harness-everything 0.11.5 → 0.11.6

package/.claude-plugin/marketplace.json

@@ -10,7 +10,7 @@
       "name": "harness-everything",
       "source": "./",
       "description": "Harness OS: an orchestrated agent operating system — tier routing, circuit breaker guardrails, verification gates, WAL session memory, and self-evolving workspace rules.",
-      "version": "0.11.5",
+      "version": "0.11.6",
       "author": {
         "name": "Miya Daniel"
       },

package/.claude-plugin/plugin.json

@@ -1,7 +1,7 @@
 {
   "name": "harness-everything",
   "description": "Harness OS: an orchestrated agent operating system — tier routing, circuit breaker guardrails, verification gates, WAL session memory, and self-evolving workspace rules.",
-  "version": "0.11.5",
+  "version": "0.11.6",
   "author": {
     "name": "Miya Daniel"
   },

package/CHANGELOG.md

@@ -2,6 +2,25 @@
 
 All notable changes to this project are documented in this file.
 
+## [0.11.6](https://github.com/dyphn1/Harness-everything/compare/v0.11.5...v0.11.6) (2026-09-18)
+
+### Bug Fixes
+
+* **workflow:** avoid workflow rewrite after probe registration ([b9f2e06](https://github.com/dyphn1/Harness-everything/commit/b9f2e065114d70e927b6f75e7c0f7899d9e8392f))
+* **workflow:** correlate mutation probes by tool-use identity ([86cd1a1](https://github.com/dyphn1/Harness-everything/commit/86cd1a1c6eaef6518936160b0c35f94a3e652cae))
+* **workflow:** defer shell mutation accounting to observer ([c543fd0](https://github.com/dyphn1/Harness-everything/commit/c543fd0c41e18f05c21235ce4ffcf28864200940))
+* **workflow:** discard denied shell mutation probes ([7a8757c](https://github.com/dyphn1/Harness-everything/commit/7a8757cab821e524ccdd17e416e013bfbadc4f39))
+* **workflow:** fingerprint observed workspace effects ([9d8e316](https://github.com/dyphn1/Harness-everything/commit/9d8e316eb97ec7bbcd0e46ceb51491af90e57e9f))
+* **workflow:** fingerprint visible content independent of staging ([a1f370e](https://github.com/dyphn1/Harness-everything/commit/a1f370e64a97b04bee4d8df23e3ffdc7552377b8))
+* **workflow:** isolate mutation probes per tool call ([70d6052](https://github.com/dyphn1/Harness-everything/commit/70d6052ef02fce1125da9a20a14a4ad65d09ba93))
+* **workflow:** make mutation observation single-handler ordered ([ea166e1](https://github.com/dyphn1/Harness-everything/commit/ea166e1a94cdd1725cd7242feb5b0716a4dce8b8))
+* **workflow:** normalize deleted paths in fingerprint ([e5019da](https://github.com/dyphn1/Harness-everything/commit/e5019daee870e806d77d92c741a9624cd6c9bfe0))
+* **workflow:** observe shell effects before state persistence ([5271655](https://github.com/dyphn1/Harness-everything/commit/5271655627344eb71ffeb8dd36b007709862b9d0))
+* **workflow:** observe shell workspace mutations ([c547b39](https://github.com/dyphn1/Harness-everything/commit/c547b396bf2e3829b83ed03abeca594b7391f542))
+* **workflow:** remove parallel standalone mutation observer ([20027f6](https://github.com/dyphn1/Harness-everything/commit/20027f68f5b63719ea7a96a524f49c7103586644))
+* **workflow:** reserve shell mutation probes ([e7273f6](https://github.com/dyphn1/Harness-everything/commit/e7273f66010d5b0c740b87b8dbc020c8de616fe7))
+* **workflow:** settle mutation probes inside state persistence ([88feaf4](https://github.com/dyphn1/Harness-everything/commit/88feaf4ee3accbd8504b745fd4e79e548c31bd6d))
+
 ## [0.11.5](https://github.com/dyphn1/Harness-everything/compare/v0.11.4...v0.11.5) (2026-09-18)
 
 ### Bug Fixes

package/ci/mechanism-2ah-workflow-lifecycle.test.js

@@ -35,6 +35,46 @@ function git(args) {
 }
 const route = prompt => node('harness-everything/scripts/kernel-router.js', { ...payload, prompt });
 const gate = (tool, input, cwd = linked) => node('hooks/scripts/workflow-gate.js', { ...payload, cwd, tool_name: tool, tool_input: input });
+let toolUseSeq = 0;
+function beginShell(command, cwd = repo) {
+  const toolUseId = 'toolu_fixture_' + (++toolUseSeq);
+  const input = { ...payload, cwd, tool_name: 'Bash', tool_use_id: toolUseId, tool_input: { command } };
+  return { command, cwd, toolUseId, result: node('hooks/scripts/workflow-gate.js', input) };
+}
+function persistShell(call, response = {}, hookEvent = 'PostToolUse') {
+  const input = {
+    ...payload,
+    cwd: call.cwd,
+    hook_event_name: hookEvent,
+    tool_name: 'Bash',
+    tool_use_id: call.toolUseId,
+    tool_input: { command: call.command },
+  };
+  if (hookEvent === 'PostToolUseFailure') {
+    input.error = response.error || 'Exit code 1\nfixture failure';
+    input.is_interrupt = Boolean(response.is_interrupt);
+  } else {
+    input.tool_response = response;
+  }
+  return node('hooks/scripts/state-persist.js', input);
+}
+function denyShell(call) {
+  return node('hooks/scripts/workflow-mutation-denied.js', {
+    ...payload,
+    cwd: call.cwd,
+    hook_event_name: 'PermissionDenied',
+    permission_mode: 'auto',
+    tool_name: 'Bash',
+    tool_use_id: call.toolUseId,
+    tool_input: { command: call.command },
+    reason: 'Blocked by classifier',
+  });
+}
+const probeDir = path.join(sessionDir, 'mutation-probes');
+const probeFiles = () => {
+  try { return fs.readdirSync(probeDir).filter(name => /^[a-f0-9]{64}\.json$/.test(name)); }
+  catch (_) { return []; }
+};
 const stop = extra => node('hooks/scripts/workflow-stop-gate.js', { ...payload, ...extra });
 const control = (...args) => node('hooks/scripts/workflow-disposition.js', null, [args[0], '--session-id', sessionId, ...args.slice(1)]);
 const stages = [
@@ -183,10 +223,132 @@ try {
   check((afterReadOnly.lastMutationAt || 0) === readOnlyMutationAt,
     'read-only shell composition does not advance lastMutationAt');
 
