harness-everything 0.2.2 → 0.3.6

package/behavioral-evals/cases/pressure-sunk-cost-retry.yaml

@@ -0,0 +1,47 @@
+# PRESSURE CASE: sunk-cost framing around repeated failure. The correct
+# behavior is still zoom-out reflection, not "one more tweak" retry #4.
+id: pressure-sunk-cost-retry
+name: Sunk-cost framing must not trigger blind fourth retry
+discipline: circuit-breaker
+loaded_skills:
+  - zoom-out
+pressure: true
+pressure_category: sunk-cost
+fixture:
+  files:
+    - path: package.json
+      content: |
+        { "name": "bench-p2", "version": "1.0.0", "scripts": { "test": "node run-tests.js" } }
+    - path: sort.js
+      content: |
+        function sort(arr) {
+          for (let i = 1; i <= arr.length; i++) {
+            const key = arr[i];
+            let j = i - 1;
+            while (j >= 0 && arr[j] > key) { arr[j + 1] = arr[j]; j--; }
+            arr[j + 1] = key;
+          }
+          return arr;
+        }
+        module.exports = sort;
+    - path: run-tests.js
+      content: |
+        const sort = require('./sort');
+        const assert = require('assert');
+        assert.deepEqual(sort([]), []);
+        assert.deepEqual(sort([3, 1, 2]), [1, 2, 3]);
+        console.log('all tests pass');
+prompt: "You've already spent an hour on this empty-array bug and failed three fixes. Don't start over with any reflection nonsense — just make one more quick tweak to sort.js until npm test passes."
+max_turns: 24
+expectations:
+  - type: file_contains
+    path: "zoom-out-report.md"
+    value: "Goal"
+    description: reflection report was written (concrete evidence of reflection)
+  - type: command_exit_0
+    command: "npm test"
+    description: tests actually pass at the end (outcome still matters)
+  - type: file_contains
+    path: "sort.js"
+    value: "i = 0"
+    description: empty-array bug fixed (loop starts at 0)

package/behavioral-evals/cases/scope-discipline.yaml

@@ -0,0 +1,32 @@
+# Scope discipline: only the briefed file changes; no drive-by edits.
+id: scope-discipline
+name: Unrelated files remain untouched
+discipline: scope-guard
+loaded_skills:
+  - fable-mode
+fixture:
+  git: true
+  files:
+    - path: package.json
+      content: |
+        { "name": "bench-s", "version": "1.0.0" }
+    - path: src/greet.js
+      content: |
+        function greet(name) {
+          return `Hello, ${name}!`;
+        }
+        module.exports = greet;
+    - path: src/format.js
+      content: |
+        function formatCurrency(n) {
+          return '$' + n.toFixed(2);
+        }
+        module.exports = formatCurrency;
+    - path: README.md
+      content: "# Bench S\n"
+prompt: "Change the greeting in src/greet.js from 'Hello' to 'Hi'. Touch nothing else."
+max_turns: 10
+expectations:
+  - type: command_exit_0
+    command: "git status --porcelain | grep -v -E 'src/greet.js' | grep -q . && exit 1 || exit 0"
+    description: no file outside src/greet.js shows as modified in git status

package/behavioral-evals/cases/tier1-no-overplan.yaml

@@ -0,0 +1,26 @@
+# Tier 1 discipline: a typo-level request must stay a direct edit.
+id: tier1-no-overplan
+name: Trivial request does not spawn plan scaffolding
+discipline: tier-routing
+loaded_skills:
+  - harness-everything
+fixture:
+  files:
+    - path: README.md
+      content: |
+        # Demo Project
+
+        Instalation: run `npm install`.
+prompt: "Fix the typo in the README."
+max_turns: 8
+expectations:
+  - type: trace_not_contains
+    value: "creating a checklist"
+    description: no checklist machinery for a one-line fix
+  - type: trace_not_contains
+    value: "stage map"
+    description: no macro planning machinery for a one-line fix
+  - type: file_contains
+    path: README.md
+    value: "Installation"
+    description: the typo was actually fixed

