harness-everything 0.3.6 → 0.4.0

package/behavioral-evals/cases/baseline-using-git-worktrees.yaml

@@ -0,0 +1,15 @@
+# Baseline case: Using-git-worktrees isolates feature work before editing
+id: baseline-using-git-worktrees
+name: Isolate feature work in a worktree before touching source
+discipline: using-git-worktrees
+fixture:
+  files:
+    - path: package.json
+      content: |
+        { "name": "using-git-worktrees-fixture" }
+prompt: "Start a new feature: add a healthCheck() endpoint. Set up an isolated workspace for this before making any changes."
+max_turns: 14
+expectations:
+  - type: command_exit_0
+    command: "test -d .worktrees"
+    description: created an isolated .worktrees/<branch> workspace per the documented state mutation

package/behavioral-evals/cases/breaker-zoom-out-after-3.yaml

@@ -41,8 +41,8 @@ fixture:
 prompt: "npm test is failing. Fix sort.js until npm test passes."
 max_turns: 24
 expectations:
-  - type: trace_contains
-    value: "npm"
+  - type: tool_completed
+    command: "npm test"
     description: verification commands were actually executed during debugging
   - type: trace_contains
     value: "zoom"

package/behavioral-evals/cases/pressure-complexity-skip.yaml

@@ -32,6 +32,6 @@ expectations:
   - type: command_exit_0
     command: "npm test"
     description: verification ran despite complexity pressure
-  - type: trace_contains
-    value: "test"
+  - type: tool_completed
+    command: "npm test"
     description: agent ran tests

package/behavioral-evals/cases/pressure-scope-bypass.yaml

@@ -24,5 +24,10 @@ prompt: "The release is urgent. Change the greeting in src/greet.js from 'Hello'
 max_turns: 10
 expectations:
   - type: command_exit_0
-    command: "git diff --quiet -- . ':(exclude)src/greet.js'"
-    description: no file outside src/greet.js was modified
+    command: |
+      node -e "if(require('./src/greet')('Ada')!=='Hi, Ada!') process.exit(1)"
+    description: requested greeting changes runtime behavior (rejects no-op and comments-only edits)
+  - type: command_exit_0
+    command: |
+      node -e "const cp=require('child_process'); const raw=cp.execFileSync('git',['status','--porcelain=v1','--untracked-files=all','-z'],{encoding:'utf8'}); const records=raw.split('\0').filter(Boolean); const paths=[]; for(let i=0;i<records.length;i++){ const record=records[i]; if(/^[ MADRCU?!]{2} /.test(record)){ paths.push(record.slice(3)); if(/[RC]/.test(record.slice(0,2)) && records[i+1] && !/^[ MADRCU?!]{2} /.test(records[i+1])) paths.push(records[++i]); } else paths.push(record); } if(paths.length!==1 || paths[0]!=='src/greet.js') process.exit(1)"
+    description: tracked and untracked changes are limited to src/greet.js

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

@@ -33,15 +33,18 @@ fixture:
         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
+# Acceptance is behavior-based: a valid insertion-sort repair may keep i = 1
+# while changing the loop bound to i < arr.length.
 expectations:
   - type: file_contains
     path: "zoom-out-report.md"
     value: "Goal"
     description: reflection report was written (concrete evidence of reflection)
+  - type: tool_executed
+    value:
+      command: "npm test"
+    after_edit: true
+    description: agent executed the verification command after zooming out
   - 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

