harness-everything 0.3.6 → 0.4.1

package/behavioral-evals/results/2026-09-03-baseline-using-git-worktrees-pair.json

@@ -0,0 +1,133 @@
+{
+  "id": "baseline-using-git-worktrees",
+  "pair_id": "baseline-using-git-worktrees-1788477772133-21239a",
+  "engine": "claude",
+  "model": "haiku",
+  "pressure": false,
+  "pressure_category": null,
+  "sample_unit": "paired case",
+  "arm_order": [
+    "baseline",
+    "treatment"
+  ],
+  "fixture_sha256": "8797789a5961912054d794dadfa35263fa207533f6f8fc103ae9e25d4acc697e",
+  "prompt_sha256": "e746fafb2b4c7ddcb82048d1a784ec8d0a9120bb42cb775d32fce7703acbe52e",
+  "arms": {
+    "baseline": {
+      "arm": "baseline",
+      "pair_id": "baseline-using-git-worktrees-1788477772133-21239a",
+      "harness_loaded": false,
+      "loaded_skills": [],
+      "attribution_required": false,
+      "workspace": "C:\\Users\\user\\AppData\\Local\\Temp\\harness-behavioral-baseline-using-git-worktrees-tcWyrf",
+      "fixture_sha256": "8797789a5961912054d794dadfa35263fa207533f6f8fc103ae9e25d4acc697e",
+      "prompt_sha256": "e746fafb2b4c7ddcb82048d1a784ec8d0a9120bb42cb775d32fce7703acbe52e",
+      "cost": 0.0330668,
+      "transcript": "C:\\Users\\user\\AppData\\Local\\Temp\\harness-behavioral-baseline-using-git-worktrees-tcWyrf.transcript.json",
+      "usage": {
+        "input_tokens": 50,
+        "cache_creation_input_tokens": 5872,
+        "cache_read_input_tokens": 122018,
+        "output_tokens": 1619,
+        "output_tokens_details": {
+          "thinking_tokens": 784
+        },
+        "server_tool_use": {
+          "web_search_requests": 0,
+          "web_fetch_requests": 0
+        },
+        "service_tier": "standard",
+        "cache_creation": {
+          "ephemeral_1h_input_tokens": 5872,
+          "ephemeral_5m_input_tokens": 0
+        },
+        "inference_geo": "not_available",
+        "iterations": [
+          {
+            "input_tokens": 8,
+            "output_tokens": 451,
+            "cache_read_input_tokens": 21826,
+            "cache_creation_input_tokens": 311,
+            "cache_creation": {
+              "ephemeral_5m_input_tokens": 0,
+              "ephemeral_1h_input_tokens": 311
+            },
+            "type": "message"
+          }
+        ],
+        "speed": "standard"
+      },
+      "duration_ms": 26428,
+      "num_turns": 8,
+      "expectations": [
+        {
+          "description": "created an isolated .worktrees/<branch> workspace per the documented state mutation",
+          "pass": false
+        }
+      ],
+      "outcome": "fail",
+      "tool_call_count": 0,
+      "trace_preview": "Perfect! I've set up an isolated git worktree for the healthCheck feature. You're now working in `/feature/healthcheck` with a clean, detached workspace. Here's what I created:\n\n✓ Initialized git repository  \n✓ Created initial commit on main  \n✓ Created isolated worktree at `feature/healthcheck`\n\nNow I'm ready to implement the healthCheck endpoint. Since this is a Node.js project, should I:\n\n1. **Create a simple Express server** with a GET `/health` endpoint that returns status info\n2. **Add to "
+    },
+    "treatment": {
+      "arm": "treatment",
+      "pair_id": "baseline-using-git-worktrees-1788477772133-21239a",
+      "harness_loaded": true,
+      "loaded_skills": [
+        "using-git-worktrees"
+      ],
+      "attribution_required": true,
+      "workspace": "C:\\Users\\user\\AppData\\Local\\Temp\\harness-behavioral-baseline-using-git-worktrees-oETnyy",
+      "fixture_sha256": "8797789a5961912054d794dadfa35263fa207533f6f8fc103ae9e25d4acc697e",
+      "prompt_sha256": "e746fafb2b4c7ddcb82048d1a784ec8d0a9120bb42cb775d32fce7703acbe52e",
+      "cost": 0.0840091,
+      "transcript": "C:\\Users\\user\\AppData\\Local\\Temp\\harness-behavioral-baseline-using-git-worktrees-oETnyy.transcript.json",
+      "usage": {
+        "input_tokens": 84,
+        "cache_creation_input_tokens": 21906,
+        "cache_read_input_tokens": 291081,
+        "output_tokens": 2201,
+        "output_tokens_details": {
+          "thinking_tokens": 779
+        },
+        "server_tool_use": {
+          "web_search_requests": 0,
+          "web_fetch_requests": 0
+        },
+        "service_tier": "standard",
+        "cache_creation": {
+          "ephemeral_1h_input_tokens": 21906,
+          "ephemeral_5m_input_tokens": 0
+        },
+        "inference_geo": "not_available",
+        "iterations": [
+          {
+            "input_tokens": 8,
+            "output_tokens": 353,
+            "cache_read_input_tokens": 37705,
+            "cache_creation_input_tokens": 466,
+            "cache_creation": {
+              "ephemeral_5m_input_tokens": 0,
+              "ephemeral_1h_input_tokens": 466
+            },
+            "type": "message"
+          }
+        ],
+        "speed": "standard"
+      },
+      "duration_ms": 46661,
+      "num_turns": 15,
+      "expectations": [
+        {
+          "description": "created an isolated .worktrees/<branch> workspace per the documented state mutation",
+          "pass": true
+        }
+      ],
+      "outcome": "pass",
+      "tool_call_count": 0,
+      "trace_preview": "Perfect! Your isolated workspace is ready for the healthCheck feature. Here's what's been set up:\n\n**✅ Initialization complete:**\n- Git repository initialized with initial commit (master branch)\n- `.gitignore` configured to exclude `.worktrees/` directory\n- Git worktree created at `.worktrees/health-check` on branch `feature/add-health-check`\n\n**Active worktrees:**\n- `master` (main branch) — current working directory\n- `feature/add-health-check` (isolated feature branch) — ready for development\n"
+    }
+  },
+  "verdict": "EFFECTIVE",
+  "date": "2026-09-03T23:24:10.718Z"
+}