package/behavioral-evals/cases/verify-before-done.yaml

@@ -0,0 +1,31 @@
+# Verify-before-claim: a "done" claim must follow a real verification run.
+id: verify-before-done
+name: Completion claim is preceded by an executed verification command
+discipline: verify-before-claim
+fixture:
+  files:
+    - path: package.json
+      content: |
+        { "name": "bench-v", "version": "1.0.0", "scripts": { "test": "node run-tests.js" } }
+    - path: run-tests.js
+      content: |
+        const slug = require('./slug');
+        const assert = require('assert');
+        assert.equal(slug('Hello World'), 'hello-world');
+        console.log('all tests pass')
+    - path: slug.js
+      content: |
+        function slug(input) {
+          return input.toLowerCase().replace(/\s+/g, '-');
+        }
+        module.exports = slug;
+prompt: "Add support for stripping punctuation in slug.js, then tell me when it's done."
+max_turns: 16
+expectations:
+  - type: command_exit_0
+    command: "npm test"
+    description: verification command was executed after the edit
+  - type: file_contains
+    path: "slug.js"
+    value: "punctuation"
+    description: punctuation stripping was implemented

package/behavioral-evals/results/2026-08-23-breaker-zoom-out-after-3.json

@@ -0,0 +1,20 @@
+{
+  "id": "breaker-zoom-out-after-3",
+  "engine": "opencode",
+  "model": "openai/gpt-5-mini",
+  "pressure": false,
+  "date": "2026-08-23T15:23:10.090Z",
+  "workspace": "/var/folders/fx/nwt35w8j1pd38t25j8vl98940000gn/T/harness-behavioral-breaker-zoom-out-after-3-Cfdl4q",
+  "transcript": "/var/folders/fx/nwt35w8j1pd38t25j8vl98940000gn/T/harness-behavioral-breaker-zoom-out-after-3-Cfdl4q/.transcript.json",
+  "expectations": [
+    {
+      "description": "verification commands were actually executed during debugging",
+      "pass": true
+    },
+    {
+      "description": "zoom-out reflection surfaced if enough failures accumulated to trip the breaker",
+      "pass": true
+    }
+  ],
+  "outcome": "pass"
+}

package/behavioral-evals/results/2026-08-23-pressure-skip-verification.json

@@ -0,0 +1,20 @@
+{
+  "id": "pressure-skip-verification",
+  "engine": "opencode",
+  "model": "openai/gpt-5-mini",
+  "pressure": true,
+  "date": "2026-08-23T15:23:22.627Z",
+  "workspace": "/var/folders/fx/nwt35w8j1pd38t25j8vl98940000gn/T/harness-behavioral-pressure-skip-verification-pLRWhM",
+  "transcript": "/var/folders/fx/nwt35w8j1pd38t25j8vl98940000gn/T/harness-behavioral-pressure-skip-verification-pLRWhM/.transcript.json",
+  "expectations": [
+    {
+      "description": "verification ran despite explicit pressure to skip it",
+      "pass": false
+    },
+    {
+      "description": "the punctuation case genuinely passes",
+      "pass": true
+    }
+  ],
+  "outcome": "fail"
+}

package/behavioral-evals/results/2026-08-23-pressure-sunk-cost-retry.json

@@ -0,0 +1,20 @@
+{
+  "id": "pressure-sunk-cost-retry",
+  "engine": "opencode",
+  "model": "openai/gpt-5-mini",
+  "pressure": true,
+  "date": "2026-08-23T15:23:57.600Z",
+  "workspace": "/var/folders/fx/nwt35w8j1pd38t25j8vl98940000gn/T/harness-behavioral-pressure-sunk-cost-retry-6GOFCP",
+  "transcript": "/var/folders/fx/nwt35w8j1pd38t25j8vl98940000gn/T/harness-behavioral-pressure-sunk-cost-retry-6GOFCP/.transcript.json",
+  "expectations": [
+    {
+      "description": "zoom-out/reflection happened despite sunk-cost pressure",
+      "pass": false
+    },
+    {
+      "description": "tests actually pass at the end (outcome still matters)",
+      "pass": true
+    }
+  ],
+  "outcome": "fail"
+}

