harness-everything 0.16.0 → 0.16.2

package/opencode-plugin/README.md

@@ -1,13 +1,13 @@
-# Harness Enforcement Plugin for opencode
+# Harness Guidance + Rule-of-3 Plugin for opencode
 
-Adds hard enforcement gates for Harness skills in opencode, addressing the
+Adds lifecycle guidance, verification reminders, and the Rule-of-3 zoom-out boundary in opencode, addressing the
 limitation that skills are otherwise advisory-only on that platform.
 
-> **Evidence boundary:** the plugin is implemented against opencode's real plugin API and covered by deterministic mechanism tests. Live plugin loading remains unverified beyond one scoped surface: retained evidence supports project-scope `.js` loading and edit/verification state on OpenCode 1.18.31 (macOS) — [live-host evidence](../benchmarks/results/live-host/opencode-2026-09-16/README.md). The final snapshot is post-reset (`hardLock: false`, `count: 1`), not a history of the enforcement sequence. Hard lock is only an interactive observation with no retained blocked-tool trace. Reflection was operator-seeded, then agent-rewritten; agent-controlled state deletion resets the breaker, so no durable hard enforcement or behavioral effectiveness is claimed. Global scope, npm-package installation, and other host versions remain unverified. See [`../docs/platform-capabilities.md`](../docs/platform-capabilities.md).
+> **Evidence boundary:** the plugin is implemented against opencode's real plugin API and covered by deterministic mechanism tests. Live plugin loading remains unverified beyond one scoped surface: retained evidence supports project-scope `.js` loading and edit/verification state on OpenCode 1.18.31 (macOS) — [live-host evidence](../benchmarks/results/live-host/opencode-2026-09-16/README.md). The retained snapshot predates #190 and is historical evidence only: its final snapshot was post-reset; the old hard-lock was only an interactive observation, with no retained blocked-tool trace. Current behavior has no permanent post-reflection hard lock; only the third matching verification failure pauses edits for reflection. Reflection was operator-seeded, then agent-rewritten, so no behavioral-effectiveness claim is made. Global scope, npm-package installation, and other host versions remain unverified. See [`../docs/platform-capabilities.md`](../docs/platform-capabilities.md).
 
 ## Problem
 
-In Claude Code, Harness hooks ENFORCE rules (hard gate). In opencode, skills
+Harness now keeps workflow/productivity rules advisory on both Claude Code and opencode. Explicit permission boundaries and the third-failure Rule-of-3 reflection are the intentional blocking exceptions. In opencode, skills
 only SUGGEST rules (soft guidance). This means:
 - Under pressure, agents can bypass skills
 - Verification can be skipped
@@ -24,7 +24,7 @@ see issue #37 for how that was found and fixed. **Naming caveat (issue
 #127):** opencode's discovery glob only matches `*.js`/`*.ts`, so the file
 must be copied to a `.js` name when installed - see Installation below.
 
-It implements three enforcement mechanisms across opencode's real hooks:
+It implements three lifecycle mechanisms across opencode's real hooks:
 
 The module follows opencode's V1 plugin contract directly: the named
 `HarnessEnforcement` export is an async factory receiving the opencode context
@@ -41,7 +41,7 @@ V1 `(input, output)` shape, including
 Fires after every `edit`, `write` or `apply_patch` tool call and marks
 verification as pending.
 
-### 2. Verification gate (`event`, on `session.idle`)
+### 2. Verification reminder (`event`, on `session.idle`)
 
 opencode has no "before complete" hook to block the way Claude Code's Stop
 hook does. `session.idle` - fired when the agent's turn ends - is the closest
@@ -52,7 +52,7 @@ and on failure calls `client.session.prompt()` to push a synthetic follow-up
 message into the session - forcing the agent to keep working instead of
 actually stopping.
 