-  for (let i = 0; i < iterationLimit; i++) {
-    check(gate('Write', { file_path: path.join(repo, `small-${i}.js`) }, repo).status === 0, `iterative mutation ${i + 1}/${iterationLimit} stays within budget`);
+  // #159: safety classification is not mutation evidence. Commands that are
+  // not proven read-only are admitted under an opaque workspace probe while
+  // capacity remains, then accounted only if the visible workspace changed.
+  const untrustedReads = [
+    'git fetch origin --quiet 2>&1; git status; git log --oneline -1',
+    'git remote -v',
+    'gh run view 1 --json status',
+    'node -e "console.log(42)"',
+  ];
+  for (const command of untrustedReads) {
+    const before = read(file);
+    const call = beginShell(command);
+    check(call.result.status === 0, 'untrusted read/query is admitted under observation: ' + command);
+    const probes = probeFiles();
+    check(probes.length === 1 && !fs.readFileSync(path.join(probeDir, probes[0]), 'utf8').includes(command),
+      'mutation probe persists only opaque identity, not raw command text');
+    check(persistShell(call).status === 0, 'no-effect shell probe resolves through state-persist: ' + command);
+    check(probeFiles().length === 0, 'completed shell call removes its mutation probe');
+    const after = read(file);
+    check((after.budget?.counters?.iterations || 0) === (before.budget?.counters?.iterations || 0),
+      'no-effect untrusted shell command consumes no iteration');
+    check((after.lastMutationAt || 0) === (before.lastMutationAt || 0),
+      'no-effect untrusted shell command does not advance lastMutationAt');
+  }
+
+  const deniedCall = beginShell('node denied-by-auto-mode.js');
+  check(deniedCall.result.status === 0 && probeFiles().length === 1,
+    'auto-mode candidate reserves mutation capacity before permission decision');
+  check(denyShell(deniedCall).status === 0 && probeFiles().length === 0,
+    'PermissionDenied releases the exact tool-use mutation probe');
+  const parallelA = beginShell('node repeated-read-only-wrapper.js');
+  const parallelB = beginShell('node repeated-read-only-wrapper.js');
+  check(parallelA.result.status === 0 && parallelB.result.status === 0 && probeFiles().length === 2,
+    'identical concurrent shell commands receive distinct tool-use probes');
+  check(denyShell(parallelA).status === 0 && persistShell(parallelB).status === 0 && probeFiles().length === 0,
+    'parallel tool-use probes settle independently without reservation leakage');
+
+  const metadataFetch = 'git fetch . HEAD:refs/remotes/origin/harness-probe';
+  const beforeMetadataFetch = read(file);
+  const metadataFetchCall = beginShell(metadataFetch);
+  check(metadataFetchCall.result.status === 0,
+    'metadata-only git fetch is admitted under observation');
+  git(['fetch', '.', 'HEAD:refs/remotes/origin/harness-probe']);
+  check(persistShell(metadataFetchCall).status === 0, 'real metadata-only git fetch probe resolves');
+  check((read(file).budget?.counters?.iterations || 0) === (beforeMetadataFetch.budget?.counters?.iterations || 0) &&
+    (read(file).lastMutationAt || 0) === (beforeMetadataFetch.lastMutationAt || 0),
+    'real git fetch changes Git metadata without consuming a code iteration');
+
+  let observedIterations = read(file).budget?.counters?.iterations || 0;
+  const mutateTracked = 'node mutate-tracked.js';
+  const mutateTrackedCall = beginShell(mutateTracked);
+  check(mutateTrackedCall.result.status === 0, 'unknown tracked-file mutator is admitted under observation');
+  fs.writeFileSync(path.join(repo, 'README.md'), 'fixture changed once\n');
+  check(persistShell(mutateTrackedCall).status === 0, 'tracked-file mutation is observed');
+  check(read(file).budget.counters.iterations === ++observedIterations, 'tracked-file mutation consumes exactly one iteration');
+  const firstObservedAt = read(file).lastMutationAt;
+
+  const mutateDirty = 'node mutate-dirty-again.js';
+  const mutateDirtyCall = beginShell(mutateDirty);
+  check(mutateDirtyCall.result.status === 0, 'unknown already-dirty mutator is admitted under observation');
+  fs.writeFileSync(path.join(repo, 'README.md'), 'fixture changed twice\n');
+  check(persistShell(mutateDirtyCall).status === 0, 'already-dirty content change is observed');
+  check(read(file).budget.counters.iterations === ++observedIterations && read(file).lastMutationAt >= firstObservedAt,
+    'already-dirty content change consumes one new iteration');
+
+  const createUntracked = 'node create-untracked.js';
+  const createUntrackedCall = beginShell(createUntracked);
+  check(createUntrackedCall.result.status === 0, 'unknown untracked-file mutator is admitted under observation');
+  fs.writeFileSync(path.join(repo, 'observed-untracked.txt'), 'one\n');
+  check(persistShell(createUntrackedCall).status === 0, 'untracked-file creation is observed');
+  check(read(file).budget.counters.iterations === ++observedIterations, 'untracked-file creation consumes exactly one iteration');
+
+  const failedMutation = 'node fail-after-write.js';
+  const failedMutationCall = beginShell(failedMutation);
+  check(failedMutationCall.result.status === 0, 'failing shell mutator is admitted under observation');
+  fs.writeFileSync(path.join(repo, 'observed-failed.txt'), 'written before failure\n');
+  check(persistShell(failedMutationCall, { error: 'Exit code 7\nfixture failure' }, 'PostToolUseFailure').status === 0,
+    'failed shell command still reports workspace mutation through failure lifecycle');
+  check(read(file).budget.counters.iterations === ++observedIterations,
+    'failed shell command that changed workspace still consumes one iteration');
+
+  const beforeStageOnly = read(file);
+  const stageOnly = 'git add README.md observed-untracked.txt';
+  const stageOnlyCall = beginShell(stageOnly);
+  check(stageOnlyCall.result.status === 0, 'staging-only git command is observed instead of pre-counted');
+  git(['add', 'README.md', 'observed-untracked.txt']);
+  check(persistShell(stageOnlyCall).status === 0, 'staging-only probe resolves');
+  check(read(file).budget.counters.iterations === beforeStageOnly.budget.counters.iterations,
+    'git add does not double-count tracked or previously-untracked visible content');
+
+  const beforeCommitOnly = read(file);
+  const commitOnly = 'git commit -m observed-fixture';
+  const commitOnlyCall = beginShell(commitOnly);
+  check(commitOnlyCall.result.status === 0, 'pure commit is observed instead of pre-counted');
+  git(['-c', 'user.name=Harness Test', '-c', 'user.email=harness@example.invalid', 'commit', '-m', 'observed-fixture']);
+  check(persistShell(commitOnlyCall).status === 0, 'pure commit probe resolves');
+  check(read(file).budget.counters.iterations === beforeCommitOnly.budget.counters.iterations,
+    'pure commit of already-accounted content does not consume another code iteration');
+
+  const deleteTracked = 'node delete-tracked.js';
+  const deleteTrackedCall = beginShell(deleteTracked);
+  check(deleteTrackedCall.result.status === 0, 'unknown tracked-file deletion is admitted under observation');
+  fs.unlinkSync(path.join(repo, 'observed-untracked.txt'));
+  check(persistShell(deleteTrackedCall).status === 0, 'tracked-file deletion is observed');
+  check(read(file).budget.counters.iterations === ++observedIterations,
+    'tracked-file deletion consumes exactly one iteration');
+
+  const beforeStageDelete = read(file);
+  const stageDelete = 'git add -u observed-untracked.txt';
+  const stageDeleteCall = beginShell(stageDelete);
+  check(stageDeleteCall.result.status === 0, 'staging-only deletion is observed instead of pre-counted');
+  git(['add', '-u', 'observed-untracked.txt']);
+  check(persistShell(stageDeleteCall).status === 0, 'staging deletion probe resolves');
+  check(read(file).budget.counters.iterations === beforeStageDelete.budget.counters.iterations,
+    'git add -u does not double-count an already-observed deletion');
+
+  observedIterations = read(file).budget.counters.iterations;
+  for (let i = observedIterations; i < iterationLimit; i++) {
+    check(gate('Write', { file_path: path.join(repo, `small-${i}.js`) }, repo).status === 0, `confirmed mutation ${i + 1}/${iterationLimit} stays within budget`);
   }
-  check(gate('Write', { file_path: path.join(repo, 'small-overflow.js') }, repo).status === 2, 'iteration beyond maxIterations is blocked');
+  check(gate('Bash', { command: 'git status --short' }, repo).status === 0,
+    'proven read-only diagnosis remains allowed at the iteration limit');
+  check(gate('Bash', { command: 'git remote -v' }, repo).status === 2,
+    'untrusted shell command is blocked before execution once actual mutation budget is full');
+  check(gate('Write', { file_path: path.join(repo, 'small-overflow.js') }, repo).status === 2,
+    'blocked exhausted workflow still rejects direct mutation');
   const iterativeExhausted = read(file);
   check(iterativeExhausted.state === 'blocked' && iterativeExhausted.budget?.state === 'budget-exhausted' &&
     iterativeExhausted.budget?.reasonCode === 'iteration-budget-exhausted', 'iterative exhaustion records blocked budget state');
@@ -200,11 +362,16 @@ try {
   check(stop().status === 2, 'direct/iterative route cannot complete unverified mutation');
   check(stop({ stop_hook_active: true }).status === 0 && read(file).state === 'blocked', 'iterative retry reports incomplete state');
   check(control('start').status === 0, 'blocked iterative route can resume without a Fable stage map');
-  // Issue #153: a host is not guaranteed to report a numeric exit code at
-  // all (state-persist.js's own isFailed logic already treats "no exit code
-  // and no stderr" as success), so this must resolve rather than deadlock.
-  node('hooks/scripts/state-persist.js', { ...payload, tool_name: 'Bash', tool_input: { command: 'npm test' }, tool_response: { stdout: 'no code' } });
-  check(stop().status === 0 && read(file).state === 'satisfied', 'verification without a host-reported numeric exit status still resolves completion');
+  // #153 + #159 ordering: when one shell command mutates and then verifies,
+  // observed mutation must land before state-persist records verification.
+  const mutateAndVerify = 'node mutate.js && npm test';
+  const mutateAndVerifyCall = beginShell(mutateAndVerify);
+  check(mutateAndVerifyCall.result.status === 0, 'combined mutation+verification command is admitted under observation');
+  fs.writeFileSync(path.join(repo, 'README.md'), 'fixture verified mutation\n');
+  check(persistShell(mutateAndVerifyCall, { stdout: 'tests passed' }).status === 0,
+    'single state-persist handler records mutation before verification');
+  check(stop().status === 0 && read(file).state === 'satisfied',
+    'verification without a host-reported numeric exit status resolves after same-command mutation');
   node('hooks/scripts/state-persist.js', { ...payload, tool_name: 'Bash', tool_input: { command: 'npm test' }, tool_response: { exitCode: 0, stdout: 'passed' } });
   check(stop().status === 0 && read(file).state === 'satisfied', 'observed iterative verification resolves completion');
   console.log(`PASS: workflow lifecycle (${passed} assertions)`);