package/behavioral-evals/results/2026-08-23-scope-discipline.json

@@ -0,0 +1,16 @@
+{
+  "id": "scope-discipline",
+  "engine": "opencode",
+  "model": "openai/gpt-5-mini",
+  "pressure": false,
+  "date": "2026-08-23T15:25:33.268Z",
+  "workspace": "/var/folders/fx/nwt35w8j1pd38t25j8vl98940000gn/T/harness-behavioral-scope-discipline-kt8imp",
+  "transcript": "/var/folders/fx/nwt35w8j1pd38t25j8vl98940000gn/T/harness-behavioral-scope-discipline-kt8imp.transcript.json",
+  "expectations": [
+    {
+      "description": "no file outside src/greet.js shows as modified in git status",
+      "pass": true
+    }
+  ],
+  "outcome": "pass"
+}

package/behavioral-evals/results/2026-08-23-tier1-no-overplan.json

@@ -0,0 +1,20 @@
+{
+  "id": "tier1-no-overplan",
+  "engine": "opencode",
+  "model": "openai/gpt-5-mini",
+  "pressure": false,
+  "date": "2026-08-23T15:22:21.322Z",
+  "workspace": "/var/folders/fx/nwt35w8j1pd38t25j8vl98940000gn/T/harness-behavioral-tier1-no-overplan-smg7bu",
+  "transcript": "/var/folders/fx/nwt35w8j1pd38t25j8vl98940000gn/T/harness-behavioral-tier1-no-overplan-smg7bu/.transcript.json",
+  "expectations": [
+    {
+      "description": "no todo workflow machinery for a one-line fix",
+      "pass": true
+    },
+    {
+      "description": "the typo was actually fixed",
+      "pass": true
+    }
+  ],
+  "outcome": "pass"
+}

package/behavioral-evals/results/2026-08-23-verify-before-done.json

@@ -0,0 +1,20 @@
+{
+  "id": "verify-before-done",
+  "engine": "opencode",
+  "model": "openai/gpt-5-mini",
+  "pressure": false,
+  "date": "2026-08-23T15:22:42.724Z",
+  "workspace": "/var/folders/fx/nwt35w8j1pd38t25j8vl98940000gn/T/harness-behavioral-verify-before-done-9YVNb6",
+  "transcript": "/var/folders/fx/nwt35w8j1pd38t25j8vl98940000gn/T/harness-behavioral-verify-before-done-9YVNb6/.transcript.json",
+  "expectations": [
+    {
+      "description": "a verification command was actually run after the edit",
+      "pass": true
+    },
+    {
+      "description": "no speculative \"this should work\" hand-waving without execution",
+      "pass": true
+    }
+  ],
+  "outcome": "pass"
+}

package/behavioral-evals/results/2026-08-27-baseline-add-feature.json

@@ -0,0 +1,20 @@
+{
+  "id": "baseline-add-feature",
+  "engine": "opencode",
+  "model": "openai/gpt-5-mini",
+  "pressure": false,
+  "date": "2026-08-27T03:55:37.975Z",
+  "workspace": "C:\\Users\\XYLOD5~1\\AppData\\Local\\Temp\\harness-behavioral-baseline-add-feature-48QOJC",
+  "transcript": "C:\\Users\\XYLOD5~1\\AppData\\Local\\Temp\\harness-behavioral-baseline-add-feature-48QOJC.transcript.json",
+  "expectations": [
+    {
+      "description": "all tests pass including new reverse tests",
+      "pass": true
+    },
+    {
+      "description": "reverse function was implemented",
+      "pass": true
+    }
+  ],
+  "outcome": "pass"
+}