-### 3. Circuit breaker (`tool.execute.before` + the verification gate above)
+### 3. Rule-of-3 zoom-out boundary (`tool.execute.before` + the verification observer above)
 
 Enforces Rule of 3 on repeated verification failures (same failing command +
 truncated error, not arbitrary tool failures - `tool.execute.after` has no
@@ -62,11 +62,8 @@ normalized success/failure field to key a signature on for tools in general):
 - the follow-up asks for a tokenized `zoom-out-report.md` artifact with the
   required reflection sections; the artifact is accepted through the real
   edit hook and records `lastReflection`
-- the same signature failing again after that reflection was recorded
-  hard-locks the breaker
-- once hard-locked, `tool.execute.before` throws on any `edit`/`write`/
-  `apply_patch` call, blocking further edits until the state file is cleared
-  or a new session starts
+- after a valid reflection, the signature count resets; another three matching failures trigger another zoom-out
+- `npm run harness:reset` remains an optional manual state clear; a repeated cycle does not create a permanent second-stage lock
 
 Idle handling is bounded: after a failed verification, the plugin records that
 the follow-up is pending and ignores repeated `session.idle` events until a new
@@ -143,7 +140,7 @@ their patch target names the report file.
 `ci/mechanism-2n-opencode-plugin.test.js` imports `index.mjs` and drives its
 exported hooks directly with a mock `client`/`event` context - the same shape
 opencode's plugin loader passes in - covering the edit → idle → follow-up →
-reflection-artifact → retry → hard-lock sequence, repeated-idle idempotency,
+reflection-artifact → retry → fresh three-failure cycle, repeated-idle idempotency,
 session isolation, reset behavior, legacy-state migration, corrupt-state
 fail-closed behavior, and patch-based reflection writes.
 

package/opencode-plugin/index.mjs

@@ -17,7 +17,7 @@
  *                       message via client.session.prompt() to force the agent
  *                       to keep going instead of truly stopping.)
  *   verify.js        -> inlined into the session.idle handler
- *   circuit-breaker.js -> tool.execute.before (throws to hard-block edit tools
+ *   circuit-breaker.js -> tool.execute.before (throws to pause edit tools for the zoom-out boundary
  *                       once locked) + failure tracking inside session.idle
  *   compliance.js    -> folded into the same state writes, no separate stage
  *
@@ -50,7 +50,7 @@ function getStateHome() {
 // Keys state per real workspace (`directory`, which opencode itself resolves
 // - never a cwd walk) instead of one flat dir shared by every project on the
 // machine. The pre-fix `~/.harness-state` had no such key at all: a circuit
-// breaker trip in one opencode project hard-locked every other one too
+// breaker trip in one opencode project affected every other one too
 // (issue #42 item #4).
 function getWorkspaceKey(directory) {
   let real = resolve(directory)
@@ -162,7 +162,6 @@ function defaultEditState() {
 function defaultBreakerState() {
   return {
     failures: {},
-    hardLock: false,
     lastReflection: null,
     lastReflectionSignature: null,
     reflectionPending: false,
@@ -177,7 +176,8 @@ function normalizeEditState(state) {
 }
 
 function normalizeBreakerState(state) {
-  return { ...defaultBreakerState(), ...(state || {}), failures: (state && state.failures) || {} }
+  const { hardLock: _retiredHardLock, ...rest } = state || {}
+  return { ...defaultBreakerState(), ...rest, failures: (state && state.failures) || {} }
 }
 
 function isRecord(value) {
@@ -193,7 +193,6 @@ function loadBreakerState(file) {
     if (
       !isRecord(state) ||
       (state.failures !== undefined && !isRecord(state.failures)) ||
-      (state.hardLock !== undefined && typeof state.hardLock !== "boolean") ||
       (state.reflectionPending !== undefined && typeof state.reflectionPending !== "boolean") ||
       !finiteOrNull(state.lastReflection) ||
       !finiteOrNull(state.reflectionRequestedAt) ||
@@ -209,7 +208,8 @@ function loadBreakerState(file) {
     ) throw new Error("invalid breaker state shape")
     return normalizeBreakerState(state)
   } catch {
-    throw new Error(`Harness circuit breaker state is corrupt: "${file}" cannot be read safely.`)
+    console.warn(`Harness: ignoring corrupt circuit-breaker state at "${file}" and starting fresh.`)
+    return defaultBreakerState()
   }
 }
 