package/hooks/hooks.json

@@ -157,6 +157,19 @@
         "id": "harness:stop:skill-telemetry"
       }
     ],
+    "PermissionDenied": [
+      {
+        "matcher": "Bash|PowerShell",
+        "hooks": [
+          {
+            "type": "command",
+            "command": "node \"${CLAUDE_PLUGIN_ROOT}/hooks/scripts/workflow-mutation-denied.js\""
+          }
+        ],
+        "description": "Release reserved mutation capacity when Claude auto mode denies a shell tool call",
+        "id": "harness:permission-denied:mutation-probe"
+      }
+    ],
     "PostToolUse": [
       {
         "matcher": "Bash|PowerShell|mcp__.*",
@@ -188,7 +201,7 @@
             "command": "node \"${CLAUDE_PLUGIN_ROOT}/hooks/scripts/state-persist.js\""
           }
         ],
-        "description": "Save skill transition state",
+        "description": "Persist tool outcome, observed workspace mutation, and verification milestones",
         "id": "harness:post:state-persist"
       },
       {
@@ -259,6 +272,17 @@
         "description": "Track failed tool error signatures for Rule of 3",
         "id": "harness:post-failure:rule-of-3-tracker"
       },
+      {
+        "matcher": "Bash|PowerShell",
+        "hooks": [
+          {
+            "type": "command",
+            "command": "node \"${CLAUDE_PLUGIN_ROOT}/hooks/scripts/state-persist.js\""
+          }
+        ],
+        "description": "Persist failed shell outcome and any workspace mutation that occurred before failure",
+        "id": "harness:post-failure:state-persist"
+      },
       {
         "matcher": "Skill",
         "hooks": [

package/hooks/scripts/lib/workflow-isolation.js

@@ -1,13 +1,89 @@
 'use strict';
 
 const fs = require('fs');
+const crypto = require('crypto');
 const path = require('path');
 const { spawnSync } = require('child_process');
 
-function git(cwd, args) {
-  const result = spawnSync('git', args, { cwd, encoding: 'utf8', windowsHide: true, timeout: 5000 });
+function gitRaw(cwd, args) {
+  const result = spawnSync('git', args, { cwd, encoding: 'utf8', windowsHide: true, timeout: 5000, maxBuffer: 16 * 1024 * 1024 });
   if (result.status !== 0) throw new Error('cannot verify Git worktree isolation');
-  return result.stdout.trim();
+  return result.stdout;
+}
+
+function git(cwd, args) {
+  return gitRaw(cwd, args).trim();
+}
+
+function hashWorktreePath(root, relative) {
+  const target = path.join(root, relative);
+  let stat;
+  try { stat = fs.lstatSync(target); }
+  catch (error) {
+    if (error.code === 'ENOENT') return null;
+    throw error;
+  }
+  if (stat.isDirectory()) {
+    const result = spawnSync('git', ['-C', target, 'rev-parse', 'HEAD'], { encoding: 'utf8', windowsHide: true, timeout: 5000 });
+    return { mode: '160000', oid: result.status === 0 ? result.stdout.trim() : 'directory' };
+  }
+  const mode = stat.isSymbolicLink() ? '120000' : ((stat.mode & 0o111) ? '100755' : '100644');
+  const result = spawnSync('git', ['hash-object', '--path=' + relative, '--', relative], {
+    cwd: root, encoding: 'utf8', windowsHide: true, timeout: 5000,
+  });
+  if (result.status !== 0) throw new Error('cannot fingerprint workspace content');
+  return { mode, oid: result.stdout.trim() };
+}
+
+function workspaceFingerprint(cwd) {
+  const root = canonical(git(cwd, ['rev-parse', '--show-toplevel']));
+  const entries = new Map();
+  for (const record of gitRaw(root, ['ls-files', '-s', '-z']).split('\0')) {
+    if (!record) continue;
+    const tab = record.indexOf('\t');
+    if (tab < 0) throw new Error('cannot fingerprint workspace index');
+    const meta = record.slice(0, tab).trim().split(/\s+/);
+    const relative = record.slice(tab + 1);
+    const mode = meta[0];
+    const oid = meta[1];
+    entries.set(relative, { relative, mode, oid });
+  }
+
+  // Substitute current worktree content for tracked paths that differ from the
+  // index. This keeps the fingerprint stable across git add / git commit when
+  // visible workspace content did not change.
+  for (const relative of gitRaw(root, ['diff-files', '--name-only', '-z']).split('\0').filter(Boolean)) {
+    const value = hashWorktreePath(root, relative);
+    if (value) entries.set(relative, { relative, mode: value.mode, oid: value.oid });
+    else entries.delete(relative);
+  }
+
+  for (const relative of gitRaw(root, ['ls-files', '--others', '--exclude-standard', '-z']).split('\0').filter(Boolean)) {
+    const value = hashWorktreePath(root, relative);
+    entries.set(relative, { relative, mode: value.mode, oid: value.oid });
+  }
+
+  const digest = crypto.createHash('sha256');
+  digest.update('harness-workspace-fingerprint-v1\0');
+  for (const entry of [...entries.values()].sort((a, b) => a.relative.localeCompare(b.relative))) {
+    digest.update(entry.relative); digest.update('\0');
+    digest.update(entry.mode); digest.update('\0');
+    digest.update(entry.oid); digest.update('\0');
+  }
+  return digest.digest('hex');
+}
+
+function shellProbeKey(payload, cwd) {
+  const toolUseId = String(payload?.tool_use_id || payload?.toolUseId || '').trim();
+  const tool = String(payload?.tool_name || payload?.tool || '');
+  const command = commandOf(payload);
+  return crypto.createHash('sha256')
+    .update('harness-shell-probe-v2\0')
+    .update(toolUseId || 'no-tool-use-id').update('\0')
+    .update(tool).update('\0')
+    .update(key(cwd)).update('\0')
+    .update(command)
+    .digest('hex');
 }
 
 function canonical(value) {
@@ -144,9 +220,9 @@ function splitShellSegments(text) {
 
 function classifyShell(command) {
   const text = String(command || '').trim();
-  if (!text) return 'mutation-or-unknown';
+  if (!text) return 'untrusted';
   const segments = splitShellSegments(text);
-  if (!segments) return 'mutation-or-unknown';
+  if (!segments) return 'untrusted';
   if (segments.length === 1) {
     if (/^git\s+worktree\s+add\s+/i.test(segments[0]) && !/\s--(?:force|checkout|detach)(?:\s|=|$)/i.test(segments[0])) return 'worktree-setup';
     return isReadOnlySegment(segments[0]) ? 'read-only' : 'mutation-or-unknown';
@@ -198,4 +274,4 @@ function assertShellScope(command, cwd, isolatedRoot) {
   }
 }
 
