johns-harness 2026.9.26 → 2026.9.27

package/CHANGELOG.md

@@ -1,5 +1,10 @@
 # Changelog
 
+## 2026.9.27
+
+- Family 74: the local active-session pressure guard compacts on exactly one trigger, estimated active-context tokens against the active model's declared context window (`ceil(0.7 x window)`, floored at `max(compaction.reserveTokensFloor, 50000)`; undeclared windows fall back to the runtime default, `DEFAULT_CONTEXT_TOKENS`). The fixed 1 MB transcript-bytes, 220-message and 3 GiB heap backstops are removed as triggers, so a session with 221 small messages and 528 KB no longer compacts early while a large context still compacts on real token pressure. The 5-minute cooldown, the 24-message floor and the failed-attempt growth gate remain as anti-thrash throttles, and the emergency compaction path is untouched. `contextRatio`, `minIntervalMs` and `minMessagesAfterCompaction` are the config surface; legacy `transcriptBytes`/`messageCount`/`heapUsedBytes` keys are accepted and ignored.
+- Roll back with `npm i -g johns-harness@2026.9.26`.
+
 ## 2026.9.26
 
 - Family 73: the `content-story`, `dashboard` and `cold-outreach` skills are removed from the bundled set on request, leaving eight generic bundled skills (`code-ops`, `web-research`, `plan-archival`, `todos`, `arch-viz`, `gcp-logs`, `activity-logging`, `outlook`). They load by description match exactly as before, workspace copies override them, and `enabled: false` and `allowBundled` still apply. Nothing else changes from 2026.9.25.

package/README.md

@@ -85,8 +85,8 @@ node johnness.mjs --help
 ### Verify
 
 ```bash
-npm test                        # 914 tests against the installed dependency tree
-bash scripts/verify-patches.sh  # 455 checks across every patch family
+npm test                        # 915 tests against the installed dependency tree
+bash scripts/verify-patches.sh  # 459 checks across every patch family
 ```
 
 ## Launch a swarm with one command
@@ -132,7 +132,7 @@ Coding agents today run as isolated processes. Two agents on the same repository
 - **Memory built for production.** Tiered recall that traces from consolidated reflections down to the verbatim record, surviving compaction and shared across the swarm. Installed and enabled by default.
 - **Production authority per commit.** Promotion is granted to an exact SHA, never to a branch name, so one worker's authorization never covers another worker's commits.
 - **An append-only action log.** Agents record every consequential action to `ACTION_LOG.md` and read it before acting.
-- **A fixed runtime surface.** The core does not grow on a schedule. New mechanisms enter only to solve a demonstrated production failure, one narrow, tested, reversible patch at a time: 73 patch families, each with a regression test and a rollback path.
+- **A fixed runtime surface.** The core does not grow on a schedule. New mechanisms enter only to solve a demonstrated production failure, one narrow, tested, reversible patch at a time: 74 patch families, each with a regression test and a rollback path.
 - **Reliability with a definition.** Bounded retries, explicit failure states, exact-once continuation, context-pressure recovery, durable delivery, and supervised gateway recovery.
 - **Coding first, phone optional.** Direct the work at a high level from a chat channel while agents plan, implement, test, coordinate, and stage releases unattended.
 - **Local voice.** Optional native Whisper transcription of inbound audio, entirely on your machine. Off by default. [Setup](docs/NATIVE-WHISPER-MUSE.md).
@@ -199,7 +199,7 @@ The log is a [`RULES.md`](RULES.md) workspace convention enforced through the ac
 
 ## Patch families
 
-The runtime carries 73 patch families plus the 65.1 safe-update amendment. Each is a production fix applied at the source with a regression test, a durable patch marker, and a rollback path. A verifier runs 455 checks against the package tree and installed dependencies, and 914 tests run against the installed dependency tree.
+The runtime carries 74 patch families plus the 65.1 safe-update amendment. Each is a production fix applied at the source with a regression test, a durable patch marker, and a rollback path. A verifier runs 459 checks against the package tree and installed dependencies, and 915 tests run against the installed dependency tree.
 
 Every family follows the same six steps: reproduce the failure, trace the exact runtime path, make the smallest source-level change that restores the invariant, add a regression test and a patch marker, run the verifier, and retain rollback artifacts. The full index: [`docs/PATCHES.md`](docs/PATCHES.md).
 
@@ -218,7 +218,7 @@ Every family follows the same six steps: reproduce the failure, trace the exact
 ## Documentation
 
 - [`docs/index.md`](docs/index.md): runtime docs, from gateway and channels to plugins, nodes, and CLI reference.
-- [`docs/PATCHES.md`](docs/PATCHES.md): the 73 patch families, one record each.
+- [`docs/PATCHES.md`](docs/PATCHES.md): the 74 patch families, one record each.
 - [`docs/NATIVE-WHISPER-MUSE.md`](docs/NATIVE-WHISPER-MUSE.md): local Whisper transcription setup.
 - [`RULES.md`](RULES.md): the default operating rules, including Rule Zero on production authorization.
 