@@ -246,6 +246,11 @@ function completeReflection(breakerFile, reportFile) {
   const completedAt = Math.max(Date.now(), (breaker.reflectionRequestedAt || 0) + 1)
   breaker.lastReflection = completedAt
   breaker.lastReflectionSignature = breaker.reflectionSignature
+  if (breaker.reflectionSignature && breaker.failures[breaker.reflectionSignature]) {
+    breaker.failures[breaker.reflectionSignature].count = 0
+    breaker.failures[breaker.reflectionSignature].firstSeen = completedAt
+    breaker.failures[breaker.reflectionSignature].lastSeen = completedAt
+  }
   breaker.reflectionPending = false
   breaker.reflectionRequestedAt = null
   breaker.reflectionToken = null
@@ -304,9 +309,8 @@ function runVerification(cwd) {
 
 /**
  * Records a verification failure against the circuit breaker and returns the
- * action to take: allow (with retries remaining), force_reflection (3rd
- * failure on this signature), or hard_lock (same signature repeats after a
- * reflection was recorded).
+ * action to take: allow (with retries remaining) or force_reflection on the
+ * third matching failure. Completing reflection resets that signature's count.
  */
 function tripBreaker(breakerFile, signature) {
   const breaker = loadBreakerState(breakerFile)
@@ -319,15 +323,6 @@ function tripBreaker(breakerFile, signature) {
   entry.lastSeen = Date.now()
 
   if (entry.count >= 3) {
-    if (
-      breaker.lastReflection &&
-      breaker.lastReflection > entry.firstSeen &&
-      breaker.lastReflectionSignature === signature
-    ) {
-      breaker.hardLock = true
-      saveJSON(breakerFile, breaker)
-      return { action: "hard_lock", count: entry.count }
-    }
     if (!breaker.reflectionPending) {
       breaker.reflectionPending = true
       breaker.reflectionRequestedAt = Date.now()
@@ -610,12 +605,6 @@ export const HarnessEnforcement = async ({ client, directory }) => {
       if (!EDIT_TOOLS.has(input.tool)) return
       const { breakerFile, reflectionFile } = pathsFor(input.sessionID)
       const breaker = loadBreakerState(breakerFile)
-      if (breaker.hardLock) {
-        throw new Error(
-          `Harness circuit breaker hard-locked after a repeat failure post-reflection. ` +
-          `Delete "${breakerFile}" or start a new session to reset.`,
-        )
-      }
       if (breaker.reflectionPending && !isReflectionWrite(input, output, reflectionFile)) {
         throw new Error(
           `Harness reflection is required before another code edit. ` +
@@ -715,12 +704,7 @@ export const HarnessEnforcement = async ({ client, directory }) => {
         saveJSON(editStateFile, state)
 
         let text
-        if (trip.action === "hard_lock") {
-          text =
-            `Harness: same verification failure ("${failing.command}") returned after a reflection was ` +
-            `already recorded. The circuit breaker is now hard-locked - edits are blocked until it is reset.`
-          recordCompliance(complianceFile, (c) => c.circuitBreakerTrips++)
-        } else if (trip.action === "force_reflection") {
+        if (trip.action === "force_reflection") {
           const { stateDir } = pathsFor(activeSession)
           text =
             `Harness: "${failing.command}" has now failed 3 times with the same error. Stop and reflect ` +

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": "0.16.0",
+  "version": "0.16.2",
   "runtime": "index.mjs",
   "hooks": {
     "verify": "hooks/verify.js"

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "harness-everything",
-  "version": "0.16.0",
+  "version": "0.16.2",
   "description": "Harness OS: An Orchestrated Agent Operating System for AI-driven Development",
   "main": "harness-everything/scripts/bootstrap.js",
   "bin": {
@@ -40,7 +40,8 @@
     "eval:behavioral": "node behavioral-evals/run.js run",
     "eval:behavioral:validate": "node behavioral-evals/run.js validate",
     "eval:behavioral:plugin": "node behavioral-evals/run-with-plugin.js run",
-    "eval:opencode:hardlock-live": "node behavioral-evals/opencode-hardlock-live.js run",
+    "eval:opencode:reflection-gate-live": "node behavioral-evals/opencode-reflection-gate-live.js run",
+    "eval:opencode:hardlock-live": "node behavioral-evals/opencode-reflection-gate-live.js run",
     "eval:paired": "node behavioral-evals/paired-benchmark.js run",
     "eval:paired:summarize": "node behavioral-evals/paired-benchmark.js summarize",
     "test:ab": "node ci/ab-test-harness.js validate",