-module.exports = { canonical, key, within, linkedWorktree, classifyShell, inputOf, cwdOf, commandOf, mutationPaths, assertTargets, assertShellScope };
+module.exports = { canonical, key, within, linkedWorktree, classifyShell, workspaceFingerprint, shellProbeKey, inputOf, cwdOf, commandOf, mutationPaths, assertTargets, assertShellScope };

package/hooks/scripts/lib/workflow-runtime.js

@@ -43,6 +43,140 @@ function loadWorkflow(payload) {
 
 function saveWorkflow(context) { atomicWriteJson(context.file, context.workflow); }
 
+const MUTATION_PROBE_KEY = /^[a-f0-9]{64}$/;
+const MUTATION_PROBE_WAIT = new Int32Array(new SharedArrayBuffer(4));
+
+function mutationProbeDir(context) {
+  return path.join(context.sessionDir, 'mutation-probes');
+}
+
+function mutationProbeFile(context, key) {
+  if (!MUTATION_PROBE_KEY.test(String(key || ''))) throw new Error('invalid mutation probe identity');
+  return path.join(mutationProbeDir(context), key + '.json');
+}
+
+function withMutationProbeLock(context, callback) {
+  fs.mkdirSync(context.sessionDir, { recursive: true });
+  const lockDir = path.join(context.sessionDir, '.mutation-probes.lock');
+  let acquired = false;
+  for (let attempt = 0; attempt < 200; attempt++) {
+    try {
+      fs.mkdirSync(lockDir);
+      acquired = true;
+      break;
+    } catch (error) {
+      if (error.code !== 'EEXIST') throw error;
+      try {
+        const stat = fs.statSync(lockDir);
+        if (Date.now() - stat.mtimeMs > 30000) {
+          fs.rmSync(lockDir, { recursive: true, force: true });
+          continue;
+        }
+      } catch (_) { /* retry below */ }
+      Atomics.wait(MUTATION_PROBE_WAIT, 0, 0, 10);
+    }
+  }
+  if (!acquired) throw new Error('cannot acquire mutation probe lock');
+  try { return callback(); }
+  finally { fs.rmSync(lockDir, { recursive: true, force: true }); }
+}
+
+function mutationProbeReservations(context) {
+  const dir = mutationProbeDir(context);
+  let files = [];
+  try { files = fs.readdirSync(dir).filter(file => /^[a-f0-9]{64}\.json$/.test(file)); }
+  catch (_) { return 0; }
+  let count = 0;
+  for (const file of files) {
+    const probe = readJson(path.join(dir, file));
+    if (probe?.schemaVersion === 1 && probe.reserveIteration === true) count++;
+  }
+  return count;
+}
+
+function refreshWorkflowForProbe(context) {
+  const latest = readJson(context.file);
+  if (!latest || latest.sessionId !== context.sessionId) throw new Error('cannot refresh workflow for mutation probe');
+  context.workflow = latest;
+  return latest;
+}
+
+function assertIterationCapacity(context, evidence) {
+  const { workflow } = context;
+  if (!workflow) throw budgetError('no workflow available for iteration capacity');
+  const budget = ensureWorkflowBudget(workflow);
+  if (budget.state !== 'active') throw budgetError(`workflow budget is ${budget.state}: ${budget.reasonCode || 'unavailable'}`);
+  const limit = budget.limits.maxIterations;
+  if (limit === null) return { applicable: false, value: budget.counters.iterations, reserved: 0 };
+  const reserved = mutationProbeReservations(context);
+  if (budget.counters.iterations + reserved >= limit) {
+    return exhaustBudget(context, 'iteration-budget-exhausted', evidence || 'untrusted-shell-capacity');
+  }
+  return { applicable: true, value: budget.counters.iterations, reserved, limit };
+}
+
+function registerMutationProbe(context, key, fingerprint, reserveIteration) {
+  if (!MUTATION_PROBE_KEY.test(String(key || '')) || !MUTATION_PROBE_KEY.test(String(fingerprint || ''))) {
+    throw new Error('invalid mutation probe identity');
+  }
+  return withMutationProbeLock(context, () => {
+    refreshWorkflowForProbe(context);
+    if (reserveIteration) assertIterationCapacity(context, 'shell:untrusted');
+    const dir = mutationProbeDir(context);
+    fs.mkdirSync(dir, { recursive: true });
+    const file = mutationProbeFile(context, key);
+    if (fs.existsSync(file)) throw new Error('mutation probe is already pending for this tool call');
+    const probe = {
+      schemaVersion: 1,
+      fingerprint,
+      reserveIteration: Boolean(reserveIteration),
+      observedAt: Date.now(),
+    };
+    atomicWriteJson(file, probe);
+    return probe;
+  });
+}
+
+function peekMutationProbe(context, key) {
+  return readJson(mutationProbeFile(context, key));
+}
+
+function settleMutationProbe(context, key, afterFingerprint, tool) {
+  if (!MUTATION_PROBE_KEY.test(String(afterFingerprint || ''))) throw new Error('invalid mutation probe fingerprint');
+  return withMutationProbeLock(context, () => {
+    const file = mutationProbeFile(context, key);
+    const probe = readJson(file);
+    if (!probe || probe.schemaVersion !== 1) return null;
+    refreshWorkflowForProbe(context);
+    fs.unlinkSync(file);
+    const changed = afterFingerprint !== probe.fingerprint;
+    if (changed) {
+      context.workflow.lastMutationAt = Date.now();
+      if (probe.reserveIteration && context.workflow.strategy === 'iterative-single') {
+        recordBudgetEvent(context, 'iteration', { evidence: String(tool || 'shell') + ':observed-workspace-mutation' });
+      } else {
+        saveWorkflow(context);
+      }
+    }
+    return { changed, probe };
+  });
+}
+
+function discardMutationProbe(context, key) {
+  return withMutationProbeLock(context, () => {
+    const file = mutationProbeFile(context, key);
+    if (!fs.existsSync(file)) return false;
+    fs.unlinkSync(file);
+    return true;
+  });
+}
+
+function clearMutationProbes(context) {
+  return withMutationProbeLock(context, () => {
+    fs.rmSync(mutationProbeDir(context), { recursive: true, force: true });
+  });
+}
+
 function budgetLimits(workflow) {
   const limits = workflow?.workflowPlan?.limits || {};
   return {
@@ -123,7 +257,8 @@ function recordBudgetEvent(context, type, options = {}) {
   const consume = (counter, limitName, reasonCode) => {
     const limit = budget.limits[limitName];
     if (limit === null) return { applicable: false, value: budget.counters[counter] };
-    if (budget.counters[counter] >= limit) return exhaustBudget(context, reasonCode, evidence);
+    const reserved = type === 'iteration' ? mutationProbeReservations(context) : 0;
+    if (budget.counters[counter] + reserved >= limit) return exhaustBudget(context, reasonCode, evidence);
     budget.counters[counter]++;
     appendBudgetEvent(budget, { type, counter, value: budget.counters[counter], limit, evidence, observedAt: now });
     persistBudget(context);
@@ -175,6 +310,7 @@ function resetWorkflowBudget(context, evidence) {
     evidence: String(evidence).trim(),
     resetAt: now,
   }].slice(-16);
+  clearMutationProbes(context);
   workflow.budget = {
     schemaVersion: 1,
     epoch: previous.epoch + 1,
@@ -260,4 +396,4 @@ function readHookInput(decide) {
   process.stdin.on('error', finish);
 }
 
