; then
1219
1203
  # extraction above emits one match per matching LINE, and `grep -Eq`
1220
1204
  # passes if ANY line matches — a multiline value must never reach
1221
1205
  # `claude --model` (parity with parseSessionModel's single-string check).
1222
- elif [ "$(printf '%s' "$_sm_model" | wc -c)" -eq 0 ] || [ "$(printf '%s' "$_sm_model" | wc -l)" -ne 0 ] || ! printf '%s' "$_sm_model" | grep -Eq '^[A-Za-z0-9][]A-Za-z0-9._[/-]{0,99}$'; then
1223
- # Invalid carrier (#3039): fall back to the configured default AND tell
1224
- # the operator once — never a silent stderr-only drop.
1225
- rm -f "{{agentDir}}/.session-model" "{{agentDir}}/.session-model-kept-notified"
1206
+ if [ "$(printf '%s' "$_sm_model" | wc -c)" -eq 0 ] || [ "$(printf '%s' "$_sm_model" | wc -l)" -ne 0 ] || ! printf '%s' "$_sm_model" | grep -Eq '^[A-Za-z0-9][]A-Za-z0-9._[/-]{0,99}$'; then
1207
+ # Invalid carrier: fall back to the configured default AND tell the
1208
+ # operator once — never a silent stderr-only drop.
1226
1209
  echo "session-model: ignoring malformed .session-model (failed shape gate) — using configured default '$_EFFECTIVE_MODEL'" >&2
1227
1210
  printf 'Your saved session model override could not be read (invalid or corrupt), so the agent booted on its configured default `%s`. Re-issue /model <name> if you want a different model.\n' "$_EFFECTIVE_MODEL" > "{{agentDir}}/.session-model-alert" 2>/dev/null || true
1228
1211
  elif [ "$_sm_cfg" != "$_EFFECTIVE_MODEL" ]; then
1229
- # switchroom.yaml `model:` changed since the switch the override is
1230
- # against a default that no longer exists (it may name a retired model).
1231
- # Invalidate + announce once (#3039 invalid-carrier fallback).
1232
- rm -f "{{agentDir}}/.session-model" "{{agentDir}}/.session-model-kept-notified"
1233
- echo "session-model: configured default changed ('$_sm_cfg''$_EFFECTIVE_MODEL') clearing session override '$_sm_model'" >&2
1234
- printf 'The configured default model changed (`%s` → `%s`), so your session override to `%s` was cleared — the agent booted on the new configured default. Re-issue /model %s if you still want it.\n' "$_sm_cfg" "$_EFFECTIVE_MODEL" "$_sm_model" "$_sm_model" > "{{agentDir}}/.session-model-alert" 2>/dev/null || true
1212
+ # switchroom.yaml `model:` changed between the switch and this apply-boot
1213
+ # the carrier is against a default that no longer exists. Invalidate +
1214
+ # announce once.
1215
+ echo "session-model: configured default changed ('$_sm_cfg' → '$_EFFECTIVE_MODEL') — dropping session override '$_sm_model'" >&2
1216
+ printf 'The configured default model changed (`%s``%s`), so your session override to `%s` was not applied — the agent booted on the new configured default. Re-issue /model %s if you still want it.\n' "$_sm_cfg" "$_EFFECTIVE_MODEL" "$_sm_model" "$_sm_model" > "{{agentDir}}/.session-model-alert" 2>/dev/null || true
1217
+ elif [ "${_sm_model#sr-}" != "$_sm_model" ] && [ -z "$_LITELLM_OK" ]; then
1218
+ # sr-* + LiteLLM unreachable at the apply-boot: booting the override would
1219
+ # 4xx against Anthropic, and consume-once means we can't retain it for a
1220
+ # later relaunch. Boot the configured default and tell the operator to
1221
+ # re-issue once the proxy is back.
1222
+ echo "session-model: LiteLLM proxy unreachable at boot — NOT applying sr-* override '$_sm_model'; booting configured default '$_EFFECTIVE_MODEL' (re-issue /model when the proxy is back)" >&2
1223
+ printf 'LiteLLM proxy was unreachable at boot, so the session model override `%s` was not applied — the agent booted on its configured default `%s`. Re-issue /model %s once the proxy is reachable.\n' "$_sm_model" "$_EFFECTIVE_MODEL" "$_sm_model" > "{{agentDir}}/.session-model-alert" 2>/dev/null || true
1235
1224
  else
1236
- case "$_sm_model" in
1237
- sr-*)
1238
- if [ -z "$_LITELLM_OK" ]; then
1239
- # sr-* + LiteLLM unreachable: booting the override would 4xx against
1240
- # Anthropic. Boot the configured default but RETAIN the durable file
1241
- # — it re-applies on the next keep relaunch once the proxy is back.