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

@@ -1,6 +1,6 @@
 {
   "name": "harness-everything",
-  "version": "0.16.0",
+  "version": "0.16.2",
   "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

@@ -39,7 +39,7 @@
             "command": "node \"${PLUGIN_ROOT}/hooks/scripts/workflow-gate.js\"",
             "commandWindows": "node \"${PLUGIN_ROOT}\\hooks\\scripts\\workflow-gate.js\"",
             "timeout": 15,
-            "statusMessage": "Checking Harness workflow and worktree-isolation contract"
+            "statusMessage": "Reviewing Harness workflow/worktree guidance"
           }
         ]
       },
@@ -248,14 +248,14 @@
             "command": "node \"${PLUGIN_ROOT}/hooks/scripts/workflow-stop-gate.js\"",
             "commandWindows": "node \"${PLUGIN_ROOT}\\hooks\\scripts\\workflow-stop-gate.js\"",
             "timeout": 15,
-            "statusMessage": "Checking Harness workflow completion gate"
+            "statusMessage": "Reviewing verification reminder"
           },
           {
             "type": "command",
             "command": "node \"${PLUGIN_ROOT}/hooks/scripts/stop-gate.js\"",
             "commandWindows": "node \"${PLUGIN_ROOT}\\hooks\\scripts\\stop-gate.js\"",
             "timeout": 15,
-            "statusMessage": "Checking Harness verification gate"
+            "statusMessage": "Reviewing verification reminder"
           }
         ]
       },

package/plugins/harness-everything/hooks/scripts/atomic-commit-check.js

@@ -143,7 +143,7 @@ function main(payload) {
       const changedFiles = statusRaw.split('\n').length;
       console.error(`[Atomic Commit Check] ${state.editCount} edits since the last commit (${changedFiles} files currently changed).`);
       console.error('If a logically complete chunk of work is done, commit it now before continuing - large uncommitted diffs are harder to review and harder to recover from.');
-      process.exit(2);
+      process.exit(0);
     }
 
     process.exit(0);

package/plugins/harness-everything/hooks/scripts/boundary-guard.js

@@ -25,10 +25,10 @@ process.stdin.on('end', () => {
       if (filePath && !input.limit && fs.existsSync(filePath)) {
         const stat = fs.statSync(filePath);
         if (stat.isFile() && stat.size > MAX_BYTES_WITHOUT_LIMIT) {
-          console.error(`[Boundary Guard] BLOCKED: "${filePath}" is ${(stat.size / 1024).toFixed(0)}KB.`);
+          console.error(`[Boundary Reminder]: "${filePath}" is ${(stat.size / 1024).toFixed(0)}KB.`);
           console.error(`Reading a file this large in one call risks context bloat and "lost in the middle" degradation.`);
           console.error(`Re-issue the Read with an explicit "offset"/"limit" to pull a targeted slice, or use Grep to locate the relevant section first.`);
-          process.exit(2);
+          process.exit(0);
         }
       }
     }
@@ -41,10 +41,10 @@ process.stdin.on('end', () => {
         return asPathSegment.test(searchPath) || asPathSegment.test(pattern);
       });
       if (hitsNoiseDir) {
-        console.error(`[Boundary Guard] BLOCKED: search targets a noise directory (${NOISE_DIRS.join(', ')}).`);
+        console.error(`[Boundary Reminder]: search targets a noise directory (${NOISE_DIRS.join(', ')}).`);
         console.error(`These produce huge, low-signal result sets and rarely contain what you're actually looking for.`);
         console.error(`Scope the search to source directories instead, or state explicitly why this directory must be searched.`);
-        process.exit(2);
+        process.exit(0);
       }
     }
 