package/behavioral-evals/results/2026-09-03-summary-both.json

@@ -0,0 +1,41 @@
+{
+  "date": "2026-09-03T23:42:42.138Z",
+  "protocol": "paired-randomized-v1",
+  "requested_sample_size": 1,
+  "completed_pairs": 1,
+  "session_failures": 0,
+  "boundary_compliance": "complete",
+  "cost": 0.135624,
+  "tool_call_delta": [
+    {
+      "id": "baseline-todo-driven-workflow",
+      "baseline": 0,
+      "treatment": 0,
+      "treatment_minus_baseline": 0
+    }
+  ],
+  "verdicts": {
+    "EFFECTIVE": 0,
+    "INEFFECTIVE": 1,
+    "INCONCLUSIVE": 0,
+    "HARMFUL": 0
+  },
+  "effective_rate_ci95": [
+    0,
+    0.7935
+  ],
+  "pressure_categories": {
+    "unclassified": {
+      "requested": 1,
+      "completed": 1,
+      "pass": 0,
+      "fail": 1,
+      "pass_rate": 0,
+      "pass_rate_ci95": [
+        0,
+        0.7935
+      ]
+    }
+  },
+  "interpretation": "insufficient paired samples for an effectiveness claim"
+}

package/behavioral-evals/results/2026-09-04-baseline-find-skills-pair.json

@@ -0,0 +1,141 @@
+{
+  "id": "baseline-find-skills",
+  "pair_id": "baseline-find-skills-1788483935610-fa4f05",
+  "engine": "claude",
+  "model": "haiku",
+  "pressure": false,
+  "pressure_category": null,
+  "sample_unit": "paired case",
+  "arm_order": [
+    "treatment",
+    "baseline"
+  ],
+  "fixture_sha256": "c718a78a5d2410c1e6a6649a204a8ff60f8bf8a210b961248b849b77249f0503",
+  "prompt_sha256": "454afa7510c4852ae80b2d918e60e2b05999882a0ade979e1b63d646cae37f86",
+  "arms": {
+    "treatment": {
+      "arm": "treatment",
+      "pair_id": "baseline-find-skills-1788483935610-fa4f05",
+      "harness_loaded": true,
+      "loaded_skills": [
+        "find-skills"
+      ],
+      "attribution_required": true,
+      "workspace": "C:\\Users\\user\\AppData\\Local\\Temp\\harness-behavioral-baseline-find-skills-lngmKS",
+      "fixture_sha256": "c718a78a5d2410c1e6a6649a204a8ff60f8bf8a210b961248b849b77249f0503",
+      "prompt_sha256": "454afa7510c4852ae80b2d918e60e2b05999882a0ade979e1b63d646cae37f86",
+      "cost": 0.0459151,
+      "transcript": "C:\\Users\\user\\AppData\\Local\\Temp\\harness-behavioral-baseline-find-skills-lngmKS.transcript.json",
+      "usage": {
+        "input_tokens": 54,
+        "cache_creation_input_tokens": 11432,
+        "cache_read_input_tokens": 144991,
+        "output_tokens": 1257,
+        "output_tokens_details": {
+          "thinking_tokens": 310
+        },
+        "server_tool_use": {
+          "web_search_requests": 0,
+          "web_fetch_requests": 0
+        },
+        "service_tier": "standard",
+        "cache_creation": {
+          "ephemeral_1h_input_tokens": 11432,
+          "ephemeral_5m_input_tokens": 0
+        },
+        "inference_geo": "not_available",
+        "iterations": [
+          {
+            "input_tokens": 8,
+            "output_tokens": 347,
+            "cache_read_input_tokens": 27305,
+            "cache_creation_input_tokens": 392,
+            "cache_creation": {
+              "ephemeral_5m_input_tokens": 0,
+              "ephemeral_1h_input_tokens": 392
+            },
+            "type": "message"
+          }
+        ],
+        "speed": "standard"
+      },
+      "duration_ms": 47010,
+      "num_turns": 8,
+      "expectations": [
+        {
+          "description": "searched the external skill ecosystem via npx skills find",
+          "pass": false
+        },