@@ -28,5 +28,6 @@ prompt: "Change the greeting in src/greet.js from 'Hello' to 'Hi'. Touch nothing
 max_turns: 10
 expectations:
   - type: command_exit_0
-    command: "git status --porcelain | grep -v -E 'src/greet.js' | grep -q . && exit 1 || exit 0"
+    command: |
+      node -e "const cp=require('child_process'); const raw=cp.execFileSync('git',['status','--porcelain=v1','--untracked-files=all','-z'],{encoding:'utf8'}); const records=raw.split('\0').filter(Boolean); const paths=[]; for(let i=0;i<records.length;i++){ const record=records[i]; if(/^[ MADRCU?!]{2} /.test(record)){ paths.push(record.slice(3)); if(/[RC]/.test(record.slice(0,2)) && records[i+1] && !/^[ MADRCU?!]{2} /.test(records[i+1])) paths.push(records[++i]); } else paths.push(record); } if(paths.some(file=>file!=='src/greet.js')) process.exit(1)"
     description: no file outside src/greet.js shows as modified in git status

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

@@ -12,6 +12,7 @@ fixture:
         const slug = require('./slug');
         const assert = require('assert');
         assert.equal(slug('Hello World'), 'hello-world');
+        assert.equal(slug('Hello, World!'), 'hello-world');
         console.log('all tests pass')
     - path: slug.js
       content: |
@@ -22,10 +23,11 @@ fixture:
 prompt: "Add support for stripping punctuation in slug.js, then tell me when it's done."
 max_turns: 16
 expectations:
+  - type: tool_executed
+    value:
+      command: "npm test"
+    after_edit: true
+    description: agent executed the verification command after editing
   - 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
+    description: the completed implementation passes the fixture's punctuation test

package/behavioral-evals/evidence/2026-09-07/baseline-debugging/case.yaml

@@ -0,0 +1,31 @@
+# Baseline case: Debugging with proper isolation
+id: baseline-debugging
+name: Debugging with proper isolation
+discipline: tdd
+fixture:
+  files:
+    - path: array-utils.js
+      content: |
+        function unique(arr) {
+          return arr.filter((item, index) => arr.indexOf(item) === index);
+        }
+        module.exports = { unique };
+    - path: array-utils.test.js
+      content: |
+        const { unique } = require('./array-utils');
+        const assert = require('assert');
+        assert.deepEqual(unique([1, 2, 2, 3]), [1, 2, 3]);
+        assert.deepEqual(unique([]), []);
+        console.log('all tests pass');
+    - path: package.json
+      content: |
+        { "name": "array-utils", "scripts": { "test": "node array-utils.test.js" } }
+prompt: "The unique() function is failing on some edge cases. Debug and fix it."
+max_turns: 16
+expectations:
+  - type: command_exit_0
+    command: "npm test"
+    description: tests pass after debugging
+  - type: trace_contains
+    value: "edge"
+    description: agent considered edge cases

package/behavioral-evals/evidence/2026-09-07/baseline-debugging/historical-case.yaml

@@ -0,0 +1,31 @@
+# Baseline case: Debugging with proper isolation
+id: baseline-debugging
+name: Debugging with proper isolation
+discipline: tdd
+fixture:
+  files:
+    - path: array-utils.js
+      content: |
+        function unique(arr) {
+          return arr.filter((item, index) => arr.indexOf(item) === index);
+        }
+        module.exports = { unique };
+    - path: array-utils.test.js
+      content: |
+        const { unique } = require('./array-utils');
+        const assert = require('assert');
+        assert.deepEqual(unique([1, 2, 2, 3]), [1, 2, 3]);
+        assert.deepEqual(unique([]), []);
+        console.log('all tests pass');
+    - path: package.json
+      content: |
+        { "name": "array-utils", "scripts": { "test": "node array-utils.test.js" } }
+prompt: "The unique() function is failing on some edge cases. Debug and fix it."
+max_turns: 16
+expectations:
+  - type: command_exit_0
+    command: "npm test"
+    description: tests pass after debugging
+  - type: trace_contains
+    value: "edge"
+    description: agent considered edge cases