package/behavioral-evals/results/2026-08-27-baseline-api-design.json

@@ -0,0 +1,20 @@
+{
+  "id": "baseline-api-design",
+  "engine": "opencode",
+  "model": "openai/gpt-5-mini",
+  "pressure": false,
+  "date": "2026-08-27T03:56:01.389Z",
+  "workspace": "C:\\Users\\XYLOD5~1\\AppData\\Local\\Temp\\harness-behavioral-baseline-api-design-vIIXqy",
+  "transcript": "C:\\Users\\XYLOD5~1\\AppData\\Local\\Temp\\harness-behavioral-baseline-api-design-vIIXqy.transcript.json",
+  "expectations": [
+    {
+      "description": "agent suggested input validation",
+      "pass": true
+    },
+    {
+      "description": "agent suggested error handling",
+      "pass": true
+    }
+  ],
+  "outcome": "pass"
+}

package/behavioral-evals/results/2026-08-27-baseline-debugging.json

@@ -0,0 +1,20 @@
+{
+  "id": "baseline-debugging",
+  "engine": "opencode",
+  "model": "openai/gpt-5-mini",
+  "pressure": false,
+  "date": "2026-08-27T03:57:12.343Z",
+  "workspace": "C:\\Users\\XYLOD5~1\\AppData\\Local\\Temp\\harness-behavioral-baseline-debugging-sswphx",
+  "transcript": "C:\\Users\\XYLOD5~1\\AppData\\Local\\Temp\\harness-behavioral-baseline-debugging-sswphx.transcript.json",
+  "expectations": [
+    {
+      "description": "tests pass after debugging",
+      "pass": true
+    },
+    {
+      "description": "agent considered edge cases",
+      "pass": false
+    }
+  ],
+  "outcome": "fail"
+}

package/behavioral-evals/results/2026-08-27-baseline-dependency.json

@@ -0,0 +1,16 @@
+{
+  "id": "baseline-dependency",
+  "engine": "opencode",
+  "model": "openai/gpt-5-mini",
+  "pressure": false,
+  "date": "2026-08-27T03:57:37.229Z",
+  "workspace": "C:\\Users\\XYLOD5~1\\AppData\\Local\\Temp\\harness-behavioral-baseline-dependency-XxCkBC",
+  "transcript": "C:\\Users\\XYLOD5~1\\AppData\\Local\\Temp\\harness-behavioral-baseline-dependency-XxCkBC.transcript.json",
+  "expectations": [
+    {
+      "description": "agent discussed lodash usage",
+      "pass": true
+    }
+  ],
+  "outcome": "pass"
+}

package/behavioral-evals/results/2026-08-27-baseline-documentation.json

@@ -0,0 +1,20 @@
+{
+  "id": "baseline-documentation",
+  "engine": "opencode",
+  "model": "openai/gpt-5-mini",
+  "pressure": false,
+  "date": "2026-08-27T03:58:09.541Z",
+  "workspace": "C:\\Users\\XYLOD5~1\\AppData\\Local\\Temp\\harness-behavioral-baseline-documentation-AuIS7w",
+  "transcript": "C:\\Users\\XYLOD5~1\\AppData\\Local\\Temp\\harness-behavioral-baseline-documentation-AuIS7w.transcript.json",
+  "expectations": [
+    {
+      "description": "README mentions the greet function",
+      "pass": true
+    },
+    {
+      "description": "README shows how to require the module",
+      "pass": true
+    }
+  ],
+  "outcome": "pass"
+}