1242
- echo "session-model: LiteLLM proxy unreachable at boot — NOT applying sr-* override '$_sm_model' this boot; booting configured default '$_EFFECTIVE_MODEL' (override retained, re-applies next relaunch)" >&2
1243
- printf 'LiteLLM proxy was unreachable at boot, so the session model override `%s` was not applied — the agent booted on its configured default `%s`. The override is retained and will re-apply on the next switchroom-managed relaunch once LiteLLM is reachable; /model default drops it.\n' "$_sm_model" "$_EFFECTIVE_MODEL" > "{{agentDir}}/.session-model-alert" 2>/dev/null || true
1244
- else
1245
- _EFFECTIVE_MODEL="$_sm_model"
1246
- fi
1247
- ;;
1248
- *)
1249
- _EFFECTIVE_MODEL="$_sm_model"
1250
- ;;
1251
- esac
1252
- if [ "$_EFFECTIVE_MODEL" = "$_sm_model" ]; then
1253
- echo "session-model: keeping session override '$_sm_model' across this relaunch${_sm_reason:+ ($_sm_reason)}" >&2
1254
- # #3042 item 4: notify the chat ONCE per kept value — a watchdog bounce
1255
- # loop must not storm the operator with identical "kept" alerts. The
1256
- # sentinel is cleared on every path that clears the override.
1257
- if [ "$(cat "{{agentDir}}/.session-model-kept-notified" 2>/dev/null || true)" != "$_sm_model" ]; then
1258
- printf 'Session model override `%s` kept across this relaunch%s. It persists across restarts and deploys; /model default clears it.\n' "$_sm_model" "${_sm_reason:+ ($_sm_reason)}" > "{{agentDir}}/.session-model-alert" 2>/dev/null || true
1259
- printf '%s\n' "$_sm_model" > "{{agentDir}}/.session-model-kept-notified" 2>/dev/null || true
1260
- fi
1261
- fi
1225
+ # Apply the carrier for THIS session. No boot alert — the gateway already
1226
+ # acked the /model switch in chat, and the override reverts on next restart.
1227
+ _EFFECTIVE_MODEL="$_sm_model"
1228
+ echo "session-model: applying session override '$_sm_model' this boot (consume-once reverts on next restart)" >&2
1262
1229
  fi
1263
- unset _smf _sm_model _sm_cfg _sm_ts
1230
+ unset _smf _sm_model _sm_cfg
1264
1231
  fi
1265
- unset _sm_keep _sm_reason _sm_revert
1266
1232
 
1267
- # --- Override crashloop self-heal (#3042 review blocker 2b) ---
1233
+ # --- Session effort resolution (consume-once .session-effort carrier, #3186) ---
1268
1234
  #
1269
- # Under keep-by-default an invalid-but-shape-valid persisted model (a typo
1270
- # persisted at shutdown, or a confirmed model later retired) would make every
1271
- # boot exec `claude --model <bad>` and die with the gateway dead recovery
1272
- # would need a host shell. Detect the loop from inside: when an override is
1273
- # ACTIVE, stamp a boot-attempt counter just before exec; a healthy run makes
1274
- # the next boot arrive much later (stamp stale counter resets), while a
1275
- # fast crashloop re-enters here within seconds (docker backoff caps ≈1 min).
1276
- # Three consecutive fast boots with the same override → clear the carrier,
1277
- # boot the configured default, and alert the chat once.
1278
- _SM_CFG_RECORDED="$(cat "{{agentDir}}/.configured-default-model" 2>/dev/null | tr -d '[:space:]' || true)"
1279
- if [ -n "$_SM_CFG_RECORDED" ] && [ "$_EFFECTIVE_MODEL" != "$_SM_CFG_RECORDED" ] && [ -f "{{agentDir}}/.session-model" ]; then
1280
- _bl_now="$(date +%s)"
1281
- _bl_cnt=0
1282
- _bl_prev=0
1283
- if [ -f "{{agentDir}}/.session-model-boot-attempts" ]; then
1284
- read -r _bl_cnt _bl_prev < "{{agentDir}}/.session-model-boot-attempts" 2>/dev/null || true
1285
- fi
1286
- case "$_bl_cnt" in (''|*[!0-9]*) _bl_cnt=0;; esac
1287
- case "$_bl_prev" in (''|*[!0-9]*) _bl_prev=0;; esac
1288
- if [ $(( _bl_now - _bl_prev )) -lt 150 ]; then _bl_cnt=$(( _bl_cnt + 1 )); else _bl_cnt=1; fi
1289
- if [ "$_bl_cnt" -ge 3 ]; then
1290
- echo "session-model: override '$_EFFECTIVE_MODEL' appears to be crashlooping the session ($_bl_cnt fast boots) — clearing it, booting configured default '$_SM_CFG_RECORDED'" >&2
1291
- printf 'Session model override `%s` was cleared automatically: the agent failed to stay up %s boots in a row with it active (the model may be invalid or retired). Booting the configured default `%s`. Re-issue /model <name> if you want a different model.\n' "$_EFFECTIVE_MODEL" "$_bl_cnt" "$_SM_CFG_RECORDED" >> "{{agentDir}}/.session-model-alert" 2>/dev/null || true
1292
- rm -f "{{agentDir}}/.session-model" "{{agentDir}}/.session-model-boot-attempts" "{{agentDir}}/.session-model-kept-notified"
1293
- _EFFECTIVE_MODEL="$_SM_CFG_RECORDED"
1294
- else
1295
- printf '%s %s\n' "$_bl_cnt" "$_bl_now" > "{{agentDir}}/.session-model-boot-attempts" 2>/dev/null || true
1296
- fi
1297
- unset _bl_now _bl_cnt _bl_prev
1298
- else
1299
- # No active override this boot — a stale counter must not bite a future one.