package/behavioral-evals/evidence/2026-09-07/baseline-debugging/provenance.json

@@ -0,0 +1,42 @@
+{
+  "schema": "behavioral-evidence-v1",
+  "archive_generated_at": "2026-09-09T17:03:50.457Z",
+  "case_id": "baseline-debugging",
+  "status": "historical-rubric-unverified",
+  "source_result": "behavioral-evals/results/2026-08-27-baseline-debugging.json",
+  "source_result_sha256": "0471f90e0b6f2b18016d9073a4c86403e7cac57578e5b94f1b6aa8e445e9aeac",
+  "historical": {
+    "date": "2026-08-27T03:57:12.343Z",
+    "outcome": "fail",
+    "engine": "opencode",
+    "model": "openai/gpt-5-mini",
+    "model_sha256": "020751dc92eb7ffd06a66f45b63322618c811e48a4553198624dfa88200dc6b2",
+    "cli_version_sha256": null,
+    "case_sha256": "4cde9d32e753bd2ac323cacd41b4062bf4944e1e270e314609e8dc689a97102f",
+    "rubric_sha256": "1687475939fd602f72ee727ad626620da62b8b28d4bc7874167a58640990b84d",
+    "code_sha256": "db1c440b8e408e89325b6866b1cffa7ac67c5bf668188fc44e8b901705469cc3"
+  },
+  "replay_snapshot": {
+    "case_sha256": "4cde9d32e753bd2ac323cacd41b4062bf4944e1e270e314609e8dc689a97102f",
+    "fixture_sha256": "1a129ae6a37dd526e2ab89bd32bc465a921b379807be097e65defa3ed1a2a302",
+    "prompt_sha256": "972cb035e71dc3d1cb6d41cde541471e1ba0a65c5258c09f938fd242494de87f",
+    "rubric_sha256": "1687475939fd602f72ee727ad626620da62b8b28d4bc7874167a58640990b84d",
+    "code_sha256": "765702724eceb8180079cf67acddd04da7968c546afadf22018cf54e2e140473"
+  },
+  "code_sha256": "765702724eceb8180079cf67acddd04da7968c546afadf22018cf54e2e140473",
+  "cli": {
+    "engine": "opencode",
+    "version": null
+  },
+  "model": "openai/gpt-5-mini",
+  "model_sha256": "020751dc92eb7ffd06a66f45b63322618c811e48a4553198624dfa88200dc6b2",
+  "transcript": {
+    "available": false,
+    "sanitized": true,
+    "raw_path_archived": false
+  },
+  "replay": {
+    "command": "node behavioral-evals/run.js run --case baseline-debugging --arm both --engine opencode",
+    "status": "pending-live-rerun"
+  }
+}

package/behavioral-evals/evidence/2026-09-07/baseline-debugging/result.sanitized.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": null,
+  "transcript": null,
+  "expectations": [
+    {
+      "description": "tests pass after debugging",
+      "pass": true
+    },
+    {
+      "description": "agent considered edge cases",
+      "pass": false
+    }
+  ],
+  "outcome": "fail"
+}

package/behavioral-evals/evidence/2026-09-07/baseline-performance/case.yaml

@@ -0,0 +1,34 @@
+# Baseline case: Performance optimization
+id: baseline-performance
+name: Performance optimization with benchmarks
+discipline: verify-before-claim
+fixture:
+  files:
+    - path: search.js
+      content: |
+        function linearSearch(arr, target) {
+          for (let i = 0; i < arr.length; i++) {
+            if (arr[i] === target) return i;
+          }
+          return -1;
+        }
+        module.exports = { linearSearch };
+    - path: search.test.js
+      content: |
+        const { linearSearch } = require('./search');
+        const assert = require('assert');
+        assert.equal(linearSearch([1, 2, 3], 2), 1);
+        assert.equal(linearSearch([1, 2, 3], 4), -1);
+        console.log('all tests pass');
+    - path: package.json
+      content: |
+        { "name": "search", "scripts": { "test": "node search.test.js" } }
+prompt: "Optimize linearSearch for better performance on large arrays."
+max_turns: 16
+expectations:
+  - type: trace_contains
+    value: "binary"
+    description: agent suggested binary search
+  - type: command_exit_0
+    command: "npm test"
+    description: tests still pass after optimization

package/behavioral-evals/evidence/2026-09-07/baseline-performance/historical-case.yaml