package/dist/compact-B247y5Qt.js

@@ -7894,9 +7894,7 @@ function resolveTranscriptPolicy(params) {
 
 //#region johnness local active-session pressure guard
 const LOCAL_PRESSURE_GUARD_DEFAULTS = {
-	transcriptBytes: 1048576,
-	messageCount: 220,
-	heapUsedBytes: 3221225472,
+	contextRatio: 0.7,
 	minIntervalMs: 300000,
 	minMessagesAfterCompaction: 24
 };
@@ -7907,14 +7905,24 @@ function localPressureGuardNumber(value, fallback) {
 function localPressureGuardSettings(cfg) {
 	const raw = cfg?.agents?.defaults?.activeSessionPressure ?? cfg?.session?.activePressure ?? {};
 	if (raw?.enabled === false) return null;
+	/* JOHNNESS_PATCH_CONTEXT_WINDOW_COMPACTION: legacy transcriptBytes/messageCount/heapUsedBytes keys are accepted without error and ignored. Config surface: enabled, contextRatio, minIntervalMs, minMessagesAfterCompaction. */
+	const contextRatio = localPressureGuardNumber(raw.contextRatio, LOCAL_PRESSURE_GUARD_DEFAULTS.contextRatio);
 	return {
-		transcriptBytes: localPressureGuardNumber(raw.transcriptBytes ?? raw.maxTranscriptBytes, LOCAL_PRESSURE_GUARD_DEFAULTS.transcriptBytes),
-		messageCount: localPressureGuardNumber(raw.messageCount ?? raw.maxMessages, LOCAL_PRESSURE_GUARD_DEFAULTS.messageCount),
-		heapUsedBytes: localPressureGuardNumber(raw.heapUsedBytes ?? raw.maxHeapUsedBytes, LOCAL_PRESSURE_GUARD_DEFAULTS.heapUsedBytes),
+		contextRatio: contextRatio > 0 ? contextRatio : LOCAL_PRESSURE_GUARD_DEFAULTS.contextRatio,
 		minIntervalMs: localPressureGuardNumber(raw.minIntervalMs, LOCAL_PRESSURE_GUARD_DEFAULTS.minIntervalMs),
 		minMessagesAfterCompaction: localPressureGuardNumber(raw.minMessagesAfterCompaction, LOCAL_PRESSURE_GUARD_DEFAULTS.minMessagesAfterCompaction)
 	};
 }
+/* JOHNNESS_PATCH_CONTEXT_WINDOW_COMPACTION: single pressure trigger — estimated active-context tokens vs the active model's declared context window; the window falls back to DEFAULT_CONTEXT_TOKENS when undeclared. See docs/PATCHES.md family 74. */
+function localPressureGuardContextWindowTokens(model) {
+	const declared = typeof model?.contextWindow === "number" && Number.isFinite(model.contextWindow) ? Math.floor(model.contextWindow) : 0;
+	return declared > 0 ? declared : DEFAULT_CONTEXT_TOKENS;
+}
+function localPressureGuardContextThreshold(settings, contextWindowTokens, cfg) {
+	const reserveTokensFloor = localPressureGuardNumber(cfg?.agents?.defaults?.compaction?.reserveTokensFloor, 2e4);
+	const floor = Math.max(reserveTokensFloor, 50000);
+	return Math.max(Math.ceil(settings.contextRatio * contextWindowTokens), floor);
+}
 /* JOHNNESS_PATCH_APPEND_ONLY_PRESSURE_BASELINE: derive a restart-safe byte baseline from the append-only JSONL so transcriptBytes pressure measures NEW growth since the last successful compaction rather than total (never-shrinking) file size. See docs/PATCHES.md family 22. Scans getEntries() newest->oldest for the newest SUCCESSFUL johnness.active-pressure-refresh marker (failed !== true) and returns its recorded post-compaction size (postCompactionBytes, else transcriptBytes). Returns 0 for legacy sessions with no such marker, so a first over-limit session still compacts once. */
 function computeActivePressureBaselineBytes(sessionManager) {
 	try {
@@ -8072,12 +8080,12 @@ function compactionRotationNumber(value, fallback) {
 	return Number.isFinite(parsed) && parsed >= 0 ? parsed : fallback;
 }
 function compactionRotationSettings(cfg) {
-	const pressure = localPressureGuardSettings(cfg) ?? { transcriptBytes: 1048576 };
 	const raw = cfg?.agents?.defaults?.activeSessionPressure ?? cfg?.session?.activePressure ?? {};
 	if (raw?.rotateAfterCompaction === false) return null;
-	const threshold = compactionRotationNumber(pressure?.transcriptBytes, 1048576);
+	/* JOHNNESS_PATCH_CONTEXT_WINDOW_COMPACTION: the legacy transcriptBytes threshold no longer exists; the rotation size budget keeps its historical 1 MiB-derived default so default rotation behavior is unchanged. */
+	const threshold = 1048576;
 	const configuredMax = compactionRotationNumber(raw.rotationMaxBytes, 0);
-	const maxBytes = configuredMax > 0 ? configuredMax : threshold > 0 ? Math.max(131072, Math.min(524288, Math.floor(threshold * .45))) : 524288;
+	const maxBytes = configuredMax > 0 ? configuredMax : Math.max(131072, Math.min(524288, Math.floor(threshold * .45)));
 	return {
 		maxBytes,
 		maxMessages: Math.max(4, Math.floor(compactionRotationNumber(raw.rotationMaxMessages, 80))),
@@ -8213,18 +8221,20 @@ async function maybeRefreshActiveSessionForPressure(params) {
 	const heapUsedBytes = typeof process?.memoryUsage === "function" ? process.memoryUsage().heapUsed : 0;
 	const __hasCompactedContext = Array.isArray(params.session.messages) && params.session.messages.some((message) => message?.role === "compactionSummary");
 	const __alreadyRotated = params.sessionManager.getEntries?.().some((entry) => entry?.type === "custom" && entry?.customType === "johnness.compaction-session-rotation");
-	if (settings.transcriptBytes > 0 && transcriptBytes >= settings.transcriptBytes && __hasCompactedContext && !__alreadyRotated) {
+	/* JOHNNESS_PATCH_CONTEXT_WINDOW_COMPACTION: the existing-compacted-lineage rotation follows the same single context-token trigger; the byte gate is gone (family 74). */
+	const __cwcWindowTokens = localPressureGuardContextWindowTokens(params.session?.model);
+	const __cwcThresholdTokens = localPressureGuardContextThreshold(settings, __cwcWindowTokens, params.config);
+	const __cwcEstimatedTokens = estimateMessagesTokens(messages);
+	if (__cwcEstimatedTokens >= __cwcThresholdTokens && __hasCompactedContext && !__alreadyRotated) {
 		const __existingRotation = await rotateCompactedSessionLineage({ config: params.config, session: params.session, sessionManager: params.sessionManager, sessionFile: params.sessionFile, sessionKey: params.sessionKey, sessionId: params.sessionId, reason: "existing-compacted-lineage" });
 		if (__existingRotation.rotated) return { refreshed: true, rotated: true, reasons: ["existing-compacted-lineage"], transcriptBytes: __existingRotation.transcriptBytes, messageCount: __existingRotation.messageCount, heapUsedBytes, sessionId: __existingRotation.sessionId, sessionFile: __existingRotation.sessionFile, oldSessionId: __existingRotation.oldSessionId, oldSessionFile: __existingRotation.oldSessionFile };
 	}
 	const reasons = [];
-	/* JOHNNESS_PATCH_APPEND_ONLY_PRESSURE_BASELINE: compare NEW growth since the last successful compaction (restart-safe baseline from the append-only JSONL), not the total never-shrinking file size (family 22). */
+	/* JOHNNESS_PATCH_CONTEXT_WINDOW_COMPACTION: single trigger — estimated active-context tokens vs the context-window threshold (family 74). Legacy byte/message/heap heuristics no longer gate; the family 22 append-only baseline below is computed and recorded for observability only. */
 	const __aopBaselineBytes = computeActivePressureBaselineBytes(params.sessionManager);
 	const __aopEffectiveBytes = effectiveTranscriptPressureBytes(transcriptBytes, __aopBaselineBytes);
-	if (settings.transcriptBytes > 0 && __aopEffectiveBytes >= settings.transcriptBytes) reasons.push("transcriptBytes=" + __aopEffectiveBytes + "/" + settings.transcriptBytes + (__aopBaselineBytes > 0 ? " (new since baseline=" + __aopBaselineBytes + "; raw=" + transcriptBytes + ")" : ""));
-	if (settings.messageCount > 0 && messageCount >= settings.messageCount) reasons.push("messages=" + messageCount + "/" + settings.messageCount);
-	if (settings.heapUsedBytes > 0 && heapUsedBytes >= settings.heapUsedBytes) reasons.push("heapUsed=" + heapUsedBytes + "/" + settings.heapUsedBytes);
-	if (reasons.length === 0) return { refreshed: false, reason: "below threshold", transcriptBytes, messageCount, heapUsedBytes };
+	if (__cwcEstimatedTokens >= __cwcThresholdTokens) reasons.push("contextTokens=" + __cwcEstimatedTokens + "/" + __cwcThresholdTokens);
+	if (reasons.length === 0) return { refreshed: false, reason: "below threshold", contextTokens: __cwcEstimatedTokens, contextThreshold: __cwcThresholdTokens, baselineBytes: __aopBaselineBytes, effectiveTranscriptBytes: __aopEffectiveBytes, transcriptBytes, messageCount, heapUsedBytes };
 	let last = 0;
 	let lastFailed = false;
 	let lastBytes = 0;
@@ -16150,13 +16160,14 @@ async function maybeCompactActiveRunForPressure(params) {
 	} catch {}
 	const heapUsedBytes = typeof process?.memoryUsage === "function" ? process.memoryUsage().heapUsed : 0;
 	const reasons = [];
-	/* JOHNNESS_PATCH_APPEND_ONLY_PRESSURE_BASELINE: compare NEW growth since the last successful compaction (restart-safe baseline from the append-only JSONL), not the total never-shrinking file size (family 22). */
+	/* JOHNNESS_PATCH_CONTEXT_WINDOW_COMPACTION: single trigger — estimated active-context tokens vs the context-window threshold (family 74). Legacy byte/message/heap heuristics no longer gate; the family 22 append-only baseline below is computed and recorded for observability only. */
 	const __aopBaselineBytes = computeActivePressureBaselineBytes(params.sessionManager);
 	const __aopEffectiveBytes = effectiveTranscriptPressureBytes(transcriptBytes, __aopBaselineBytes);
-	if (settings.transcriptBytes > 0 && __aopEffectiveBytes >= settings.transcriptBytes) reasons.push("transcriptBytes=" + __aopEffectiveBytes + "/" + settings.transcriptBytes + (__aopBaselineBytes > 0 ? " (new since baseline=" + __aopBaselineBytes + "; raw=" + transcriptBytes + ")" : ""));
-	if (settings.messageCount > 0 && messageCount >= settings.messageCount) reasons.push("messages=" + messageCount + "/" + settings.messageCount);
-	if (settings.heapUsedBytes > 0 && heapUsedBytes >= settings.heapUsedBytes) reasons.push("heapUsed=" + heapUsedBytes + "/" + settings.heapUsedBytes);
-	if (reasons.length === 0) return { refreshed: false, reason: "below threshold", transcriptBytes, messageCount, heapUsedBytes };
+	const __cwcWindowTokens = localPressureGuardContextWindowTokens(params.session?.model);
+	const __cwcThresholdTokens = localPressureGuardContextThreshold(settings, __cwcWindowTokens, params.config);
+	const __cwcEstimatedTokens = estimateMessagesTokens(messages);
+	if (__cwcEstimatedTokens >= __cwcThresholdTokens) reasons.push("contextTokens=" + __cwcEstimatedTokens + "/" + __cwcThresholdTokens);
+	if (reasons.length === 0) return { refreshed: false, reason: "below threshold", contextTokens: __cwcEstimatedTokens, contextThreshold: __cwcThresholdTokens, baselineBytes: __aopBaselineBytes, effectiveTranscriptBytes: __aopEffectiveBytes, transcriptBytes, messageCount, heapUsedBytes };
 	let last = 0;
 	let lastFailed = false;
 	let lastBytes = 0;