1300
- rm -f "{{agentDir}}/.session-model-boot-attempts"
1301
- fi
1302
- unset _SM_CFG_RECORDED
1303
-
1304
- # --- Session effort resolution (durable .session-effort, #3039) ---
1235
+ # The `/effort` sibling of the block above, session-scoped like /model
1236
+ # (rev 4). A live `/effort <level>` applies in-session via the applyEffort
1237
+ # driver and writes NO carrier the explicit `--effort <configured>` below
1238
+ # reverts it on the next boot. The ONLY boot-applied case is a queued
1239
+ # mid-turn `/effort` the gateway persisted at shutdown: this block APPLIES
1240
+ # that carrier on the single boot that reads it and then DELETES it
1241
+ # (consume-once), so any subsequent restart reverts to the configured
1242
+ # `thinking_effort`. Invalidation (corrupt file / configured default
1243
+ # changed) appends to `.session-model-alert` so the gateway's boot relay
1244
+ # tells the operator once.
1305
1245
  #
1306
- # The `/effort` sibling of the block above. A positively-confirmed `/effort`
1307
- # apply persists `{{agentDir}}/.session-effort` (one-line JSON
1308
- # {"level","configuredDefaultAtWrite","ts"} written by the gateway). It is
1309
- # honored on every boot and cleared only by `/effort default` (live delete)
1310
- # or invalidation here (corrupt file / configured `thinking_effort:` changed)
1311
- # — each invalidation appends to `.session-model-alert` so the gateway's
1312
- # boot relay tells the operator once.
1246
+ # NB: the CONFIGURED default resolution is untouched (#3186 constraint)
1247
+ # the fleet `thinking_effort: low` pin (see #1978 /
1248
+ # src/config/thinking-effort-risk.ts) resolves exactly as before.
1313
1249
  _EFFECTIVE_EFFORT={{#if thinkingEffort}}'{{thinkingEffort}}'{{else}}''{{/if}}
1314
1250
  if [ -f "{{agentDir}}/.session-effort" ]; then
1315
1251
  _sef="$(cat "{{agentDir}}/.session-effort" 2>/dev/null || true)"
1316
1252
  _se_level="$(printf '%s' "$_sef" | sed -n 's/.*"level"[[:space:]]*:[[:space:]]*"\([^"]*\)".*/\1/p')"
1317
1253
  _se_cfg="$(printf '%s' "$_sef" | sed -n 's/.*"configuredDefaultAtWrite"[[:space:]]*:[[:space:]]*"\([^"]*\)".*/\1/p')"
1254
+ # CONSUME-ONCE: delete the carrier now, before any apply/exec — it governs
1255
+ # exactly this one boot (parity with the .session-model block above).
1256
+ rm -f "{{agentDir}}/.session-effort"