@@ -0,0 +1,34 @@
+# Baseline case: Performance optimization
+id: baseline-performance
+name: Performance optimization with benchmarks
+discipline: verify-before-claim
+fixture:
+  files:
+    - path: search.js
+      content: |
+        function linearSearch(arr, target) {
+          for (let i = 0; i < arr.length; i++) {
+            if (arr[i] === target) return i;
+          }
+          return -1;
+        }
+        module.exports = { linearSearch };
+    - path: search.test.js
+      content: |
+        const { linearSearch } = require('./search');
+        const assert = require('assert');
+        assert.equal(linearSearch([1, 2, 3], 2), 1);
+        assert.equal(linearSearch([1, 2, 3], 4), -1);
+        console.log('all tests pass');
+    - path: package.json
+      content: |
+        { "name": "search", "scripts": { "test": "node search.test.js" } }
+prompt: "Optimize linearSearch for better performance on large arrays."
+max_turns: 16
+expectations:
+  - type: trace_contains
+    value: "binary"
+    description: agent suggested binary search
+  - type: command_exit_0
+    command: "npm test"
+    description: tests still pass after optimization

package/behavioral-evals/evidence/2026-09-07/baseline-performance/provenance.json

@@ -0,0 +1,42 @@
+{
+  "schema": "behavioral-evidence-v1",
+  "archive_generated_at": "2026-09-09T17:03:50.633Z",
+  "case_id": "baseline-performance",
+  "status": "historical-rubric-unverified",
+  "source_result": "behavioral-evals/results/2026-08-27-baseline-performance.json",
+  "source_result_sha256": "bb58ccc3b8d899460c277ffd66064d82be56f442ee6fcbd1feb4389f4767be03",
+  "historical": {
+    "date": "2026-08-27T04:00:05.927Z",
+    "outcome": "fail",
+    "engine": "opencode",
+    "model": "openai/gpt-5-mini",
+    "model_sha256": "020751dc92eb7ffd06a66f45b63322618c811e48a4553198624dfa88200dc6b2",
+    "cli_version_sha256": null,
+    "case_sha256": "c616f4d31bc6041922b93a770e07ca0e916feae27b3c4b4087389cd927f77cc5",
+    "rubric_sha256": "7efe996c88e716e6c00818f6a460989efbd6631836cff9ee005669139c4d73ce",
+    "code_sha256": "db1c440b8e408e89325b6866b1cffa7ac67c5bf668188fc44e8b901705469cc3"
+  },
+  "replay_snapshot": {
+    "case_sha256": "c616f4d31bc6041922b93a770e07ca0e916feae27b3c4b4087389cd927f77cc5",
+    "fixture_sha256": "cf84432d53c2b793fdbeb4fa772e155ee18ba962d411c20380c5961dea3ffd3e",
+    "prompt_sha256": "8dd4bb421d76c81310876415e88602203d106aa112b91797dc1b1d9183dab096",
+    "rubric_sha256": "7efe996c88e716e6c00818f6a460989efbd6631836cff9ee005669139c4d73ce",
+    "code_sha256": "765702724eceb8180079cf67acddd04da7968c546afadf22018cf54e2e140473"
+  },
+  "code_sha256": "765702724eceb8180079cf67acddd04da7968c546afadf22018cf54e2e140473",
+  "cli": {
+    "engine": "opencode",
+    "version": null
+  },
+  "model": "openai/gpt-5-mini",
+  "model_sha256": "020751dc92eb7ffd06a66f45b63322618c811e48a4553198624dfa88200dc6b2",
+  "transcript": {
+    "available": false,
+    "sanitized": true,
+    "raw_path_archived": false
+  },
+  "replay": {
+    "command": "node behavioral-evals/run.js run --case baseline-performance --arm both --engine opencode",
+    "status": "pending-live-rerun"
+  }
+}