package/dist/pi-embedded-C6ITuRXf.js

@@ -96130,9 +96130,7 @@ function buildEmbeddedExtensionFactories(params) {
 
 //#region johnness local active-session pressure guard
 const LOCAL_PRESSURE_GUARD_DEFAULTS = {
-	transcriptBytes: 1048576,
-	messageCount: 220,
-	heapUsedBytes: 3221225472,
+	contextRatio: 0.7,
 	minIntervalMs: 300000,
 	minMessagesAfterCompaction: 24
 };
@@ -96143,14 +96141,24 @@ function localPressureGuardNumber(value, fallback) {
 function localPressureGuardSettings(cfg) {
 	const raw = cfg?.agents?.defaults?.activeSessionPressure ?? cfg?.session?.activePressure ?? {};
 	if (raw?.enabled === false) return null;
+	/* JOHNNESS_PATCH_CONTEXT_WINDOW_COMPACTION: legacy transcriptBytes/messageCount/heapUsedBytes keys are accepted without error and ignored. Config surface: enabled, contextRatio, minIntervalMs, minMessagesAfterCompaction. */
+	const contextRatio = localPressureGuardNumber(raw.contextRatio, LOCAL_PRESSURE_GUARD_DEFAULTS.contextRatio);
 	return {
-		transcriptBytes: localPressureGuardNumber(raw.transcriptBytes ?? raw.maxTranscriptBytes, LOCAL_PRESSURE_GUARD_DEFAULTS.transcriptBytes),
-		messageCount: localPressureGuardNumber(raw.messageCount ?? raw.maxMessages, LOCAL_PRESSURE_GUARD_DEFAULTS.messageCount),
-		heapUsedBytes: localPressureGuardNumber(raw.heapUsedBytes ?? raw.maxHeapUsedBytes, LOCAL_PRESSURE_GUARD_DEFAULTS.heapUsedBytes),
+		contextRatio: contextRatio > 0 ? contextRatio : LOCAL_PRESSURE_GUARD_DEFAULTS.contextRatio,
 		minIntervalMs: localPressureGuardNumber(raw.minIntervalMs, LOCAL_PRESSURE_GUARD_DEFAULTS.minIntervalMs),
 		minMessagesAfterCompaction: localPressureGuardNumber(raw.minMessagesAfterCompaction, LOCAL_PRESSURE_GUARD_DEFAULTS.minMessagesAfterCompaction)
 	};
 }
+/* JOHNNESS_PATCH_CONTEXT_WINDOW_COMPACTION: single pressure trigger — estimated active-context tokens vs the active model's declared context window; the window falls back to DEFAULT_CONTEXT_TOKENS when undeclared. See docs/PATCHES.md family 74. */
+function localPressureGuardContextWindowTokens(model) {
+	const declared = typeof model?.contextWindow === "number" && Number.isFinite(model.contextWindow) ? Math.floor(model.contextWindow) : 0;
+	return declared > 0 ? declared : DEFAULT_CONTEXT_TOKENS;
+}
+function localPressureGuardContextThreshold(settings, contextWindowTokens, cfg) {
+	const reserveTokensFloor = localPressureGuardNumber(cfg?.agents?.defaults?.compaction?.reserveTokensFloor, 2e4);
+	const floor = Math.max(reserveTokensFloor, 50000);
+	return Math.max(Math.ceil(settings.contextRatio * contextWindowTokens), floor);
+}
 /* JOHNNESS_PATCH_APPEND_ONLY_PRESSURE_BASELINE: derive a restart-safe byte baseline from the append-only JSONL so transcriptBytes pressure measures NEW growth since the last successful compaction rather than total (never-shrinking) file size. See docs/PATCHES.md family 22. Scans getEntries() newest->oldest for the newest SUCCESSFUL johnness.active-pressure-refresh marker (failed !== true) and returns its recorded post-compaction size (postCompactionBytes, else transcriptBytes). Returns 0 for legacy sessions with no such marker, so a first over-limit session still compacts once. */
 function computeActivePressureBaselineBytes(sessionManager) {
 	try {
@@ -96308,12 +96316,12 @@ function compactionRotationNumber(value, fallback) {
 	return Number.isFinite(parsed) && parsed >= 0 ? parsed : fallback;
 }
 function compactionRotationSettings(cfg) {
-	const pressure = localPressureGuardSettings(cfg) ?? { transcriptBytes: 1048576 };
 	const raw = cfg?.agents?.defaults?.activeSessionPressure ?? cfg?.session?.activePressure ?? {};
 	if (raw?.rotateAfterCompaction === false) return null;
-	const threshold = compactionRotationNumber(pressure?.transcriptBytes, 1048576);
+	/* JOHNNESS_PATCH_CONTEXT_WINDOW_COMPACTION: the legacy transcriptBytes threshold no longer exists; the rotation size budget keeps its historical 1 MiB-derived default so default rotation behavior is unchanged. */
+	const threshold = 1048576;
 	const configuredMax = compactionRotationNumber(raw.rotationMaxBytes, 0);
-	const maxBytes = configuredMax > 0 ? configuredMax : threshold > 0 ? Math.max(131072, Math.min(524288, Math.floor(threshold * .45))) : 524288;
+	const maxBytes = configuredMax > 0 ? configuredMax : Math.max(131072, Math.min(524288, Math.floor(threshold * .45)));
 	return {
 		maxBytes,
 		maxMessages: Math.max(4, Math.floor(compactionRotationNumber(raw.rotationMaxMessages, 80))),
@@ -96449,18 +96457,20 @@ async function maybeRefreshActiveSessionForPressure(params) {
 	const heapUsedBytes = typeof process?.memoryUsage === "function" ? process.memoryUsage().heapUsed : 0;
 	const __hasCompactedContext = Array.isArray(params.session.messages) && params.session.messages.some((message) => message?.role === "compactionSummary");
 	const __alreadyRotated = params.sessionManager.getEntries?.().some((entry) => entry?.type === "custom" && entry?.customType === "johnness.compaction-session-rotation");
-	if (settings.transcriptBytes > 0 && transcriptBytes >= settings.transcriptBytes && __hasCompactedContext && !__alreadyRotated) {
+	/* JOHNNESS_PATCH_CONTEXT_WINDOW_COMPACTION: the existing-compacted-lineage rotation follows the same single context-token trigger; the byte gate is gone (family 74). */
+	const __cwcWindowTokens = localPressureGuardContextWindowTokens(params.session?.model);