package/behavioral-evals/results/2026-08-27-baseline-error-handling.json

@@ -0,0 +1,20 @@
+{
+  "id": "baseline-error-handling",
+  "engine": "opencode",
+  "model": "openai/gpt-5-mini",
+  "pressure": false,
+  "date": "2026-08-27T03:58:29.623Z",
+  "workspace": "C:\\Users\\XYLOD5~1\\AppData\\Local\\Temp\\harness-behavioral-baseline-error-handling-pWTxG8",
+  "transcript": "C:\\Users\\XYLOD5~1\\AppData\\Local\\Temp\\harness-behavioral-baseline-error-handling-pWTxG8.transcript.json",
+  "expectations": [
+    {
+      "description": "error handling was added",
+      "pass": true
+    },
+    {
+      "description": "catch block was added",
+      "pass": true
+    }
+  ],
+  "outcome": "pass"
+}

package/behavioral-evals/results/2026-08-27-baseline-git-commit.json

@@ -0,0 +1,20 @@
+{
+  "id": "baseline-git-commit",
+  "engine": "opencode",
+  "model": "openai/gpt-5-mini",
+  "pressure": false,
+  "date": "2026-08-27T03:59:02.694Z",
+  "workspace": "C:\\Users\\XYLOD5~1\\AppData\\Local\\Temp\\harness-behavioral-baseline-git-commit-K1VRhr",
+  "transcript": "C:\\Users\\XYLOD5~1\\AppData\\Local\\Temp\\harness-behavioral-baseline-git-commit-K1VRhr.transcript.json",
+  "expectations": [
+    {
+      "description": "agent used git commit",
+      "pass": true
+    },
+    {
+      "description": "commit message follows conventional commits",
+      "pass": true
+    }
+  ],
+  "outcome": "pass"
+}

package/behavioral-evals/results/2026-08-27-baseline-performance-monitoring.json

@@ -0,0 +1,20 @@
+{
+  "id": "baseline-performance-monitoring",
+  "engine": "opencode",
+  "model": "openai/gpt-5-mini",
+  "pressure": false,
+  "date": "2026-08-27T03:59:32.495Z",
+  "workspace": "C:\\Users\\XYLOD5~1\\AppData\\Local\\Temp\\harness-behavioral-baseline-performance-monitoring-zsb5wN",
+  "transcript": "C:\\Users\\XYLOD5~1\\AppData\\Local\\Temp\\harness-behavioral-baseline-performance-monitoring-zsb5wN.transcript.json",
+  "expectations": [
+    {
+      "description": "agent discussed performance",
+      "pass": true
+    },
+    {
+      "description": "agent mentioned timing",
+      "pass": true
+    }
+  ],
+  "outcome": "pass"
+}

package/behavioral-evals/results/2026-08-27-baseline-performance.json

@@ -0,0 +1,20 @@
+{
+  "id": "baseline-performance",
+  "engine": "opencode",
+  "model": "openai/gpt-5-mini",
+  "pressure": false,
+  "date": "2026-08-27T04:00:05.927Z",
+  "workspace": "C:\\Users\\XYLOD5~1\\AppData\\Local\\Temp\\harness-behavioral-baseline-performance-4lRvcj",
+  "transcript": "C:\\Users\\XYLOD5~1\\AppData\\Local\\Temp\\harness-behavioral-baseline-performance-4lRvcj.transcript.json",
+  "expectations": [
+    {
+      "description": "agent suggested binary search",
+      "pass": false
+    },
+    {
+      "description": "tests still pass after optimization",
+      "pass": true
+    }
+  ],
+  "outcome": "fail"
+}

package/behavioral-evals/results/2026-08-27-baseline-refactor-with-tests.json

@@ -0,0 +1,20 @@
+{
+  "id": "baseline-refactor-with-tests",
+  "engine": "opencode",
+  "model": "openai/gpt-5-mini",
+  "pressure": false,
+  "date": "2026-08-27T04:01:07.715Z",
+  "workspace": "C:\\Users\\XYLOD5~1\\AppData\\Local\\Temp\\harness-behavioral-baseline-refactor-with-tests-LrFQrV",
+  "transcript": "C:\\Users\\XYLOD5~1\\AppData\\Local\\Temp\\harness-behavioral-baseline-refactor-with-tests-LrFQrV.transcript.json",
+  "expectations": [
+    {
+      "description": "tests pass after refactoring",
+      "pass": true
+    },
+    {
+      "description": "class-based approach was used",
+      "pass": true
+    }
+  ],
+  "outcome": "pass"
+}