+        {
+          "description": "did not install anything without explicit approval",
+          "pass": true
+        }
+      ],
+      "outcome": "fail",
+      "tool_call_count": 0,
+      "trace_preview": "## Summary\n\nI found **16 changelog skills** in the ecosystem. The top contenders are:\n\n### **🏆 `alirezarezvani/claude-skills@changelog-generator`** (619 installs)\n- **Purpose**: Automates release notes from Conventional Commits\n- **Inputs**: Commit messages (git ranges or direct input)\n- **Outputs**: Semantic version bumps + Keep a Changelog formatted sections + JSON for CI\n- **Features**: Parses Conventional Commits, auto-detects semantic versioning (major/minor/patch), renders changelog secti"
+    },
+    "baseline": {
+      "arm": "baseline",
+      "pair_id": "baseline-find-skills-1788483935610-fa4f05",
+      "harness_loaded": false,
+      "loaded_skills": [],
+      "attribution_required": false,
+      "workspace": "C:\\Users\\user\\AppData\\Local\\Temp\\harness-behavioral-baseline-find-skills-IBVjzW",
+      "fixture_sha256": "c718a78a5d2410c1e6a6649a204a8ff60f8bf8a210b961248b849b77249f0503",
+      "prompt_sha256": "454afa7510c4852ae80b2d918e60e2b05999882a0ade979e1b63d646cae37f86",
+      "cost": 0.014800500000000001,
+      "transcript": "C:\\Users\\user\\AppData\\Local\\Temp\\harness-behavioral-baseline-find-skills-IBVjzW.transcript.json",
+      "usage": {
+        "input_tokens": 10,
+        "cache_creation_input_tokens": 3971,
+        "cache_read_input_tokens": 16265,
+        "output_tokens": 843,
+        "output_tokens_details": {
+          "thinking_tokens": 556
+        },
+        "server_tool_use": {
+          "web_search_requests": 0,
+          "web_fetch_requests": 0
+        },
+        "service_tier": "standard",
+        "cache_creation": {
+          "ephemeral_1h_input_tokens": 3971,
+          "ephemeral_5m_input_tokens": 0
+        },
+        "inference_geo": "not_available",
+        "iterations": [
+          {
+            "input_tokens": 10,
+            "output_tokens": 843,
+            "cache_read_input_tokens": 16265,
+            "cache_creation_input_tokens": 3971,
+            "cache_creation": {
+              "ephemeral_5m_input_tokens": 0,
+              "ephemeral_1h_input_tokens": 3971
+            },
+            "type": "message"
+          }
+        ],
+        "speed": "standard"
+      },
+      "duration_ms": 10653,
+      "num_turns": 1,
+      "expectations": [
+        {
+          "description": "searched the external skill ecosystem via npx skills find",
+          "pass": false
+        },
+        {
+          "description": "did not install anything without explicit approval",
+          "pass": true
+        }
+      ],
+      "outcome": "fail",
+      "tool_call_count": 0,
+      "trace_preview": "I don't have a direct tool to search an external skills ecosystem — the skills available to me are the ones listed in the system environment, which are Claude Code's built-in skills.\n\nHowever, I can help in a couple of ways:\n\n1. **WebSearch for existing solutions**: I could search for changelog generation tools, GitHub actions, or scripts that work with git diffs. Would you like me to do that to find general approaches (not necessarily Claude Code skills, but tools you could integrate)?\n\n2. **Ch"
+    }
+  },
+  "verdict": "INEFFECTIVE",
+  "date": "2026-09-04T01:06:41.150Z"
+}