+	const __cwcThresholdTokens = localPressureGuardContextThreshold(settings, __cwcWindowTokens, params.config);
+	const __cwcEstimatedTokens = estimateMessagesTokens(messages);
+	if (__cwcEstimatedTokens >= __cwcThresholdTokens && __hasCompactedContext && !__alreadyRotated) {
 		const __existingRotation = await rotateCompactedSessionLineage({ config: params.config, session: params.session, sessionManager: params.sessionManager, sessionFile: params.sessionFile, sessionKey: params.sessionKey, sessionId: params.sessionId, reason: "existing-compacted-lineage" });
 		if (__existingRotation.rotated) return { refreshed: true, rotated: true, reasons: ["existing-compacted-lineage"], transcriptBytes: __existingRotation.transcriptBytes, messageCount: __existingRotation.messageCount, heapUsedBytes, sessionId: __existingRotation.sessionId, sessionFile: __existingRotation.sessionFile, oldSessionId: __existingRotation.oldSessionId, oldSessionFile: __existingRotation.oldSessionFile };
 	}
 	const reasons = [];
-	/* JOHNNESS_PATCH_APPEND_ONLY_PRESSURE_BASELINE: compare NEW growth since the last successful compaction (restart-safe baseline from the append-only JSONL), not the total never-shrinking file size (family 22). */
+	/* JOHNNESS_PATCH_CONTEXT_WINDOW_COMPACTION: single trigger — estimated active-context tokens vs the context-window threshold (family 74). Legacy byte/message/heap heuristics no longer gate; the family 22 append-only baseline below is computed and recorded for observability only. */
 	const __aopBaselineBytes = computeActivePressureBaselineBytes(params.sessionManager);
 	const __aopEffectiveBytes = effectiveTranscriptPressureBytes(transcriptBytes, __aopBaselineBytes);
-	if (settings.transcriptBytes > 0 && __aopEffectiveBytes >= settings.transcriptBytes) reasons.push("transcriptBytes=" + __aopEffectiveBytes + "/" + settings.transcriptBytes + (__aopBaselineBytes > 0 ? " (new since baseline=" + __aopBaselineBytes + "; raw=" + transcriptBytes + ")" : ""));
-	if (settings.messageCount > 0 && messageCount >= settings.messageCount) reasons.push("messages=" + messageCount + "/" + settings.messageCount);
-	if (settings.heapUsedBytes > 0 && heapUsedBytes >= settings.heapUsedBytes) reasons.push("heapUsed=" + heapUsedBytes + "/" + settings.heapUsedBytes);
-	if (reasons.length === 0) return { refreshed: false, reason: "below threshold", transcriptBytes, messageCount, heapUsedBytes };
+	if (__cwcEstimatedTokens >= __cwcThresholdTokens) reasons.push("contextTokens=" + __cwcEstimatedTokens + "/" + __cwcThresholdTokens);
+	if (reasons.length === 0) return { refreshed: false, reason: "below threshold", contextTokens: __cwcEstimatedTokens, contextThreshold: __cwcThresholdTokens, baselineBytes: __aopBaselineBytes, effectiveTranscriptBytes: __aopEffectiveBytes, transcriptBytes, messageCount, heapUsedBytes };
 	let last = 0;
 	let lastFailed = false;
 	let lastBytes = 0;
@@ -101878,13 +101888,14 @@ async function maybeCompactActiveRunForPressure(params) {
 	} catch {}
 	const heapUsedBytes = typeof process?.memoryUsage === "function" ? process.memoryUsage().heapUsed : 0;
 	const reasons = [];
-	/* JOHNNESS_PATCH_APPEND_ONLY_PRESSURE_BASELINE: compare NEW growth since the last successful compaction (restart-safe baseline from the append-only JSONL), not the total never-shrinking file size (family 22). */
+	/* JOHNNESS_PATCH_CONTEXT_WINDOW_COMPACTION: single trigger — estimated active-context tokens vs the context-window threshold (family 74). Legacy byte/message/heap heuristics no longer gate; the family 22 append-only baseline below is computed and recorded for observability only. */
 	const __aopBaselineBytes = computeActivePressureBaselineBytes(params.sessionManager);
 	const __aopEffectiveBytes = effectiveTranscriptPressureBytes(transcriptBytes, __aopBaselineBytes);
-	if (settings.transcriptBytes > 0 && __aopEffectiveBytes >= settings.transcriptBytes) reasons.push("transcriptBytes=" + __aopEffectiveBytes + "/" + settings.transcriptBytes + (__aopBaselineBytes > 0 ? " (new since baseline=" + __aopBaselineBytes + "; raw=" + transcriptBytes + ")" : ""));
-	if (settings.messageCount > 0 && messageCount >= settings.messageCount) reasons.push("messages=" + messageCount + "/" + settings.messageCount);
-	if (settings.heapUsedBytes > 0 && heapUsedBytes >= settings.heapUsedBytes) reasons.push("heapUsed=" + heapUsedBytes + "/" + settings.heapUsedBytes);
-	if (reasons.length === 0) return { refreshed: false, reason: "below threshold", transcriptBytes, messageCount, heapUsedBytes };
+	const __cwcWindowTokens = localPressureGuardContextWindowTokens(params.session?.model);
+	const __cwcThresholdTokens = localPressureGuardContextThreshold(settings, __cwcWindowTokens, params.config);
+	const __cwcEstimatedTokens = estimateMessagesTokens(messages);
+	if (__cwcEstimatedTokens >= __cwcThresholdTokens) reasons.push("contextTokens=" + __cwcEstimatedTokens + "/" + __cwcThresholdTokens);
+	if (reasons.length === 0) return { refreshed: false, reason: "below threshold", contextTokens: __cwcEstimatedTokens, contextThreshold: __cwcThresholdTokens, baselineBytes: __aopBaselineBytes, effectiveTranscriptBytes: __aopEffectiveBytes, transcriptBytes, messageCount, heapUsedBytes };
 	let last = 0;
 	let lastFailed = false;
 	let lastBytes = 0;