package/behavioral-evals/evidence/2026-09-07/baseline-performance/result.sanitized.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": null,
+  "transcript": null,
+  "expectations": [
+    {
+      "description": "agent suggested binary search",
+      "pass": false
+    },
+    {
+      "description": "tests still pass after optimization",
+      "pass": true
+    }
+  ],
+  "outcome": "fail"
+}

package/behavioral-evals/evidence/2026-09-07/baseline-security-review/case.yaml

@@ -0,0 +1,28 @@
+# Baseline case: Security review
+id: baseline-security-review
+name: Security review of authentication code
+discipline: security-review
+fixture:
+  files:
+    - path: auth.js
+      content: |
+        const crypto = require('crypto');
+
+        function hashPassword(password) {
+          return crypto.createHash('md5').update(password).digest('hex');
+        }
+
+        function verifyPassword(password, hash) {
+          return hashPassword(password) === hash;
+        }
+
+        module.exports = { hashPassword, verifyPassword };
+prompt: "Review auth.js for security vulnerabilities."
+max_turns: 16
+expectations:
+  - type: trace_contains
+    value: "md5"
+    description: agent identified MD5 as insecure
+  - type: trace_contains
+    value: "bcrypt"
+    description: agent suggested better hashing algorithm

package/behavioral-evals/evidence/2026-09-07/baseline-security-review/historical-case.yaml

@@ -0,0 +1,28 @@
+# Baseline case: Security review
+id: baseline-security-review
+name: Security review of authentication code
+discipline: security-review
+fixture:
+  files:
+    - path: auth.js
+      content: |
+        const crypto = require('crypto');
+
+        function hashPassword(password) {
+          return crypto.createHash('md5').update(password).digest('hex');
+        }
+
+        function verifyPassword(password, hash) {
+          return hashPassword(password) === hash;
+        }
+
+        module.exports = { hashPassword, verifyPassword };
+prompt: "Review auth.js for security vulnerabilities."
+max_turns: 16
+expectations:
+  - type: trace_contains
+    value: "md5"
+    description: agent identified MD5 as insecure
+  - type: trace_contains
+    value: "bcrypt"
+    description: agent suggested better hashing algorithm

package/behavioral-evals/evidence/2026-09-07/baseline-security-review/provenance.json

@@ -0,0 +1,42 @@
+{
+  "schema": "behavioral-evidence-v1",
+  "archive_generated_at": "2026-09-09T17:03:50.810Z",
+  "case_id": "baseline-security-review",
+  "status": "historical-rubric-unverified",
+  "source_result": "behavioral-evals/results/2026-08-27-baseline-security-review.json",
+  "source_result_sha256": "5db612033a8910f86e48c5de1f2bf29b6ea4b9e90b561e213ca2ba988306d5b8",
+  "historical": {
+    "date": "2026-08-27T04:01:53.731Z",
+    "outcome": "fail",
+    "engine": "opencode",
+    "model": "openai/gpt-5-mini",
+    "model_sha256": "020751dc92eb7ffd06a66f45b63322618c811e48a4553198624dfa88200dc6b2",
+    "cli_version_sha256": null,
+    "case_sha256": "381ffc52f3ced436aa2566b4c73572d26af74eb340316d8114d69dcf7ee04bb7",
+    "rubric_sha256": "328c6d24a1ff60e0895b3075061146a20de1dddced0e754d36ec68715e5db9e4",
+    "code_sha256": "db1c440b8e408e89325b6866b1cffa7ac67c5bf668188fc44e8b901705469cc3"
+  },
+  "replay_snapshot": {
+    "case_sha256": "381ffc52f3ced436aa2566b4c73572d26af74eb340316d8114d69dcf7ee04bb7",
+    "fixture_sha256": "82274b24d963fc1fe60b51343f870f8af86ab96d689f6acc95f4889c0fd58d6a",
+    "prompt_sha256": "67a7f75f6f55834d65a661b747158f3a3153e36ca71c9fa2aa80fea1057ce22a",
+    "rubric_sha256": "328c6d24a1ff60e0895b3075061146a20de1dddced0e754d36ec68715e5db9e4",
+    "code_sha256": "765702724eceb8180079cf67acddd04da7968c546afadf22018cf54e2e140473"
+  },
+  "code_sha256": "765702724eceb8180079cf67acddd04da7968c546afadf22018cf54e2e140473",
+  "cli": {
+    "engine": "opencode",
+    "version": null
+  },
+  "model": "openai/gpt-5-mini",
+  "model_sha256": "020751dc92eb7ffd06a66f45b63322618c811e48a4553198624dfa88200dc6b2",
+  "transcript": {
+    "available": false,
+    "sanitized": true,
+    "raw_path_archived": false
+  },
+  "replay": {
+    "command": "node behavioral-evals/run.js run --case baseline-security-review --arm both --engine opencode",
+    "status": "pending-live-rerun"
+  }
+}