-module.exports = { OPEN_STATES, SAFE_ID, WORKFLOW_CONTROLLER_COMMANDS, isMajorWorkflow, loadWorkflow, saveWorkflow, budgetLimits, ensureWorkflowBudget, recordBudgetEvent, resetWorkflowBudget, syncBudgetToRun, matchingRun, unresolvedStages, readHookInput };
+module.exports = { OPEN_STATES, SAFE_ID, WORKFLOW_CONTROLLER_COMMANDS, isMajorWorkflow, loadWorkflow, saveWorkflow, budgetLimits, ensureWorkflowBudget, mutationProbeReservations, assertIterationCapacity, registerMutationProbe, peekMutationProbe, settleMutationProbe, discardMutationProbe, clearMutationProbes, recordBudgetEvent, resetWorkflowBudget, syncBudgetToRun, matchingRun, unresolvedStages, readHookInput };

package/hooks/scripts/state-persist.js

@@ -2,6 +2,8 @@
 const fs = require('fs');
 const path = require('path');
 const { getWorkspaceRoot, getSessionDir } = require('./lib/harness-state');
+const { loadWorkflow, saveWorkflow, peekMutationProbe, settleMutationProbe } = require('./lib/workflow-runtime');
+const { cwdOf, shellProbeKey, workspaceFingerprint } = require('./lib/workflow-isolation');
 const { observeTool } = require('./lib/telemetry');
 
 // Commands that count as "verification ran" for the Stop gate
@@ -28,7 +30,31 @@ process.stdin.on('end', () => {
   }
 });
 
+function observeWorkspaceMutation(payload, root) {
+  const toolName = payload && (payload.tool_name || payload.tool);
+  if (toolName !== 'Bash' && toolName !== 'PowerShell' && toolName !== 'exec_command') return false;
+  const context = loadWorkflow(payload);
+  if (!context.workflow || context.workflow.state === 'deferred') return false;
+  const cwd = cwdOf(payload, root);
+  const probeKey = shellProbeKey(payload, cwd);
+  if (!peekMutationProbe(context, probeKey)) return false;
+  let after;
+  try {
+    after = workspaceFingerprint(cwd);
+  } catch (error) {
+    context.workflow.state = 'blocked';
+    context.workflow.blockReason = 'mutation-observation-failed';
+    saveWorkflow(context);
+    const critical = new Error('cannot verify whether shell execution changed workspace content');
+    critical.code = 'HARNESS_MUTATION_OBSERVATION';
+    throw critical;
+  }
+  const result = settleMutationProbe(context, probeKey, after, toolName);
+  return Boolean(result && result.changed);
+}
+
 function processState(payload) {
+  let exitCodeForHook = 0;
   try {
     const sessionId = payload && (payload.session_id || payload.sessionId);
     const root = getWorkspaceRoot(payload);
@@ -48,12 +74,16 @@ function processState(payload) {
       // it's actually a number; fall back to a stderr signal otherwise.
       const toolResponse = payload.tool_response || {};
       const stdout = toolResponse.stdout ?? payload.stdout ?? '';
-      const stderr = toolResponse.stderr ?? payload.stderr ?? '';
+      const stderr = toolResponse.stderr ?? payload.stderr ?? payload.error ?? '';
       const rawExitCode = toolResponse.exitCode ?? toolResponse.exit_code ?? payload.exitCode;
       const exitCode = typeof rawExitCode === 'number' ? rawExitCode : undefined;
       const stderrSignal = typeof stderr === 'string' && stderr.trim().length > 0;
-      const isFailed = (exitCode !== undefined && exitCode !== 0) || (exitCode === undefined && stderrSignal);
+      const hookEvent = payload.hook_event_name || payload.hookEventName || '';
+      const isFailed = hookEvent === 'PostToolUseFailure' ||
+        (exitCode !== undefined && exitCode !== 0) ||
+        (exitCode === undefined && stderrSignal);
       const toolName = payload.tool_name || payload.tool || 'command';
+      const observedMutation = observeWorkspaceMutation(payload, root);
 
       if (isFailed) {
         // Truncate output to avoid state bloat
@@ -78,9 +108,10 @@ function processState(payload) {
 
       // Milestones for the Stop gate: when did the last mutation happen, and
       // has any verification-ish command succeeded since.
-      if (toolName === 'Edit' || toolName === 'Write' || toolName === 'apply_patch') {
+      if (toolName === 'Edit' || toolName === 'Write' || toolName === 'apply_patch' || observedMutation) {
         state.lastEditAt = Date.now();
-      } else if ((toolName === 'Bash' || toolName === 'PowerShell') && !isFailed) {
+      }
+      if ((toolName === 'Bash' || toolName === 'PowerShell') && !isFailed) {
         const command = (payload.tool_input && payload.tool_input.command) || '';
         if (VERIFY_COMMAND_RE.test(command)) {
           state.lastVerifyAt = Date.now();
@@ -100,7 +131,11 @@ function processState(payload) {
       }
     }
   } catch (err) {
-    // Fail silently in hooks
+    if (err && (err.code === 'HARNESS_MUTATION_OBSERVATION' || err.code === 'HARNESS_WORKFLOW_BUDGET')) {
+      console.error('[State Persist] ' + err.message);
+      exitCodeForHook = 2;
+    }
+    // Non-critical persistence/telemetry failures remain fail-open.
   }
-  process.exit(0);
+  process.exit(exitCodeForHook);
 }

package/hooks/scripts/workflow-gate.js

@@ -3,8 +3,8 @@
 
 const path = require('path');
 const fs = require('fs');
-const { loadWorkflow, saveWorkflow, isMajorWorkflow, matchingRun, recordBudgetEvent, readHookInput, WORKFLOW_CONTROLLER_COMMANDS } = require('./lib/workflow-runtime');
-const { key, classifyShell, cwdOf, commandOf, mutationPaths, linkedWorktree, assertTargets, assertShellScope } = require('./lib/workflow-isolation');
+const { loadWorkflow, saveWorkflow, isMajorWorkflow, matchingRun, registerMutationProbe, recordBudgetEvent, readHookInput, WORKFLOW_CONTROLLER_COMMANDS } = require('./lib/workflow-runtime');
+const { key, classifyShell, workspaceFingerprint, shellProbeKey, cwdOf, commandOf, mutationPaths, linkedWorktree, assertTargets, assertShellScope } = require('./lib/workflow-isolation');
 
 const DIRECT = new Set(['Edit', 'Write', 'apply_patch']);
 const SHELL = new Set(['Bash', 'PowerShell', 'exec_command']);
@@ -27,8 +27,8 @@ function decide(payload) {
   const cwd = cwdOf(payload, root);
   const command = commandOf(payload);
   const commandClass = SHELL.has(tool) ? classifyShell(command) : 'direct-mutation';
-  if (commandClass === 'read-only' || commandClass === 'worktree-setup') return;
   if (SHELL.has(tool) && isController(command, cwd)) return;
+  if (commandClass === 'read-only' || commandClass === 'worktree-setup') return;
   // Only the stage specification is allowed before run entry. Other runtime
   // state, including workflow-run.json, cannot use this bootstrap exception.
   const targets = DIRECT.has(tool) ? mutationPaths(payload) : [];
@@ -47,11 +47,22 @@ function decide(payload) {
   if (String(workflow.strategy || '').startsWith('fable-') && !matchingRun(context)) {
     throw new Error('selected Fable workflow has no correlated run. Write workflow-stages.json at the displayed session path, then use workflow-disposition.js start.');
   }
-  if (workflow.strategy === 'iterative-single') {
-    recordBudgetEvent(context, 'iteration', { evidence: `${tool}:${commandClass}` });
+  if (DIRECT.has(tool)) {
+    if (workflow.strategy === 'iterative-single') {
+      recordBudgetEvent(context, 'iteration', { evidence: `${tool}:direct-mutation` });
+    }
+    workflow.lastMutationAt = Date.now();
+    saveWorkflow(context);
+    return;
   }
-  workflow.lastMutationAt = Date.now();
-  saveWorkflow(context);
+
+  // Shell safety and shell mutation accounting are intentionally separate.
+  // `untrusted` means "not proven read-only", not "a mutation occurred".
+  // Capture an opaque pre-execution workspace fingerprint; the post-tool
+  // observer consumes an iteration only when workspace content actually changed.
+  const fingerprint = workspaceFingerprint(cwd);
+  const probeKey = shellProbeKey(payload, cwd);
+  registerMutationProbe(context, probeKey, fingerprint, workflow.strategy === 'iterative-single');
 }
 
 readHookInput(decide);

package/hooks/scripts/workflow-mutation-denied.js

@@ -0,0 +1,17 @@
+#!/usr/bin/env node
+'use strict';
+
+const { loadWorkflow, discardMutationProbe, readHookInput } = require('./lib/workflow-runtime');
+const { cwdOf, shellProbeKey } = require('./lib/workflow-isolation');
+
+function discardDenied(payload) {
+  if (!payload) return;
+  const tool = payload.tool_name || payload.tool;
+  if (tool !== 'Bash' && tool !== 'PowerShell' && tool !== 'exec_command') return;
+  const context = loadWorkflow(payload);
+  if (!context.workflow || context.workflow.state === 'deferred') return;
+  const cwd = cwdOf(payload, context.root);
+  discardMutationProbe(context, shellProbeKey(payload, cwd));
+}
+
+readHookInput(discardDenied);

package/opencode-plugin/plugin.json

@@ -1,7 +1,7 @@
 {
   "name": "harness-enforcement",
   "description": "Repository inventory for the opencode enforcement plugin and its standalone helper",
-  "version": "0.11.5",
+  "version": "0.11.6",
   "runtime": "index.mjs",
   "hooks": {
     "verify": "hooks/verify.js"