package/dist/pi-embedded-DoQsYfIY.js

@@ -96120,9 +96120,7 @@ function buildEmbeddedExtensionFactories(params) {
 
 //#region johnness local active-session pressure guard
 const LOCAL_PRESSURE_GUARD_DEFAULTS = {
-	transcriptBytes: 1048576,
-	messageCount: 220,
-	heapUsedBytes: 3221225472,
+	contextRatio: 0.7,
 	minIntervalMs: 300000,
 	minMessagesAfterCompaction: 24
 };
@@ -96133,14 +96131,24 @@ function localPressureGuardNumber(value, fallback) {
 function localPressureGuardSettings(cfg) {
 	const raw = cfg?.agents?.defaults?.activeSessionPressure ?? cfg?.session?.activePressure ?? {};
 	if (raw?.enabled === false) return null;
+	/* JOHNNESS_PATCH_CONTEXT_WINDOW_COMPACTION: legacy transcriptBytes/messageCount/heapUsedBytes keys are accepted without error and ignored. Config surface: enabled, contextRatio, minIntervalMs, minMessagesAfterCompaction. */
+	const contextRatio = localPressureGuardNumber(raw.contextRatio, LOCAL_PRESSURE_GUARD_DEFAULTS.contextRatio);
 	return {
-		transcriptBytes: localPressureGuardNumber(raw.transcriptBytes ?? raw.maxTranscriptBytes, LOCAL_PRESSURE_GUARD_DEFAULTS.transcriptBytes),
-		messageCount: localPressureGuardNumber(raw.messageCount ?? raw.maxMessages, LOCAL_PRESSURE_GUARD_DEFAULTS.messageCount),
-		heapUsedBytes: localPressureGuardNumber(raw.heapUsedBytes ?? raw.maxHeapUsedBytes, LOCAL_PRESSURE_GUARD_DEFAULTS.heapUsedBytes),
+		contextRatio: contextRatio > 0 ? contextRatio : LOCAL_PRESSURE_GUARD_DEFAULTS.contextRatio,
 		minIntervalMs: localPressureGuardNumber(raw.minIntervalMs, LOCAL_PRESSURE_GUARD_DEFAULTS.minIntervalMs),
 		minMessagesAfterCompaction: localPressureGuardNumber(raw.minMessagesAfterCompaction, LOCAL_PRESSURE_GUARD_DEFAULTS.minMessagesAfterCompaction)
 	};
 }
+/* JOHNNESS_PATCH_CONTEXT_WINDOW_COMPACTION: single pressure trigger — estimated active-context tokens vs the active model's declared context window; the window falls back to DEFAULT_CONTEXT_TOKENS when undeclared. See docs/PATCHES.md family 74. */
+function localPressureGuardContextWindowTokens(model) {
+	const declared = typeof model?.contextWindow === "number" && Number.isFinite(model.contextWindow) ? Math.floor(model.contextWindow) : 0;
+	return declared > 0 ? declared : DEFAULT_CONTEXT_TOKENS;
+}
+function localPressureGuardContextThreshold(settings, contextWindowTokens, cfg) {
+	const reserveTokensFloor = localPressureGuardNumber(cfg?.agents?.defaults?.compaction?.reserveTokensFloor, 2e4);
+	const floor = Math.max(reserveTokensFloor, 50000);
+	return Math.max(Math.ceil(settings.contextRatio * contextWindowTokens), floor);
+}
 /* JOHNNESS_PATCH_APPEND_ONLY_PRESSURE_BASELINE: derive a restart-safe byte baseline from the append-only JSONL so transcriptBytes pressure measures NEW growth since the last successful compaction rather than total (never-shrinking) file size. See docs/PATCHES.md family 22. Scans getEntries() newest->oldest for the newest SUCCESSFUL johnness.active-pressure-refresh marker (failed !== true) and returns its recorded post-compaction size (postCompactionBytes, else transcriptBytes). Returns 0 for legacy sessions with no such marker, so a first over-limit session still compacts once. */
 function computeActivePressureBaselineBytes(sessionManager) {
 	try {
@@ -96298,12 +96306,12 @@ function compactionRotationNumber(value, fallback) {
 	return Number.isFinite(parsed) && parsed >= 0 ? parsed : fallback;
 }
 function compactionRotationSettings(cfg) {
-	const pressure = localPressureGuardSettings(cfg) ?? { transcriptBytes: 1048576 };
 	const raw = cfg?.agents?.defaults?.activeSessionPressure ?? cfg?.session?.activePressure ?? {};
 	if (raw?.rotateAfterCompaction === false) return null;
-	const threshold = compactionRotationNumber(pressure?.transcriptBytes, 1048576);
+	/* JOHNNESS_PATCH_CONTEXT_WINDOW_COMPACTION: the legacy transcriptBytes threshold no longer exists; the rotation size budget keeps its historical 1 MiB-derived default so default rotation behavior is unchanged. */
+	const threshold = 1048576;
 	const configuredMax = compactionRotationNumber(raw.rotationMaxBytes, 0);
-	const maxBytes = configuredMax > 0 ? configuredMax : threshold > 0 ? Math.max(131072, Math.min(524288, Math.floor(threshold * .45))) : 524288;
+	const maxBytes = configuredMax > 0 ? configuredMax : Math.max(131072, Math.min(524288, Math.floor(threshold * .45)));
 	return {
 		maxBytes,
 		maxMessages: Math.max(4, Math.floor(compactionRotationNumber(raw.rotationMaxMessages, 80))),
@@ -96439,18 +96447,20 @@ async function maybeRefreshActiveSessionForPressure(params) {
 	const heapUsedBytes = typeof process?.memoryUsage === "function" ? process.memoryUsage().heapUsed : 0;
 	const __hasCompactedContext = Array.isArray(params.session.messages) && params.session.messages.some((message) => message?.role === "compactionSummary");
 	const __alreadyRotated = params.sessionManager.getEntries?.().some((entry) => entry?.type === "custom" && entry?.customType === "johnness.compaction-session-rotation");
-	if (settings.transcriptBytes > 0 && transcriptBytes >= settings.transcriptBytes && __hasCompactedContext && !__alreadyRotated) {
+	/* JOHNNESS_PATCH_CONTEXT_WINDOW_COMPACTION: the existing-compacted-lineage rotation follows the same single context-token trigger; the byte gate is gone (family 74). */
+	const __cwcWindowTokens = localPressureGuardContextWindowTokens(params.session?.model);