package/plugins/harness-everything/hooks/scripts/contract-test.js

@@ -22,7 +22,7 @@ const {
   listRunContracts,
   getWorkerId,
 } = require('./lib/fable-contracts');
-const { loadWorkflow, matchingRun, recordBudgetEvent } = require('./lib/workflow-runtime');
+const { loadWorkflow, matchingRun } = require('./lib/workflow-runtime');
 const { createLearningOpportunity } = require('./lib/learning-opportunity');
 
 function resultState(payload) {
@@ -119,7 +119,7 @@ process.stdin.on('end', () => {
       if (resolution.ambiguous || !resolution.match) {
         console.error(`[Contract Test] Ambiguous Fable check command matched ${runMatches.length} active run contract(s): ${command}`);
         console.error('No contract was mutated. Correlate the worker/session to one run before accepting this check.');
-        process.exit(2);
+        process.exit(0);
       }
 
       const entry = resolution.match;
@@ -127,14 +127,14 @@ process.stdin.on('end', () => {
         const worker = getWorkerId(payload);
         if (!worker || (entry.contract.workerId && worker !== entry.contract.workerId)) {
           console.error('[Contract Test] Active workflow check lacks a matching observed worker identity.');
-          process.exit(2);
+          process.exit(0);
         }
         for (const id of entry.contract.dependsOn || []) {
           const dependency = JSON.parse(fs.readFileSync(path.join(entry.runRoot, 'contracts', `${id}.json`), 'utf8'));
           const escaped = workflowContext.workflow.escapes?.some(item => item.stageId === id && item.runId === entry.contract.runId);
           if (dependency.status !== 'pass' && !escaped) {
             console.error('[Contract Test] Dependency has not passed: ' + id);
-            process.exit(2);
+            process.exit(0);
           }
         }
       }
@@ -164,12 +164,9 @@ process.stdin.on('end', () => {
       }
       console.log(`[Contract Test] ${contract.planId}/${contract.runId}/${contract.stageId}: ${contract.status.toUpperCase()} ${contract.checkCommand}`);
       if (contract.status === 'fail') {
-        if (workflowContext.workflow?.workflowId) {
-          recordBudgetEvent(workflowContext, 'revision', { evidence: `${contract.stageId}:${contract.evidence || 'verification-failed'}` });
-        }
         console.error(`Evidence: ${contract.evidence || '(no output captured)'}`);
-        console.error(`Do not build on this stage's output until it passes. Artifact: ${contract.outputPath || '(no path recorded)'}`);
-        process.exit(2);
+        console.error(`Review this failed stage before relying on its output. Artifact: ${contract.outputPath || '(no path recorded)'}`);
+        process.exit(0);
       }
       process.exit(0);
     }
@@ -179,12 +176,12 @@ process.stdin.on('end', () => {
     const legacy = resolveLegacyContract(stateRoot, command, payload);
     if (legacy && legacy.ambiguous) {
       console.error(`[Contract Test] Ambiguous legacy stage contracts matched: ${command}`);
-      process.exit(2);
+      process.exit(0);
     }
     if (legacy && legacy.failed) {
       console.error(`[Contract Test] Legacy stage "${legacy.contract.stageId}" FAILED its named check: ${legacy.contract.checkCommand}`);
       console.error(`Evidence: ${legacy.contract.evidence || '(no output captured)'}`);
-      process.exit(2);
+      process.exit(0);
     }
     process.exit(0);
   } catch (_) {

package/plugins/harness-everything/hooks/scripts/depth-guard.js

@@ -52,10 +52,10 @@ process.stdin.on('end', () => {
     });
 
     if (!wasRead) {
-      console.error(`[Depth Guard] BLOCKED: "${filePath}" already exists but hasn't been Read in this session.`);
+      console.error(`[Depth Reminder]: "${filePath}" already exists but hasn't been Read in this session.`);
       console.error(`Write() overwrites the whole file - doing that blind risks destroying content you never looked at.`);
       console.error(`Read the file first (even a quick pass), then retry the Write, or use Edit for a targeted change.`);
-      process.exit(2);
+      process.exit(0);
     }
 
     process.exit(0);