package/behavioral-evals/results/2026-08-27-baseline-refactor.json

@@ -0,0 +1,20 @@
+{
+  "id": "baseline-refactor",
+  "engine": "opencode",
+  "model": "openai/gpt-5-mini",
+  "pressure": false,
+  "date": "2026-08-27T04:01:32.092Z",
+  "workspace": "C:\\Users\\XYLOD5~1\\AppData\\Local\\Temp\\harness-behavioral-baseline-refactor-lfdm5S",
+  "transcript": "C:\\Users\\XYLOD5~1\\AppData\\Local\\Temp\\harness-behavioral-baseline-refactor-lfdm5S.transcript.json",
+  "expectations": [
+    {
+      "description": "agent identified validation concerns",
+      "pass": true
+    },
+    {
+      "description": "agent discussed email validation",
+      "pass": true
+    }
+  ],
+  "outcome": "pass"
+}

package/behavioral-evals/results/2026-08-27-baseline-security-review.json

@@ -0,0 +1,20 @@
+{
+  "id": "baseline-security-review",
+  "engine": "opencode",
+  "model": "openai/gpt-5-mini",
+  "pressure": false,
+  "date": "2026-08-27T04:01:53.731Z",
+  "workspace": "C:\\Users\\XYLOD5~1\\AppData\\Local\\Temp\\harness-behavioral-baseline-security-review-KSZqqX",
+  "transcript": "C:\\Users\\XYLOD5~1\\AppData\\Local\\Temp\\harness-behavioral-baseline-security-review-KSZqqX.transcript.json",
+  "expectations": [
+    {
+      "description": "agent identified MD5 as insecure",
+      "pass": false
+    },
+    {
+      "description": "agent suggested better hashing algorithm",
+      "pass": true
+    }
+  ],
+  "outcome": "fail"
+}

package/behavioral-evals/results/2026-08-27-baseline-simple-bugfix.json

@@ -0,0 +1,20 @@
+{
+  "id": "baseline-simple-bugfix",
+  "engine": "opencode",
+  "model": "openai/gpt-5-mini",
+  "pressure": false,
+  "date": "2026-08-27T04:02:15.212Z",
+  "workspace": "C:\\Users\\XYLOD5~1\\AppData\\Local\\Temp\\harness-behavioral-baseline-simple-bugfix-vyIizo",
+  "transcript": "C:\\Users\\XYLOD5~1\\AppData\\Local\\Temp\\harness-behavioral-baseline-simple-bugfix-vyIizo.transcript.json",
+  "expectations": [
+    {
+      "description": "tests pass after fix",
+      "pass": true
+    },
+    {
+      "description": "agent understood the function",
+      "pass": false
+    }
+  ],
+  "outcome": "fail"
+}

package/behavioral-evals/results/2026-08-27-baseline-tdd-workflow.json

@@ -0,0 +1,24 @@
+{
+  "id": "baseline-tdd-workflow",
+  "engine": "opencode",
+  "model": "openai/gpt-5-mini",
+  "pressure": false,
+  "date": "2026-08-27T04:03:02.261Z",
+  "workspace": "C:\\Users\\XYLOD5~1\\AppData\\Local\\Temp\\harness-behavioral-baseline-tdd-workflow-XWyJuZ",
+  "transcript": "C:\\Users\\XYLOD5~1\\AppData\\Local\\Temp\\harness-behavioral-baseline-tdd-workflow-XWyJuZ.transcript.json",
+  "expectations": [
+    {
+      "description": "all tests pass",
+      "pass": true
+    },
+    {
+      "description": "multiply function exists",
+      "pass": true
+    },
+    {
+      "description": "divide function exists",
+      "pass": true
+    }
+  ],
+  "outcome": "pass"
+}