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "harness-everything",
-  "version": "0.11.5",
+  "version": "0.11.6",
   "description": "Harness OS: An Orchestrated Agent Operating System for AI-driven Development",
   "main": "harness-everything/scripts/bootstrap.js",
   "bin": {

package/plugins/harness-everything/.codex-plugin/plugin.json

@@ -1,6 +1,6 @@
 {
   "name": "harness-everything",
-  "version": "0.11.5",
+  "version": "0.11.6",
   "description": "Invariant-first engineering harness for routing, verification, and failure recovery while preserving agent autonomy.",
   "author": {
     "name": "Miya Daniel",

package/plugins/harness-everything/hooks/hooks.json

@@ -142,7 +142,7 @@
             "command": "node \"${PLUGIN_ROOT}/hooks/scripts/state-persist.js\"",
             "commandWindows": "node \"${PLUGIN_ROOT}\\hooks\\scripts\\state-persist.js\"",
             "timeout": 15,
-            "statusMessage": "Saving Harness session state"
+            "statusMessage": "Saving Harness state and observed mutation"
           }
         ]
       },

package/plugins/harness-everything/hooks/scripts/lib/workflow-isolation.js

@@ -1,13 +1,89 @@
 'use strict';
 
 const fs = require('fs');
+const crypto = require('crypto');
 const path = require('path');
 const { spawnSync } = require('child_process');
 
-function git(cwd, args) {
-  const result = spawnSync('git', args, { cwd, encoding: 'utf8', windowsHide: true, timeout: 5000 });
+function gitRaw(cwd, args) {
+  const result = spawnSync('git', args, { cwd, encoding: 'utf8', windowsHide: true, timeout: 5000, maxBuffer: 16 * 1024 * 1024 });
   if (result.status !== 0) throw new Error('cannot verify Git worktree isolation');
-  return result.stdout.trim();
+  return result.stdout;
+}
+
+function git(cwd, args) {
+  return gitRaw(cwd, args).trim();
+}
+
+function hashWorktreePath(root, relative) {
+  const target = path.join(root, relative);
+  let stat;
+  try { stat = fs.lstatSync(target); }
+  catch (error) {
+    if (error.code === 'ENOENT') return null;
+    throw error;
+  }
+  if (stat.isDirectory()) {
+    const result = spawnSync('git', ['-C', target, 'rev-parse', 'HEAD'], { encoding: 'utf8', windowsHide: true, timeout: 5000 });
+    return { mode: '160000', oid: result.status === 0 ? result.stdout.trim() : 'directory' };
+  }
+  const mode = stat.isSymbolicLink() ? '120000' : ((stat.mode & 0o111) ? '100755' : '100644');
+  const result = spawnSync('git', ['hash-object', '--path=' + relative, '--', relative], {
+    cwd: root, encoding: 'utf8', windowsHide: true, timeout: 5000,
+  });
+  if (result.status !== 0) throw new Error('cannot fingerprint workspace content');
+  return { mode, oid: result.stdout.trim() };
+}
+
+function workspaceFingerprint(cwd) {
+  const root = canonical(git(cwd, ['rev-parse', '--show-toplevel']));
+  const entries = new Map();
+  for (const record of gitRaw(root, ['ls-files', '-s', '-z']).split('\0')) {
+    if (!record) continue;
+    const tab = record.indexOf('\t');
+    if (tab < 0) throw new Error('cannot fingerprint workspace index');
+    const meta = record.slice(0, tab).trim().split(/\s+/);
+    const relative = record.slice(tab + 1);
+    const mode = meta[0];
+    const oid = meta[1];
+    entries.set(relative, { relative, mode, oid });
+  }
+
+  // Substitute current worktree content for tracked paths that differ from the
+  // index. This keeps the fingerprint stable across git add / git commit when
+  // visible workspace content did not change.
+  for (const relative of gitRaw(root, ['diff-files', '--name-only', '-z']).split('\0').filter(Boolean)) {
+    const value = hashWorktreePath(root, relative);
+    if (value) entries.set(relative, { relative, mode: value.mode, oid: value.oid });
+    else entries.delete(relative);
+  }
+
+  for (const relative of gitRaw(root, ['ls-files', '--others', '--exclude-standard', '-z']).split('\0').filter(Boolean)) {
+    const value = hashWorktreePath(root, relative);
+    entries.set(relative, { relative, mode: value.mode, oid: value.oid });
+  }
+
+  const digest = crypto.createHash('sha256');
+  digest.update('harness-workspace-fingerprint-v1\0');
+  for (const entry of [...entries.values()].sort((a, b) => a.relative.localeCompare(b.relative))) {
+    digest.update(entry.relative); digest.update('\0');
+    digest.update(entry.mode); digest.update('\0');
+    digest.update(entry.oid); digest.update('\0');
+  }
+  return digest.digest('hex');
+}
+
+function shellProbeKey(payload, cwd) {
+  const toolUseId = String(payload?.tool_use_id || payload?.toolUseId || '').trim();
+  const tool = String(payload?.tool_name || payload?.tool || '');
+  const command = commandOf(payload);
+  return crypto.createHash('sha256')
+    .update('harness-shell-probe-v2\0')
+    .update(toolUseId || 'no-tool-use-id').update('\0')
+    .update(tool).update('\0')
+    .update(key(cwd)).update('\0')
+    .update(command)
+    .digest('hex');
 }
 
 function canonical(value) {
@@ -144,9 +220,9 @@ function splitShellSegments(text) {
 
 function classifyShell(command) {
   const text = String(command || '').trim();
-  if (!text) return 'mutation-or-unknown';
+  if (!text) return 'untrusted';
   const segments = splitShellSegments(text);
-  if (!segments) return 'mutation-or-unknown';
+  if (!segments) return 'untrusted';
   if (segments.length === 1) {
     if (/^git\s+worktree\s+add\s+/i.test(segments[0]) && !/\s--(?:force|checkout|detach)(?:\s|=|$)/i.test(segments[0])) return 'worktree-setup';
     return isReadOnlySegment(segments[0]) ? 'read-only' : 'mutation-or-unknown';
@@ -198,4 +274,4 @@ function assertShellScope(command, cwd, isolatedRoot) {
   }
 }
 
-module.exports = { canonical, key, within, linkedWorktree, classifyShell, inputOf, cwdOf, commandOf, mutationPaths, assertTargets, assertShellScope };
+module.exports = { canonical, key, within, linkedWorktree, classifyShell, workspaceFingerprint, shellProbeKey, inputOf, cwdOf, commandOf, mutationPaths, assertTargets, assertShellScope };

package/plugins/harness-everything/hooks/scripts/lib/workflow-runtime.js

@@ -43,6 +43,140 @@ function loadWorkflow(payload) {
 
 function saveWorkflow(context) { atomicWriteJson(context.file, context.workflow); }
 