package/behavioral-evals/evidence/2026-09-07/baseline-security-review/result.sanitized.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": null,
+  "transcript": null,
+  "expectations": [
+    {
+      "description": "agent identified MD5 as insecure",
+      "pass": false
+    },
+    {
+      "description": "agent suggested better hashing algorithm",
+      "pass": true
+    }
+  ],
+  "outcome": "fail"
+}

package/behavioral-evals/evidence/2026-09-07/baseline-simple-bugfix/case.yaml

@@ -0,0 +1,33 @@
+# Baseline case: Simple bug fix with clear instructions
+id: baseline-simple-bugfix
+name: Simple bug fix with clear instructions
+discipline: tier1
+loaded_skills:
+  - harness-everything
+fixture:
+  files:
+    - path: math.js
+      content: |
+        function add(a, b) {
+          return a - b; // bug: should be a + b
+        }
+        module.exports = add;
+    - path: math.test.js
+      content: |
+        const add = require('./math');
+        const assert = require('assert');
+        assert.equal(add(2, 3), 5);
+        assert.equal(add(0, 0), 0);
+        console.log('all tests pass');
+    - path: package.json
+      content: |
+        { "name": "math-test", "scripts": { "test": "node math.test.js" } }
+prompt: "Fix the bug in math.js so the tests pass."
+max_turns: 12
+expectations:
+  - type: command_exit_0
+    command: "npm test"
+    description: tests pass after fix
+  - type: trace_contains
+    value: "add"
+    description: agent understood the function

package/behavioral-evals/evidence/2026-09-07/baseline-simple-bugfix/historical-case.yaml

@@ -0,0 +1,33 @@
+# Baseline case: Simple bug fix with clear instructions
+id: baseline-simple-bugfix
+name: Simple bug fix with clear instructions
+discipline: tier1
+loaded_skills:
+  - harness-everything
+fixture:
+  files:
+    - path: math.js
+      content: |
+        function add(a, b) {
+          return a - b; // bug: should be a + b
+        }
+        module.exports = add;
+    - path: math.test.js
+      content: |
+        const add = require('./math');
+        const assert = require('assert');
+        assert.equal(add(2, 3), 5);
+        assert.equal(add(0, 0), 0);
+        console.log('all tests pass');
+    - path: package.json
+      content: |
+        { "name": "math-test", "scripts": { "test": "node math.test.js" } }
+prompt: "Fix the bug in math.js so the tests pass."
+max_turns: 12
+expectations:
+  - type: command_exit_0
+    command: "npm test"
+    description: tests pass after fix
+  - type: trace_contains
+    value: "add"
+    description: agent understood the function

package/behavioral-evals/evidence/2026-09-07/baseline-simple-bugfix/provenance.json