1318
1257
  # Allowlist gate — kept in sync with EFFORT_LEVELS in
1319
1258
  # telegram-plugin/gateway/effort-command.ts. The level is passed verbatim
1320
1259
  # to `claude --effort`.
1321
1260
  if [ "$(printf '%s' "$_se_level" | wc -l)" -ne 0 ] || ! printf '%s' "$_se_level" | grep -Eq '^(low|medium|high|xhigh|max)$'; then
1322
- rm -f "{{agentDir}}/.session-effort"
1323
1261
  echo "session-effort: ignoring malformed .session-effort (failed allowlist gate) — using configured default '${_EFFECTIVE_EFFORT:-<unset>}'" >&2
1324
1262
  printf 'Your saved session effort override could not be read (invalid or corrupt), so the agent booted on its configured default effort. Re-issue /effort <level> if you want a different one.\n' >> "{{agentDir}}/.session-model-alert" 2>/dev/null || true
1325
1263
  elif [ "$_se_cfg" != "$_EFFECTIVE_EFFORT" ]; then
1326
- rm -f "{{agentDir}}/.session-effort"
1327
- echo "session-effort: configured thinking_effort changed ('$_se_cfg' → '${_EFFECTIVE_EFFORT:-<unset>}') clearing session override '$_se_level'" >&2
1328
- printf 'The configured default effort changed (`%s` → `%s`), so your session effort override `%s` was cleared — the agent booted on the new configured default. Re-issue /effort %s if you still want it.\n' "${_se_cfg:-<unset>}" "${_EFFECTIVE_EFFORT:-<unset>}" "$_se_level" "$_se_level" >> "{{agentDir}}/.session-model-alert" 2>/dev/null || true
1264
+ echo "session-effort: configured thinking_effort changed ('$_se_cfg' → '${_EFFECTIVE_EFFORT:-<unset>}') — dropping session override '$_se_level'" >&2
1265
+ printf 'The configured default effort changed (`%s``%s`), so your session effort override `%s` was not applied — the agent booted on the new configured default. Re-issue /effort %s if you still want it.\n' "${_se_cfg:-<unset>}" "${_EFFECTIVE_EFFORT:-<unset>}" "$_se_level" "$_se_level" >> "{{agentDir}}/.session-model-alert" 2>/dev/null || true
1329
1266
  else
1330
1267
  _EFFECTIVE_EFFORT="$_se_level"
1331
- echo "session-effort: keeping session effort override '$_se_level' across this relaunch" >&2
1268
+ echo "session-effort: applying session effort override '$_se_level' this boot (consume-once — reverts on next restart)" >&2
1332
1269
  fi
1333
1270
  unset _sef _se_level _se_cfg
1334
1271
  fi
@@ -1337,6 +1274,11 @@ if [ -n "$_EFFECTIVE_EFFORT" ]; then
1337
1274
  else
1338
1275
  _EFFORT_ARG=""
1339
1276
  fi
1277
+ # Record the EFFECTIVE launched effort so the gateway can re-hydrate its
1278
+ # in-memory session-effort state after a queued-apply boot, keeping the
1279
+ # /effort menu highlight honest. Overwrite (not consumed) — the effort
1280
+ # sibling of `.active-session-model` below.
1281
+ printf '%s\n' "$_EFFECTIVE_EFFORT" > "{{agentDir}}/.active-session-effort" 2>/dev/null || true
1340
1282
 
1341
1283
  # sr-* passthrough→router repoint — ONE post-resolution gate covering BOTH the
1342
1284
  # /model override path AND the configured-default path (`model: sr-*` in
@@ -23061,6 +23061,26 @@ function isTransientUpstreamSignal(text) {
23061
23061
  const lower = sample.toLowerCase();
23062
23062
  return transientUpstreamSignals.some((s) => lower.includes(s));
23063
23063
  }
23064
+ var litellmProxyLocal429Signals = [
23065
+ "deployment over user-defined ratelimit",
23066
+ "model rate limit exceeded. tpm limit",
23067
+ "model rate limit exceeded. rpm limit",
23068
+ "deployment over defined rpm limit",
23069
+ "no deployments available for selected model",
23070
+ "litellm rate limit handler",
23071
+ "crossed tpm / rpm",
23072
+ "max parallel request limit reached"
23073
+ ];
23074
+ var litellmV3LimiterSignalPair = ["rate limit exceeded for ", "limit type:"];
23075
+ function isLitellmProxyLocal429(text) {
23076
+ if (typeof text !== "string" || text.length === 0)
23077
+ return false;
23078
+ const sample = text.length > 16384 ? text.slice(0, 16384) : text;
23079
+ const lower = sample.toLowerCase();
23080
+ if (litellmProxyLocal429Signals.some((s) => lower.includes(s)))
23081
+ return true;
23082
+ return litellmV3LimiterSignalPair.every((s) => lower.includes(s));
23083
+ }
23064
23084
 
23065
23085
  // tool-label-sidecar.ts
23066
23086
  import { existsSync as existsSync2, readFileSync, statSync as statSync2 } from "node:fs";
@@ -23481,6 +23501,7 @@ function detectErrorInTranscriptLine(line) {
23481
23501
  const errStr = typeof obj.error === "string" ? obj.error : "";
23482
23502
  const text = extractAssistantText(obj);
23483
23503
  const kind2 = status === 429 ? isTransientUpstreamSignal(`${text}
23504
+ ${errStr}`) || isLitellmProxyLocal429(`${text}
23484
23505
  ${errStr}`) ? "rate-limited" : "quota-exhausted" : classifyClaudeError({ type: errStr, status, message: text });
23485
23506
  return {
23486
23507
  kind: kind2,