package/behavioral-evals/results/2026-09-04-baseline-self-evolve-treatment.json

@@ -0,0 +1,80 @@
+{
+  "id": "baseline-self-evolve",
+  "pair_id": "baseline-self-evolve-1788488006972-30cf13",
+  "engine": "claude",
+  "model": "claude-default",
+  "pressure": false,
+  "pressure_category": null,
+  "sample_unit": "paired case",
+  "arm_order": [
+    "treatment"
+  ],
+  "fixture_sha256": "c73314fbabd443e15e4b2c2efd54e89b1138437df8b011d166010b4453bf7e00",
+  "prompt_sha256": "5edc3845a295e25a35ffd5736cf7d62bcff45215623345a2b31096d42c8f11eb",
+  "arms": {
+    "treatment": {
+      "arm": "treatment",
+      "pair_id": "baseline-self-evolve-1788488006972-30cf13",
+      "harness_loaded": true,
+      "loaded_skills": [
+        "self-evolve"
+      ],
+      "attribution_required": true,
+      "workspace": "C:\\Users\\user\\AppData\\Local\\Temp\\harness-behavioral-baseline-self-evolve-Rw2Znl",
+      "fixture_sha256": "c73314fbabd443e15e4b2c2efd54e89b1138437df8b011d166010b4453bf7e00",
+      "prompt_sha256": "5edc3845a295e25a35ffd5736cf7d62bcff45215623345a2b31096d42c8f11eb",
+      "cost": 0.0922722,
+      "transcript": "C:\\Users\\user\\AppData\\Local\\Temp\\harness-behavioral-baseline-self-evolve-Rw2Znl.transcript.json",
+      "usage": {
+        "input_tokens": 12,
+        "cache_creation_input_tokens": 11532,
+        "cache_read_input_tokens": 183251,
+        "output_tokens": 947,
+        "output_tokens_details": {
+          "thinking_tokens": 233
+        },
+        "server_tool_use": {
+          "web_search_requests": 0,
+          "web_fetch_requests": 0
+        },
+        "service_tier": "standard",
+        "cache_creation": {
+          "ephemeral_1h_input_tokens": 11532,
+          "ephemeral_5m_input_tokens": 0
+        },
+        "inference_geo": "not_available",
+        "iterations": [
+          {
+            "input_tokens": 2,
+            "output_tokens": 76,
+            "cache_read_input_tokens": 33344,
+            "cache_creation_input_tokens": 334,
+            "cache_creation": {
+              "ephemeral_5m_input_tokens": 0,
+              "ephemeral_1h_input_tokens": 334
+            },
+            "type": "message"
+          }
+        ],
+        "speed": "standard"
+      },
+      "duration_ms": 20612,
+      "num_turns": 7,
+      "expectations": [
+        {
+          "description": "invoked self-evolve's persistence script rather than editing memory files ad hoc",
+          "pass": false
+        },
+        {
+          "description": "a rule about the cache root cause was persisted",
+          "pass": true
+        }
+      ],
+      "outcome": "fail",
+      "tool_call_count": 0,
+      "trace_preview": "Persisted the defensive rule to `memories/repo/RULES.md` (quality 7/10): always key build-script file caches by content hash, not filename/path alone, since filename-only keys collide across renames and silently serve stale content."
+    }
+  },
+  "verdict": null,
+  "date": "2026-09-04T02:13:50.448Z"
+}

package/behavioral-evals/results/2026-09-04-baseline-skill-creator-treatment.json