+	const __cwcThresholdTokens = localPressureGuardContextThreshold(settings, __cwcWindowTokens, params.config);
+	const __cwcEstimatedTokens = estimateMessagesTokens(messages);
+	if (__cwcEstimatedTokens >= __cwcThresholdTokens && __hasCompactedContext && !__alreadyRotated) {
 		const __existingRotation = await rotateCompactedSessionLineage({ config: params.config, session: params.session, sessionManager: params.sessionManager, sessionFile: params.sessionFile, sessionKey: params.sessionKey, sessionId: params.sessionId, reason: "existing-compacted-lineage" });
 		if (__existingRotation.rotated) return { refreshed: true, rotated: true, reasons: ["existing-compacted-lineage"], transcriptBytes: __existingRotation.transcriptBytes, messageCount: __existingRotation.messageCount, heapUsedBytes, sessionId: __existingRotation.sessionId, sessionFile: __existingRotation.sessionFile, oldSessionId: __existingRotation.oldSessionId, oldSessionFile: __existingRotation.oldSessionFile };
 	}
 	const reasons = [];
-	/* JOHNNESS_PATCH_APPEND_ONLY_PRESSURE_BASELINE: compare NEW growth since the last successful compaction (restart-safe baseline from the append-only JSONL), not the total never-shrinking file size (family 22). */
+	/* JOHNNESS_PATCH_CONTEXT_WINDOW_COMPACTION: single trigger — estimated active-context tokens vs the context-window threshold (family 74). Legacy byte/message/heap heuristics no longer gate; the family 22 append-only baseline below is computed and recorded for observability only. */
 	const __aopBaselineBytes = computeActivePressureBaselineBytes(params.sessionManager);
 	const __aopEffectiveBytes = effectiveTranscriptPressureBytes(transcriptBytes, __aopBaselineBytes);
-	if (settings.transcriptBytes > 0 && __aopEffectiveBytes >= settings.transcriptBytes) reasons.push("transcriptBytes=" + __aopEffectiveBytes + "/" + settings.transcriptBytes + (__aopBaselineBytes > 0 ? " (new since baseline=" + __aopBaselineBytes + "; raw=" + transcriptBytes + ")" : ""));
-	if (settings.messageCount > 0 && messageCount >= settings.messageCount) reasons.push("messages=" + messageCount + "/" + settings.messageCount);
-	if (settings.heapUsedBytes > 0 && heapUsedBytes >= settings.heapUsedBytes) reasons.push("heapUsed=" + heapUsedBytes + "/" + settings.heapUsedBytes);
-	if (reasons.length === 0) return { refreshed: false, reason: "below threshold", transcriptBytes, messageCount, heapUsedBytes };
+	if (__cwcEstimatedTokens >= __cwcThresholdTokens) reasons.push("contextTokens=" + __cwcEstimatedTokens + "/" + __cwcThresholdTokens);
+	if (reasons.length === 0) return { refreshed: false, reason: "below threshold", contextTokens: __cwcEstimatedTokens, contextThreshold: __cwcThresholdTokens, baselineBytes: __aopBaselineBytes, effectiveTranscriptBytes: __aopEffectiveBytes, transcriptBytes, messageCount, heapUsedBytes };
 	let last = 0;
 	let lastFailed = false;
 	let lastBytes = 0;
@@ -101868,13 +101878,14 @@ async function maybeCompactActiveRunForPressure(params) {
 	} catch {}
 	const heapUsedBytes = typeof process?.memoryUsage === "function" ? process.memoryUsage().heapUsed : 0;
 	const reasons = [];
-	/* JOHNNESS_PATCH_APPEND_ONLY_PRESSURE_BASELINE: compare NEW growth since the last successful compaction (restart-safe baseline from the append-only JSONL), not the total never-shrinking file size (family 22). */
+	/* JOHNNESS_PATCH_CONTEXT_WINDOW_COMPACTION: single trigger — estimated active-context tokens vs the context-window threshold (family 74). Legacy byte/message/heap heuristics no longer gate; the family 22 append-only baseline below is computed and recorded for observability only. */
 	const __aopBaselineBytes = computeActivePressureBaselineBytes(params.sessionManager);
 	const __aopEffectiveBytes = effectiveTranscriptPressureBytes(transcriptBytes, __aopBaselineBytes);
-	if (settings.transcriptBytes > 0 && __aopEffectiveBytes >= settings.transcriptBytes) reasons.push("transcriptBytes=" + __aopEffectiveBytes + "/" + settings.transcriptBytes + (__aopBaselineBytes > 0 ? " (new since baseline=" + __aopBaselineBytes + "; raw=" + transcriptBytes + ")" : ""));
-	if (settings.messageCount > 0 && messageCount >= settings.messageCount) reasons.push("messages=" + messageCount + "/" + settings.messageCount);
-	if (settings.heapUsedBytes > 0 && heapUsedBytes >= settings.heapUsedBytes) reasons.push("heapUsed=" + heapUsedBytes + "/" + settings.heapUsedBytes);
-	if (reasons.length === 0) return { refreshed: false, reason: "below threshold", transcriptBytes, messageCount, heapUsedBytes };
+	const __cwcWindowTokens = localPressureGuardContextWindowTokens(params.session?.model);
+	const __cwcThresholdTokens = localPressureGuardContextThreshold(settings, __cwcWindowTokens, params.config);
+	const __cwcEstimatedTokens = estimateMessagesTokens(messages);
+	if (__cwcEstimatedTokens >= __cwcThresholdTokens) reasons.push("contextTokens=" + __cwcEstimatedTokens + "/" + __cwcThresholdTokens);
+	if (reasons.length === 0) return { refreshed: false, reason: "below threshold", contextTokens: __cwcEstimatedTokens, contextThreshold: __cwcThresholdTokens, baselineBytes: __aopBaselineBytes, effectiveTranscriptBytes: __aopEffectiveBytes, transcriptBytes, messageCount, heapUsedBytes };
 	let last = 0;
 	let lastFailed = false;
 	let lastBytes = 0;