+const MUTATION_PROBE_KEY = /^[a-f0-9]{64}$/;
+const MUTATION_PROBE_WAIT = new Int32Array(new SharedArrayBuffer(4));
+
+function mutationProbeDir(context) {
+  return path.join(context.sessionDir, 'mutation-probes');
+}
+
+function mutationProbeFile(context, key) {
+  if (!MUTATION_PROBE_KEY.test(String(key || ''))) throw new Error('invalid mutation probe identity');
+  return path.join(mutationProbeDir(context), key + '.json');
+}
+
+function withMutationProbeLock(context, callback) {
+  fs.mkdirSync(context.sessionDir, { recursive: true });
+  const lockDir = path.join(context.sessionDir, '.mutation-probes.lock');
+  let acquired = false;
+  for (let attempt = 0; attempt < 200; attempt++) {
+    try {
+      fs.mkdirSync(lockDir);
+      acquired = true;
+      break;
+    } catch (error) {
+      if (error.code !== 'EEXIST') throw error;
+      try {
+        const stat = fs.statSync(lockDir);
+        if (Date.now() - stat.mtimeMs > 30000) {
+          fs.rmSync(lockDir, { recursive: true, force: true });
+          continue;
+        }
+      } catch (_) { /* retry below */ }
+      Atomics.wait(MUTATION_PROBE_WAIT, 0, 0, 10);
+    }
+  }
+  if (!acquired) throw new Error('cannot acquire mutation probe lock');
+  try { return callback(); }
+  finally { fs.rmSync(lockDir, { recursive: true, force: true }); }
+}
+
+function mutationProbeReservations(context) {
+  const dir = mutationProbeDir(context);
+  let files = [];
+  try { files = fs.readdirSync(dir).filter(file => /^[a-f0-9]{64}\.json$/.test(file)); }
+  catch (_) { return 0; }
+  let count = 0;
+  for (const file of files) {
+    const probe = readJson(path.join(dir, file));
+    if (probe?.schemaVersion === 1 && probe.reserveIteration === true) count++;
+  }
+  return count;
+}
+
+function refreshWorkflowForProbe(context) {
+  const latest = readJson(context.file);
+  if (!latest || latest.sessionId !== context.sessionId) throw new Error('cannot refresh workflow for mutation probe');
+  context.workflow = latest;
+  return latest;
+}
+
+function assertIterationCapacity(context, evidence) {
+  const { workflow } = context;
+  if (!workflow) throw budgetError('no workflow available for iteration capacity');
+  const budget = ensureWorkflowBudget(workflow);
+  if (budget.state !== 'active') throw budgetError(`workflow budget is ${budget.state}: ${budget.reasonCode || 'unavailable'}`);
+  const limit = budget.limits.maxIterations;
+  if (limit === null) return { applicable: false, value: budget.counters.iterations, reserved: 0 };
+  const reserved = mutationProbeReservations(context);
+  if (budget.counters.iterations + reserved >= limit) {
+    return exhaustBudget(context, 'iteration-budget-exhausted', evidence || 'untrusted-shell-capacity');
+  }
+  return { applicable: true, value: budget.counters.iterations, reserved, limit };
+}
+
+function registerMutationProbe(context, key, fingerprint, reserveIteration) {
+  if (!MUTATION_PROBE_KEY.test(String(key || '')) || !MUTATION_PROBE_KEY.test(String(fingerprint || ''))) {
+    throw new Error('invalid mutation probe identity');
+  }
+  return withMutationProbeLock(context, () => {
+    refreshWorkflowForProbe(context);
+    if (reserveIteration) assertIterationCapacity(context, 'shell:untrusted');
+    const dir = mutationProbeDir(context);
+    fs.mkdirSync(dir, { recursive: true });
+    const file = mutationProbeFile(context, key);
+    if (fs.existsSync(file)) throw new Error('mutation probe is already pending for this tool call');
+    const probe = {
+      schemaVersion: 1,
+      fingerprint,
+      reserveIteration: Boolean(reserveIteration),
+      observedAt: Date.now(),
+    };
+    atomicWriteJson(file, probe);
+    return probe;
+  });
+}
+
+function peekMutationProbe(context, key) {
+  return readJson(mutationProbeFile(context, key));
+}
+
+function settleMutationProbe(context, key, afterFingerprint, tool) {
+  if (!MUTATION_PROBE_KEY.test(String(afterFingerprint || ''))) throw new Error('invalid mutation probe fingerprint');
+  return withMutationProbeLock(context, () => {
+    const file = mutationProbeFile(context, key);
+    const probe = readJson(file);
+    if (!probe || probe.schemaVersion !== 1) return null;
+    refreshWorkflowForProbe(context);
+    fs.unlinkSync(file);
+    const changed = afterFingerprint !== probe.fingerprint;
+    if (changed) {
+      context.workflow.lastMutationAt = Date.now();
+      if (probe.reserveIteration && context.workflow.strategy === 'iterative-single') {
+        recordBudgetEvent(context, 'iteration', { evidence: String(tool || 'shell') + ':observed-workspace-mutation' });
+      } else {
+        saveWorkflow(context);
+      }
+    }
+    return { changed, probe };
+  });
+}
+
+function discardMutationProbe(context, key) {
+  return withMutationProbeLock(context, () => {
+    const file = mutationProbeFile(context, key);
+    if (!fs.existsSync(file)) return false;
+    fs.unlinkSync(file);
+    return true;
+  });
+}
+
+function clearMutationProbes(context) {
+  return withMutationProbeLock(context, () => {
+    fs.rmSync(mutationProbeDir(context), { recursive: true, force: true });
+  });
+}
+
 function budgetLimits(workflow) {
   const limits = workflow?.workflowPlan?.limits || {};
   return {
@@ -123,7 +257,8 @@ function recordBudgetEvent(context, type, options = {}) {
   const consume = (counter, limitName, reasonCode) => {
     const limit = budget.limits[limitName];
     if (limit === null) return { applicable: false, value: budget.counters[counter] };
-    if (budget.counters[counter] >= limit) return exhaustBudget(context, reasonCode, evidence);
+    const reserved = type === 'iteration' ? mutationProbeReservations(context) : 0;
+    if (budget.counters[counter] + reserved >= limit) return exhaustBudget(context, reasonCode, evidence);
     budget.counters[counter]++;
     appendBudgetEvent(budget, { type, counter, value: budget.counters[counter], limit, evidence, observedAt: now });
     persistBudget(context);
@@ -175,6 +310,7 @@ function resetWorkflowBudget(context, evidence) {
     evidence: String(evidence).trim(),
     resetAt: now,
   }].slice(-16);
+  clearMutationProbes(context);
   workflow.budget = {
     schemaVersion: 1,
     epoch: previous.epoch + 1,
@@ -260,4 +396,4 @@ function readHookInput(decide) {
   process.stdin.on('error', finish);
 }
 
-module.exports = { OPEN_STATES, SAFE_ID, WORKFLOW_CONTROLLER_COMMANDS, isMajorWorkflow, loadWorkflow, saveWorkflow, budgetLimits, ensureWorkflowBudget, recordBudgetEvent, resetWorkflowBudget, syncBudgetToRun, matchingRun, unresolvedStages, readHookInput };
+module.exports = { OPEN_STATES, SAFE_ID, WORKFLOW_CONTROLLER_COMMANDS, isMajorWorkflow, loadWorkflow, saveWorkflow, budgetLimits, ensureWorkflowBudget, mutationProbeReservations, assertIterationCapacity, registerMutationProbe, peekMutationProbe, settleMutationProbe, discardMutationProbe, clearMutationProbes, recordBudgetEvent, resetWorkflowBudget, syncBudgetToRun, matchingRun, unresolvedStages, readHookInput };

package/plugins/harness-everything/hooks/scripts/state-persist.js

@@ -2,6 +2,8 @@
 const fs = require('fs');
 const path = require('path');
 const { getWorkspaceRoot, getSessionDir } = require('./lib/harness-state');
+const { loadWorkflow, saveWorkflow, peekMutationProbe, settleMutationProbe } = require('./lib/workflow-runtime');
+const { cwdOf, shellProbeKey, workspaceFingerprint } = require('./lib/workflow-isolation');
 const { observeTool } = require('./lib/telemetry');
 
 // Commands that count as "verification ran" for the Stop gate
@@ -28,7 +30,31 @@ process.stdin.on('end', () => {
   }
 });
 