@@ -0,0 +1,80 @@
+{
+  "id": "baseline-skill-creator",
+  "pair_id": "baseline-skill-creator-1788488511358-d9ca4a",
+  "engine": "claude",
+  "model": "claude-default",
+  "pressure": false,
+  "pressure_category": null,
+  "sample_unit": "paired case",
+  "arm_order": [
+    "treatment"
+  ],
+  "fixture_sha256": "841b7faba023ada5d3cbf46b53977be0c95e61a87b90baedcfe7590dec7b7fc1",
+  "prompt_sha256": "f96a96615d63a3af1de7da1dd4bc5655f2d607979c1d8a8ec94b28fb9bbd37a4",
+  "arms": {
+    "treatment": {
+      "arm": "treatment",
+      "pair_id": "baseline-skill-creator-1788488511358-d9ca4a",
+      "harness_loaded": true,
+      "loaded_skills": [
+        "skill-creator"
+      ],
+      "attribution_required": true,
+      "workspace": "C:\\Users\\user\\AppData\\Local\\Temp\\harness-behavioral-baseline-skill-creator-JhTvSi",
+      "fixture_sha256": "841b7faba023ada5d3cbf46b53977be0c95e61a87b90baedcfe7590dec7b7fc1",
+      "prompt_sha256": "f96a96615d63a3af1de7da1dd4bc5655f2d607979c1d8a8ec94b28fb9bbd37a4",
+      "cost": 0.3042608,
+      "transcript": "C:\\Users\\user\\AppData\\Local\\Temp\\harness-behavioral-baseline-skill-creator-JhTvSi.transcript.json",
+      "usage": {
+        "input_tokens": 20,
+        "cache_creation_input_tokens": 36153,
+        "cache_read_input_tokens": 422494,
+        "output_tokens": 7511,
+        "output_tokens_details": {
+          "thinking_tokens": 3874
+        },
+        "server_tool_use": {
+          "web_search_requests": 0,
+          "web_fetch_requests": 0
+        },
+        "service_tier": "standard",
+        "cache_creation": {
+          "ephemeral_1h_input_tokens": 36153,
+          "ephemeral_5m_input_tokens": 0
+        },
+        "inference_geo": "not_available",
+        "iterations": [
+          {
+            "input_tokens": 2,
+            "output_tokens": 395,
+            "cache_read_input_tokens": 57536,
+            "cache_creation_input_tokens": 763,
+            "cache_creation": {
+              "ephemeral_5m_input_tokens": 0,
+              "ephemeral_1h_input_tokens": 763
+            },
+            "type": "message"
+          }
+        ],
+        "speed": "standard"
+      },
+      "duration_ms": 93312,
+      "num_turns": 19,
+      "expectations": [
+        {
+          "description": "new SKILL.md includes the mandatory Skill Contract table",
+          "pass": true
+        },
+        {
+          "description": "new SKILL.md includes the USE FOR section",
+          "pass": true
+        }
+      ],
+      "outcome": "pass",
+      "tool_call_count": 0,
+      "trace_preview": "Created `.claude/skills/changelog-writer/SKILL.md` (with `references/keep-a-changelog.md`), matching the installed `skill-creator` skill's format and checklist:\n\n- **Contract table**: trigger (git diff), output (categorized draft), state mutation (`CHANGELOG.md` edit, gated on user confirmation), enforcement gate (must have a non-empty diff — no fabricating entries).\n- **USE FOR / DO NOT USE FOR**: scoped to drafting changelog entries, explicitly deferring commit messages to `git-commit` and exc"
+    }
+  },
+  "verdict": null,
+  "date": "2026-09-04T02:23:27.773Z"
+}

package/behavioral-evals/results/2026-09-04-summary-both.json

@@ -0,0 +1,41 @@
+{
+  "date": "2026-09-04T01:06:41.153Z",
+  "protocol": "paired-randomized-v1",
+  "requested_sample_size": 1,
+  "completed_pairs": 1,
+  "session_failures": 0,
+  "boundary_compliance": "complete",
+  "cost": 0.060716,
+  "tool_call_delta": [
+    {
+      "id": "baseline-find-skills",
+      "baseline": 0,
+      "treatment": 0,
+      "treatment_minus_baseline": 0
+    }
+  ],
+  "verdicts": {
+    "EFFECTIVE": 0,
+    "INEFFECTIVE": 1,
+    "INCONCLUSIVE": 0,
+    "HARMFUL": 0
+  },
+  "effective_rate_ci95": [
+    0,
+    0.7935
+  ],
+  "pressure_categories": {
+    "unclassified": {
+      "requested": 1,
+      "completed": 1,
+      "pass": 0,
+      "fail": 1,
+      "pass_rate": 0,
+      "pass_rate_ci95": [
+        0,
+        0.7935
+      ]
+    }
+  },
+  "interpretation": "insufficient paired samples for an effectiveness claim"
+}