package/dist/reply-C5LKjXcC.js

@@ -74132,9 +74132,7 @@ function buildEmbeddedExtensionFactories(params) {
 
 //#region johnness local active-session pressure guard
 const LOCAL_PRESSURE_GUARD_DEFAULTS = {
-	transcriptBytes: 1048576,
-	messageCount: 220,
-	heapUsedBytes: 3221225472,
+	contextRatio: 0.7,
 	minIntervalMs: 300000,
 	minMessagesAfterCompaction: 24
 };
@@ -74145,14 +74143,24 @@ function localPressureGuardNumber(value, fallback) {
 function localPressureGuardSettings(cfg) {
 	const raw = cfg?.agents?.defaults?.activeSessionPressure ?? cfg?.session?.activePressure ?? {};
 	if (raw?.enabled === false) return null;
+	/* JOHNNESS_PATCH_CONTEXT_WINDOW_COMPACTION: legacy transcriptBytes/messageCount/heapUsedBytes keys are accepted without error and ignored. Config surface: enabled, contextRatio, minIntervalMs, minMessagesAfterCompaction. */
+	const contextRatio = localPressureGuardNumber(raw.contextRatio, LOCAL_PRESSURE_GUARD_DEFAULTS.contextRatio);
 	return {
-		transcriptBytes: localPressureGuardNumber(raw.transcriptBytes ?? raw.maxTranscriptBytes, LOCAL_PRESSURE_GUARD_DEFAULTS.transcriptBytes),
-		messageCount: localPressureGuardNumber(raw.messageCount ?? raw.maxMessages, LOCAL_PRESSURE_GUARD_DEFAULTS.messageCount),
-		heapUsedBytes: localPressureGuardNumber(raw.heapUsedBytes ?? raw.maxHeapUsedBytes, LOCAL_PRESSURE_GUARD_DEFAULTS.heapUsedBytes),
+		contextRatio: contextRatio > 0 ? contextRatio : LOCAL_PRESSURE_GUARD_DEFAULTS.contextRatio,
 		minIntervalMs: localPressureGuardNumber(raw.minIntervalMs, LOCAL_PRESSURE_GUARD_DEFAULTS.minIntervalMs),
 		minMessagesAfterCompaction: localPressureGuardNumber(raw.minMessagesAfterCompaction, LOCAL_PRESSURE_GUARD_DEFAULTS.minMessagesAfterCompaction)
 	};
 }
+/* JOHNNESS_PATCH_CONTEXT_WINDOW_COMPACTION: single pressure trigger — estimated active-context tokens vs the active model's declared context window; the window falls back to DEFAULT_CONTEXT_TOKENS when undeclared. See docs/PATCHES.md family 74. */
+function localPressureGuardContextWindowTokens(model) {
+	const declared = typeof model?.contextWindow === "number" && Number.isFinite(model.contextWindow) ? Math.floor(model.contextWindow) : 0;
+	return declared > 0 ? declared : DEFAULT_CONTEXT_TOKENS;
+}
+function localPressureGuardContextThreshold(settings, contextWindowTokens, cfg) {
+	const reserveTokensFloor = localPressureGuardNumber(cfg?.agents?.defaults?.compaction?.reserveTokensFloor, 2e4);
+	const floor = Math.max(reserveTokensFloor, 50000);
+	return Math.max(Math.ceil(settings.contextRatio * contextWindowTokens), floor);
+}
 /* JOHNNESS_PATCH_APPEND_ONLY_PRESSURE_BASELINE: derive a restart-safe byte baseline from the append-only JSONL so transcriptBytes pressure measures NEW growth since the last successful compaction rather than total (never-shrinking) file size. See docs/PATCHES.md family 22. Scans getEntries() newest->oldest for the newest SUCCESSFUL johnness.active-pressure-refresh marker (failed !== true) and returns its recorded post-compaction size (postCompactionBytes, else transcriptBytes). Returns 0 for legacy sessions with no such marker, so a first over-limit session still compacts once. */
 function computeActivePressureBaselineBytes(sessionManager) {
 	try {
@@ -74310,12 +74318,12 @@ function compactionRotationNumber(value, fallback) {
 	return Number.isFinite(parsed) && parsed >= 0 ? parsed : fallback;
 }
 function compactionRotationSettings(cfg) {
-	const pressure = localPressureGuardSettings(cfg) ?? { transcriptBytes: 1048576 };
 	const raw = cfg?.agents?.defaults?.activeSessionPressure ?? cfg?.session?.activePressure ?? {};
 	if (raw?.rotateAfterCompaction === false) return null;
-	const threshold = compactionRotationNumber(pressure?.transcriptBytes, 1048576);
+	/* JOHNNESS_PATCH_CONTEXT_WINDOW_COMPACTION: the legacy transcriptBytes threshold no longer exists; the rotation size budget keeps its historical 1 MiB-derived default so default rotation behavior is unchanged. */
+	const threshold = 1048576;
 	const configuredMax = compactionRotationNumber(raw.rotationMaxBytes, 0);
-	const maxBytes = configuredMax > 0 ? configuredMax : threshold > 0 ? Math.max(131072, Math.min(524288, Math.floor(threshold * .45))) : 524288;
+	const maxBytes = configuredMax > 0 ? configuredMax : Math.max(131072, Math.min(524288, Math.floor(threshold * .45)));
 	return {
 		maxBytes,
 		maxMessages: Math.max(4, Math.floor(compactionRotationNumber(raw.rotationMaxMessages, 80))),
@@ -74451,18 +74459,20 @@ async function maybeRefreshActiveSessionForPressure(params) {
 	const heapUsedBytes = typeof process?.memoryUsage === "function" ? process.memoryUsage().heapUsed : 0;
 	const __hasCompactedContext = Array.isArray(params.session.messages) && params.session.messages.some((message) => message?.role === "compactionSummary");
 	const __alreadyRotated = params.sessionManager.getEntries?.().some((entry) => entry?.type === "custom" && entry?.customType === "johnness.compaction-session-rotation");
-	if (settings.transcriptBytes > 0 && transcriptBytes >= settings.transcriptBytes && __hasCompactedContext && !__alreadyRotated) {
+	/* JOHNNESS_PATCH_CONTEXT_WINDOW_COMPACTION: the existing-compacted-lineage rotation follows the same single context-token trigger; the byte gate is gone (family 74). */
+	const __cwcWindowTokens = localPressureGuardContextWindowTokens(params.session?.model);
+	const __cwcThresholdTokens = localPressureGuardContextThreshold(settings, __cwcWindowTokens, params.config);
+	const __cwcEstimatedTokens = estimateMessagesTokens(messages);
+	if (__cwcEstimatedTokens >= __cwcThresholdTokens && __hasCompactedContext && !__alreadyRotated) {
 		const __existingRotation = await rotateCompactedSessionLineage({ config: params.config, session: params.session, sessionManager: params.sessionManager, sessionFile: params.sessionFile, sessionKey: params.sessionKey, sessionId: params.sessionId, reason: "existing-compacted-lineage" });
 		if (__existingRotation.rotated) return { refreshed: true, rotated: true, reasons: ["existing-compacted-lineage"], transcriptBytes: __existingRotation.transcriptBytes, messageCount: __existingRotation.messageCount, heapUsedBytes, sessionId: __existingRotation.sessionId, sessionFile: __existingRotation.sessionFile, oldSessionId: __existingRotation.oldSessionId, oldSessionFile: __existingRotation.oldSessionFile };
 	}
 	const reasons = [];
-	/* JOHNNESS_PATCH_APPEND_ONLY_PRESSURE_BASELINE: compare NEW growth since the last successful compaction (restart-safe baseline from the append-only JSONL), not the total never-shrinking file size (family 22). */
+	/* JOHNNESS_PATCH_CONTEXT_WINDOW_COMPACTION: single trigger — estimated active-context tokens vs the context-window threshold (family 74). Legacy byte/message/heap heuristics no longer gate; the family 22 append-only baseline below is computed and recorded for observability only. */
 	const __aopBaselineBytes = computeActivePressureBaselineBytes(params.sessionManager);
 	const __aopEffectiveBytes = effectiveTranscriptPressureBytes(transcriptBytes, __aopBaselineBytes);
-	if (settings.transcriptBytes > 0 && __aopEffectiveBytes >= settings.transcriptBytes) reasons.push("transcriptBytes=" + __aopEffectiveBytes + "/" + settings.transcriptBytes + (__aopBaselineBytes > 0 ? " (new since baseline=" + __aopBaselineBytes + "; raw=" + transcriptBytes + ")" : ""));
-	if (settings.messageCount > 0 && messageCount >= settings.messageCount) reasons.push("messages=" + messageCount + "/" + settings.messageCount);
-	if (settings.heapUsedBytes > 0 && heapUsedBytes >= settings.heapUsedBytes) reasons.push("heapUsed=" + heapUsedBytes + "/" + settings.heapUsedBytes);
-	if (reasons.length === 0) return { refreshed: false, reason: "below threshold", transcriptBytes, messageCount, heapUsedBytes };
+	if (__cwcEstimatedTokens >= __cwcThresholdTokens) reasons.push("contextTokens=" + __cwcEstimatedTokens + "/" + __cwcThresholdTokens);
+	if (reasons.length === 0) return { refreshed: false, reason: "below threshold", contextTokens: __cwcEstimatedTokens, contextThreshold: __cwcThresholdTokens, baselineBytes: __aopBaselineBytes, effectiveTranscriptBytes: __aopEffectiveBytes, transcriptBytes, messageCount, heapUsedBytes };
 	let last = 0;
 	let lastFailed = false;
 	let lastBytes = 0;
@@ -79883,13 +79893,14 @@ async function maybeCompactActiveRunForPressure(params) {
 	} catch {}
 	const heapUsedBytes = typeof process?.memoryUsage === "function" ? process.memoryUsage().heapUsed : 0;
 	const reasons = [];
-	/* JOHNNESS_PATCH_APPEND_ONLY_PRESSURE_BASELINE: compare NEW growth since the last successful compaction (restart-safe baseline from the append-only JSONL), not the total never-shrinking file size (family 22). */
+	/* JOHNNESS_PATCH_CONTEXT_WINDOW_COMPACTION: single trigger — estimated active-context tokens vs the context-window threshold (family 74). Legacy byte/message/heap heuristics no longer gate; the family 22 append-only baseline below is computed and recorded for observability only. */
 	const __aopBaselineBytes = computeActivePressureBaselineBytes(params.sessionManager);
 	const __aopEffectiveBytes = effectiveTranscriptPressureBytes(transcriptBytes, __aopBaselineBytes);
-	if (settings.transcriptBytes > 0 && __aopEffectiveBytes >= settings.transcriptBytes) reasons.push("transcriptBytes=" + __aopEffectiveBytes + "/" + settings.transcriptBytes + (__aopBaselineBytes > 0 ? " (new since baseline=" + __aopBaselineBytes + "; raw=" + transcriptBytes + ")" : ""));
-	if (settings.messageCount > 0 && messageCount >= settings.messageCount) reasons.push("messages=" + messageCount + "/" + settings.messageCount);
-	if (settings.heapUsedBytes > 0 && heapUsedBytes >= settings.heapUsedBytes) reasons.push("heapUsed=" + heapUsedBytes + "/" + settings.heapUsedBytes);
-	if (reasons.length === 0) return { refreshed: false, reason: "below threshold", transcriptBytes, messageCount, heapUsedBytes };
+	const __cwcWindowTokens = localPressureGuardContextWindowTokens(params.session?.model);
+	const __cwcThresholdTokens = localPressureGuardContextThreshold(settings, __cwcWindowTokens, params.config);
+	const __cwcEstimatedTokens = estimateMessagesTokens(messages);
+	if (__cwcEstimatedTokens >= __cwcThresholdTokens) reasons.push("contextTokens=" + __cwcEstimatedTokens + "/" + __cwcThresholdTokens);
+	if (reasons.length === 0) return { refreshed: false, reason: "below threshold", contextTokens: __cwcEstimatedTokens, contextThreshold: __cwcThresholdTokens, baselineBytes: __aopBaselineBytes, effectiveTranscriptBytes: __aopEffectiveBytes, transcriptBytes, messageCount, heapUsedBytes };
 	let last = 0;
 	let lastFailed = false;
 	let lastBytes = 0;