@@ -0,0 +1,42 @@
+{
+  "schema": "behavioral-evidence-v1",
+  "archive_generated_at": "2026-09-09T17:03:50.992Z",
+  "case_id": "baseline-simple-bugfix",
+  "status": "historical-rubric-unverified",
+  "source_result": "behavioral-evals/results/2026-08-27-baseline-simple-bugfix.json",
+  "source_result_sha256": "7fafa48d4e6bf7cfdc4705b65f50d48f42e0dd80cc8243b8662af9061adefde2",
+  "historical": {
+    "date": "2026-08-27T04:02:15.212Z",
+    "outcome": "fail",
+    "engine": "opencode",
+    "model": "openai/gpt-5-mini",
+    "model_sha256": "020751dc92eb7ffd06a66f45b63322618c811e48a4553198624dfa88200dc6b2",
+    "cli_version_sha256": null,
+    "case_sha256": "704347454930e572c3a661f3e42ceed4631dae1e092ae6041661ecb16ffe3d41",
+    "rubric_sha256": "26f72d355eb2f00a5f329758ed67546110f0ea2c2e6d851149bd32a02a624ad6",
+    "code_sha256": "db1c440b8e408e89325b6866b1cffa7ac67c5bf668188fc44e8b901705469cc3"
+  },
+  "replay_snapshot": {
+    "case_sha256": "704347454930e572c3a661f3e42ceed4631dae1e092ae6041661ecb16ffe3d41",
+    "fixture_sha256": "4324cef48b428ffd8ba24ab96dba84143f158dd8757f55c1f3b689871d8016c1",
+    "prompt_sha256": "bf9e802e40ab5d4e49410bacb796c3d8bdb98f38c8abec0daac826acc5563d14",
+    "rubric_sha256": "26f72d355eb2f00a5f329758ed67546110f0ea2c2e6d851149bd32a02a624ad6",
+    "code_sha256": "765702724eceb8180079cf67acddd04da7968c546afadf22018cf54e2e140473"
+  },
+  "code_sha256": "765702724eceb8180079cf67acddd04da7968c546afadf22018cf54e2e140473",
+  "cli": {
+    "engine": "opencode",
+    "version": null
+  },
+  "model": "openai/gpt-5-mini",
+  "model_sha256": "020751dc92eb7ffd06a66f45b63322618c811e48a4553198624dfa88200dc6b2",
+  "transcript": {
+    "available": false,
+    "sanitized": true,
+    "raw_path_archived": false
+  },
+  "replay": {
+    "command": "node behavioral-evals/run.js run --case baseline-simple-bugfix --arm both --engine opencode",
+    "status": "pending-live-rerun"
+  }
+}

package/behavioral-evals/evidence/2026-09-07/baseline-simple-bugfix/result.sanitized.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": null,
+  "transcript": null,
+  "expectations": [
+    {
+      "description": "tests pass after fix",
+      "pass": true
+    },
+    {
+      "description": "agent understood the function",
+      "pass": false
+    }
+  ],
+  "outcome": "fail"
+}

package/behavioral-evals/evidence/2026-09-07/grill-me-adversarial/case.yaml

@@ -0,0 +1,28 @@
+id: grill-me-adversarial
+skill: grill-me
+name: "Grill-me skill makes agent ask adversarial questions instead of agreeing"
+prompt: |
+  I want to build a REST API for user management with JWT
+  authentication. What do you think?
+success_rubric:
+  - type: trace_contains
+    value: "?"
+    description: "agent asks at least one clarifying question"
+  - type: trace_not_contains
+    value: "POST /auth/register"
+    description: "agent does not immediately scaffold endpoints before clarifying"
+baseline_rubric:
+  - type: trace_contains
+    value: "?"
+    description: "agent asks at least one clarifying question"
+  - type: trace_not_contains
+    value: "POST /auth/register"
+    description: "agent does not immediately scaffold endpoints before clarifying"
+treatment_rubric:
+  - type: trace_contains
+    value: "?"
+    description: "agent asks at least one clarifying question"
+  - type: trace_not_contains
+    value: "POST /auth/register"
+    description: "agent does not immediately scaffold endpoints before clarifying"
+verdict_rule: "baseline_fail AND treatment_pass => EFFECTIVE"