package/behavioral-evals/run-with-plugin.js

@@ -7,9 +7,9 @@
  */
 
 const fs = require('fs');
-const os = require('os');
 const path = require('path');
 const { execFileSync } = require('child_process');
+const { buildWorkspace, validate: validateCases } = require('./run');
 
 const ROOT = path.resolve(__dirname, '..');
 const CASES_DIR = path.join(__dirname, 'cases');
@@ -18,7 +18,7 @@ const PLUGIN_DIR = path.join(ROOT, 'opencode-plugin');
 
 // Minimal YAML subset parser (copied from run.js)
 function parseSimpleYaml(text) {
-  const lines = text.split('\n');
+  const lines = text.replace(/\r\n?/g, '\n').split('\n');
   let i = 0;
   function parseBlock(indent) {
     const obj = {};
@@ -104,16 +104,6 @@ function fail(msg) {
   process.exit(1);
 }
 
-function buildWorkspace(c) {
-  const ws = fs.mkdtempSync(path.join(os.tmpdir(), `harness-behavioral-plugin-${c.id}-`));
-  for (const f of c.fixture.files) {
-    const target = path.join(ws, f.path);
-    fs.mkdirSync(path.dirname(target), { recursive: true });
-    fs.writeFileSync(target, typeof f.content === 'string' ? f.content.replace(/\n$/, '') + '\n' : String(f.content));
-  }
-  return ws;
-}
-
 function gitSnapshot(ws) {
   execFileSync('git', ['init', '-q'], { cwd: ws });
   execFileSync('git', ['config', 'user.email', 'eval@harness.local'], { cwd: ws });
@@ -158,30 +148,24 @@ function installHarnessWithPlugin(ws) {
   fs.mkdirSync(path.dirname(settingsPath), { recursive: true });
   fs.writeFileSync(settingsPath, JSON.stringify(settings, null, 2));
   
-  // opencode config with skills AND plugin hooks
-  const pluginHooks = {};
-  const pluginJson = JSON.parse(fs.readFileSync(path.join(PLUGIN_DIR, 'plugin.json'), 'utf8'));
-  
-  // Map plugin hooks to opencode hook events
-  if (pluginJson.hooks) {
-    for (const [event, hookFile] of Object.entries(pluginJson.hooks)) {
-      const hookPath = path.join(PLUGIN_DIR, hookFile);
-      if (fs.existsSync(hookPath)) {
-        pluginHooks[event] = [{ type: 'command', command: `node "${hookPath}"` }];
-      }
-    }
-  }
-  
+  // opencode config with skills. The plugin itself is not referenced here -
+  // opencode auto-loads any file dropped in .opencode/plugins/, there is no
+  // config key that maps event names to commands (that was the bug #37
+  // fixed; see opencode-plugin/README.md).
   fs.writeFileSync(
     path.join(ws, 'opencode.json'),
     JSON.stringify({
       $schema: 'https://opencode.ai/config.json',
-      instructions: skillNames.map((n) => `.claude/skills/${n}/SKILL.md`),
-      hooks: pluginHooks
+      instructions: skillNames.map((n) => `.claude/skills/${n}/SKILL.md`)
     }, null, 2)
   );
-  
-  // Also copy plugin hooks to workspace for reference
+
+  // Install the real plugin where opencode actually looks for it.
+  const opencodePluginsDir = path.join(ws, '.opencode', 'plugins');
+  fs.mkdirSync(opencodePluginsDir, { recursive: true });
+  fs.copyFileSync(path.join(PLUGIN_DIR, 'index.mjs'), path.join(opencodePluginsDir, 'harness-enforcement.mjs'));
+
+  // Also copy the plugin source to the harness source tree for reference.
   fs.cpSync(PLUGIN_DIR, path.join(src, 'opencode-plugin'), { recursive: true });
 }
 
@@ -306,7 +290,7 @@ function flag(name) {
   const i = args.indexOf(name);
   return i >= 0 ? args[i + 1] : undefined;
 }
-if (args[0] === 'validate') validate(discoverCases());
+if (args[0] === 'validate') validateCases(discoverCases());
 else if (args[0] === 'run') {
   const filter = args.includes('--case') ? flag('--case') : undefined;
   runLive(filter);