package/docs/PATCHES.md

@@ -17,7 +17,7 @@ they predate ~2026-06-06).
 
 ## Index
 
-Family 01 is retired. Families 02 through 41, 43 through 72, including the 65.1 amendment, are live on `main`; the `latest` and `extended` release lines converged at 2026.9.24. Family 42 is omitted; native Codex OAuth ownership is Family 58.
+Family 01 is retired. Families 02 through 41, 43 through 74, including the 65.1 amendment, are live on `main`; the `latest` and `extended` release lines converged at 2026.9.24. Family 42 is omitted; native Codex OAuth ownership is Family 58.
 
 1. **01, announce queue TTL, retired:** Removed the five-minute age filter that could discard delayed sub-agent completion announcements. [Details](#01--announce-ttl-retired-2026-07-30)
 2. **02, failover rotation:** Returns model selection to the highest-priority profile after its cooldown ends. [Details](#02--failover-rotation-undocumented)
@@ -99,6 +99,7 @@ Family 01 is retired. Families 02 through 41, 43 through 72, including the 65.1
 
 72. **72, bundled generic skills:** Eight workspace skills (code operations, research, todos, planning, logs, and more) ship as generic bundled skills that load by description match, with no owner names or personal paths. [Details](#family-72-bundled-generic-skills)
 73. **73, prune bundled skills:** `content-story`, `dashboard` and `cold-outreach` are removed from the Family 72 bundled set on request after `2026.9.25`; the remaining eight ship unchanged. [Details](#family-73-prune-bundled-skills)
+74. **74, context-window compaction:** The local active-session pressure guard becomes single-trigger and model-aware: estimated active-context tokens against 70% of the active model's declared context window, floored at `max(compaction.reserveTokensFloor, 50000)`; the 1 MB byte, 220-message and 3 GiB heap triggers are removed. [Details](#family-74-context-window-compaction)
 
 ---
 
@@ -729,6 +730,7 @@ already-compacting / compaction-disabled / cooldown are no-ops.
 ---
 
 ## 22 - append-only-pressure-baseline (2026-07-29)
+**Superseded by Family 74:** the append-only baseline is still computed and recorded on the `active-pressure-refresh` marker (see below), but nothing gates on it any more; the pressure trigger is now estimated context tokens against the active model's declared context window.
 **What:** The active-pressure `transcriptBytes` trigger now measures **NEW transcript
 growth since the last successful compaction** instead of the total, never-shrinking
 append-only file size. This removes the false-retrigger loop where a session that has
@@ -2669,3 +2671,17 @@ Three of the Family 72 bundled skills are removed from the bundled set on reques
 The counters move with the set: the verifier emits one check per remaining skill through the same 72.1 loop (455 checks in total), `tests/bundled-skills.test.mjs` asserts exactly eight bundled skills, `scripts/check-packed-artifact.mjs` fails closed unless a tarball carries all eight with frontmatter, and `README.md` and the pinned copy in `patches/59-migrate-cli.patch` carry the new counts together so check 59c stays green. `RULES.md` section 5 drops the three rows.
 
 Roll back with `npm i -g johns-harness@2026.9.25`.
+
+## Family 74: context-window compaction
+
+**Release:** `johns-harness@2026.9.27`. Published as `latest` and the `extended` dist-tag moves to it.
+
+A live gateway session compacted at `messages=221/220` while its transcript held only 528 KB of a 1,048,576-byte budget. The guard was firing on a fixed message count, which is the wrong unit: 221 small messages can sit comfortably inside a large context window, and a few large tool outputs can overflow a small one. Family 74 makes the local active-session pressure guard single-trigger and model-aware.
+
+The only trigger is estimated active-context tokens (`estimateMessagesTokens`, the estimator the compaction path already uses) against `max(ceil(contextRatio x activeModel.contextWindow), max(compaction.reserveTokensFloor, 50000))`, with `contextRatio` defaulting to 0.7. A model that declares no `contextWindow` falls back to the runtime default. Both the startup rotation gate and the in-loop refresh push only `contextTokens=<n>/<threshold>`.
+
+The `transcriptBytes` (1 MB), `messageCount` (220) and `heapUsedBytes` (3 GiB) triggers are removed by design; the legacy config keys are accepted without error and ignored. The 5-minute cooldown, the 24-message post-compaction floor and the failed-attempt growth gate stay as anti-thrash throttles, and the emergency compaction path (Family 13) is untouched, so a genuine overflow is still caught. The Family 22 append-only baseline is still computed and recorded on the `active-pressure-refresh` marker for observability, but nothing gates on it. The config surface is `enabled`, `contextRatio`, `minIntervalMs`, `minMessagesAfterCompaction`.
+
+Verifier checks 74.1 through 74.4 prove the marker, the absence of the byte/message/heap triggers, the threshold math and the config parsing in all four pressure bundles; `scripts/patch-context-window-compaction.py` applies the same change to any install root (both naming eras), is idempotent with `--check`, and fails closed on anchor drift. `patches/provider-catalog/changes.json` pins the four pressure bundles by hash, so `scripts/patch-provider-catalog.py` maps the Family 74 rewrite back to its pre-image in `family67_surface`, exactly as it already does for Family 68.
+
+Roll back with `npm i -g johns-harness@2026.9.26`.

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "johns-harness",
-  "version": "2026.9.26",
+  "version": "2026.9.27",
   "description": "John's Harness: a production agent harness that runs autonomous coding agents as one cooperative swarm.",
   "keywords": [
     "ai-agents",