+function observeWorkspaceMutation(payload, root) {
+  const toolName = payload && (payload.tool_name || payload.tool);
+  if (toolName !== 'Bash' && toolName !== 'PowerShell' && toolName !== 'exec_command') return false;
+  const context = loadWorkflow(payload);
+  if (!context.workflow || context.workflow.state === 'deferred') return false;
+  const cwd = cwdOf(payload, root);
+  const probeKey = shellProbeKey(payload, cwd);
+  if (!peekMutationProbe(context, probeKey)) return false;
+  let after;
+  try {
+    after = workspaceFingerprint(cwd);
+  } catch (error) {
+    context.workflow.state = 'blocked';
+    context.workflow.blockReason = 'mutation-observation-failed';
+    saveWorkflow(context);
+    const critical = new Error('cannot verify whether shell execution changed workspace content');
+    critical.code = 'HARNESS_MUTATION_OBSERVATION';
+    throw critical;
+  }
+  const result = settleMutationProbe(context, probeKey, after, toolName);
+  return Boolean(result && result.changed);
+}
+
 function processState(payload) {
+  let exitCodeForHook = 0;
   try {
     const sessionId = payload && (payload.session_id || payload.sessionId);
     const root = getWorkspaceRoot(payload);
@@ -48,12 +74,16 @@ function processState(payload) {
       // it's actually a number; fall back to a stderr signal otherwise.
       const toolResponse = payload.tool_response || {};
       const stdout = toolResponse.stdout ?? payload.stdout ?? '';
-      const stderr = toolResponse.stderr ?? payload.stderr ?? '';
+      const stderr = toolResponse.stderr ?? payload.stderr ?? payload.error ?? '';
       const rawExitCode = toolResponse.exitCode ?? toolResponse.exit_code ?? payload.exitCode;
       const exitCode = typeof rawExitCode === 'number' ? rawExitCode : undefined;
       const stderrSignal = typeof stderr === 'string' && stderr.trim().length > 0;
-      const isFailed = (exitCode !== undefined && exitCode !== 0) || (exitCode === undefined && stderrSignal);
+      const hookEvent = payload.hook_event_name || payload.hookEventName || '';
+      const isFailed = hookEvent === 'PostToolUseFailure' ||
+        (exitCode !== undefined && exitCode !== 0) ||
+        (exitCode === undefined && stderrSignal);
       const toolName = payload.tool_name || payload.tool || 'command';
+      const observedMutation = observeWorkspaceMutation(payload, root);
 
       if (isFailed) {
         // Truncate output to avoid state bloat
@@ -78,9 +108,10 @@ function processState(payload) {
 
       // Milestones for the Stop gate: when did the last mutation happen, and
       // has any verification-ish command succeeded since.
-      if (toolName === 'Edit' || toolName === 'Write' || toolName === 'apply_patch') {
+      if (toolName === 'Edit' || toolName === 'Write' || toolName === 'apply_patch' || observedMutation) {
         state.lastEditAt = Date.now();
-      } else if ((toolName === 'Bash' || toolName === 'PowerShell') && !isFailed) {
+      }
+      if ((toolName === 'Bash' || toolName === 'PowerShell') && !isFailed) {
         const command = (payload.tool_input && payload.tool_input.command) || '';
         if (VERIFY_COMMAND_RE.test(command)) {
           state.lastVerifyAt = Date.now();
@@ -100,7 +131,11 @@ function processState(payload) {
       }
     }
   } catch (err) {
-    // Fail silently in hooks
+    if (err && (err.code === 'HARNESS_MUTATION_OBSERVATION' || err.code === 'HARNESS_WORKFLOW_BUDGET')) {
+      console.error('[State Persist] ' + err.message);
+      exitCodeForHook = 2;
+    }
+    // Non-critical persistence/telemetry failures remain fail-open.
   }
-  process.exit(0);
+  process.exit(exitCodeForHook);
 }

package/plugins/harness-everything/hooks/scripts/workflow-gate.js

@@ -3,8 +3,8 @@
 
 const path = require('path');
 const fs = require('fs');
-const { loadWorkflow, saveWorkflow, isMajorWorkflow, matchingRun, recordBudgetEvent, readHookInput, WORKFLOW_CONTROLLER_COMMANDS } = require('./lib/workflow-runtime');
-const { key, classifyShell, cwdOf, commandOf, mutationPaths, linkedWorktree, assertTargets, assertShellScope } = require('./lib/workflow-isolation');
+const { loadWorkflow, saveWorkflow, isMajorWorkflow, matchingRun, registerMutationProbe, recordBudgetEvent, readHookInput, WORKFLOW_CONTROLLER_COMMANDS } = require('./lib/workflow-runtime');
+const { key, classifyShell, workspaceFingerprint, shellProbeKey, cwdOf, commandOf, mutationPaths, linkedWorktree, assertTargets, assertShellScope } = require('./lib/workflow-isolation');
 
 const DIRECT = new Set(['Edit', 'Write', 'apply_patch']);
 const SHELL = new Set(['Bash', 'PowerShell', 'exec_command']);
@@ -27,8 +27,8 @@ function decide(payload) {
   const cwd = cwdOf(payload, root);
   const command = commandOf(payload);
   const commandClass = SHELL.has(tool) ? classifyShell(command) : 'direct-mutation';
-  if (commandClass === 'read-only' || commandClass === 'worktree-setup') return;
   if (SHELL.has(tool) && isController(command, cwd)) return;
+  if (commandClass === 'read-only' || commandClass === 'worktree-setup') return;
   // Only the stage specification is allowed before run entry. Other runtime
   // state, including workflow-run.json, cannot use this bootstrap exception.
   const targets = DIRECT.has(tool) ? mutationPaths(payload) : [];
@@ -47,11 +47,22 @@ function decide(payload) {
   if (String(workflow.strategy || '').startsWith('fable-') && !matchingRun(context)) {
     throw new Error('selected Fable workflow has no correlated run. Write workflow-stages.json at the displayed session path, then use workflow-disposition.js start.');
   }
-  if (workflow.strategy === 'iterative-single') {
-    recordBudgetEvent(context, 'iteration', { evidence: `${tool}:${commandClass}` });
+  if (DIRECT.has(tool)) {
+    if (workflow.strategy === 'iterative-single') {
+      recordBudgetEvent(context, 'iteration', { evidence: `${tool}:direct-mutation` });
+    }
+    workflow.lastMutationAt = Date.now();
+    saveWorkflow(context);
+    return;
   }
-  workflow.lastMutationAt = Date.now();
-  saveWorkflow(context);
+
+  // Shell safety and shell mutation accounting are intentionally separate.
+  // `untrusted` means "not proven read-only", not "a mutation occurred".
+  // Capture an opaque pre-execution workspace fingerprint; the post-tool
+  // observer consumes an iteration only when workspace content actually changed.
+  const fingerprint = workspaceFingerprint(cwd);
+  const probeKey = shellProbeKey(payload, cwd);
+  registerMutationProbe(context, probeKey, fingerprint, workflow.strategy === 'iterative-single');
 }
 
 readHookInput(decide);

package/plugins/harness-everything/hooks/scripts/workflow-mutation-denied.js

@@ -0,0 +1,17 @@
+#!/usr/bin/env node
+'use strict';
+
+const { loadWorkflow, discardMutationProbe, readHookInput } = require('./lib/workflow-runtime');
+const { cwdOf, shellProbeKey } = require('./lib/workflow-isolation');
+
+function discardDenied(payload) {
+  if (!payload) return;
+  const tool = payload.tool_name || payload.tool;
+  if (tool !== 'Bash' && tool !== 'PowerShell' && tool !== 'exec_command') return;
+  const context = loadWorkflow(payload);
+  if (!context.workflow || context.workflow.state === 'deferred') return;
+  const cwd = cwdOf(payload, context.root);
+  discardMutationProbe(context, shellProbeKey(payload, cwd));
+}
+
+readHookInput(discardDenied);

package/plugins/harness-everything/plugin.json

@@ -1,7 +1,7 @@
 {
   "$schema": "https://agent-plugins.org/schemas/1.0.0/plugin.schema.json",
   "name": "harness-everything",
-  "version": "0.11.5",
+  "version": "0.11.6",
   "description": "Invariant-first engineering harness for routing, verification, and failure recovery while preserving agent autonomy.",
   "author": {
     "name": "Miya Daniel",