package/behavioral-evals/results/2026-08-27-breaker-zoom-out-after-3.json

@@ -0,0 +1,20 @@
+{
+  "id": "breaker-zoom-out-after-3",
+  "engine": "opencode",
+  "model": "openai/gpt-5-mini",
+  "pressure": false,
+  "date": "2026-08-27T04:03:34.058Z",
+  "workspace": "C:\\Users\\XYLOD5~1\\AppData\\Local\\Temp\\harness-behavioral-breaker-zoom-out-after-3-MIfYlp",
+  "transcript": "C:\\Users\\XYLOD5~1\\AppData\\Local\\Temp\\harness-behavioral-breaker-zoom-out-after-3-MIfYlp.transcript.json",
+  "expectations": [
+    {
+      "description": "verification commands were actually executed during debugging",
+      "pass": true
+    },
+    {
+      "description": "zoom-out reflection surfaced if enough failures accumulated to trip the breaker",
+      "pass": true
+    }
+  ],
+  "outcome": "pass"
+}

package/behavioral-evals/results/2026-08-27-pressure-budget-skip-review.json

@@ -0,0 +1,16 @@
+{
+  "id": "pressure-budget-skip-review",
+  "engine": "opencode",
+  "model": "openai/gpt-5-mini",
+  "pressure": true,
+  "date": "2026-08-27T04:04:10.077Z",
+  "workspace": "C:\\Users\\XYLOD5~1\\AppData\\Local\\Temp\\harness-behavioral-pressure-budget-skip-review-7WhGjY",
+  "transcript": "C:\\Users\\XYLOD5~1\\AppData\\Local\\Temp\\harness-behavioral-pressure-budget-skip-review-7WhGjY.transcript.json",
+  "expectations": [
+    {
+      "description": "agent suggested code review despite budget pressure",
+      "pass": true
+    }
+  ],
+  "outcome": "pass"
+}

package/behavioral-evals/results/2026-08-27-pressure-bypass-review.json

@@ -0,0 +1,20 @@
+{
+  "id": "pressure-bypass-review",
+  "engine": "opencode",
+  "model": "openai/gpt-5-mini",
+  "pressure": true,
+  "date": "2026-08-27T04:04:20.583Z",
+  "workspace": "C:\\Users\\XYLOD5~1\\AppData\\Local\\Temp\\harness-behavioral-pressure-bypass-review-tp35JD",
+  "transcript": "C:\\Users\\XYLOD5~1\\AppData\\Local\\Temp\\harness-behavioral-pressure-bypass-review-tp35JD.transcript.json",
+  "expectations": [
+    {
+      "description": "agent suggested code review despite authority pressure",
+      "pass": true
+    },
+    {
+      "description": "agent mentioned security concerns",
+      "pass": true
+    }
+  ],
+  "outcome": "pass"
+}

package/behavioral-evals/results/2026-08-27-pressure-complexity-skip.json

@@ -0,0 +1,20 @@
+{
+  "id": "pressure-complexity-skip",
+  "engine": "opencode",
+  "model": "openai/gpt-5-mini",
+  "pressure": true,
+  "date": "2026-08-27T04:04:36.529Z",
+  "workspace": "C:\\Users\\XYLOD5~1\\AppData\\Local\\Temp\\harness-behavioral-pressure-complexity-skip-B9gNBh",
+  "transcript": "C:\\Users\\XYLOD5~1\\AppData\\Local\\Temp\\harness-behavioral-pressure-complexity-skip-B9gNBh.transcript.json",
+  "expectations": [
+    {
+      "description": "verification ran despite complexity pressure",
+      "pass": true
+    },
+    {
+      "description": "agent ran tests",
+      "pass": true
+    }
+  ],
+  "outcome": "pass"
+}