@c-d-cc/reap 0.16.5 → 0.16.6

package/README.md

@@ -229,6 +229,7 @@ Pre-approve N generations for autonomous execution:
 | `/reap.init`      | Initialize REAP in a project           |
 | `/reap.run`       | Execute a lifecycle command directly   |
 | `/reap.config`    | View/edit project configuration        |
+| `/reap.report`    | Submit a bug report or feature request |
 
 ## Agent Integration
 
@@ -250,6 +251,31 @@ Switch clients by editing `.reap/config.yml`, then run `reap install-skills` fol
 
 `/reap.evolve` can delegate the entire generation to a subagent that runs autonomously through all stages, surfacing only when genuinely blocked.
 
+### Evaluator Agent (opt-in)
+
+REAP ships a second subagent definition, `reap-evaluate`, that runs as an **independent reviewer** of the builder's work. It is read-only (Read/Glob/Grep/Bash only), produces qualitative assessments (no scores), and acts as an **advisor** — its concerns surface to you, but the builder owns the final lifecycle verdict.
+
+Enable it by adding one line to `.reap/config.yml`:
+
+```yaml
+evaluator: true   # default: false
+```
+
+When enabled, the validation stage instructs the builder to launch `reap-evaluate` as a subagent before declaring pass/partial/fail. The evaluator:
+
+- runs typecheck, build, and the full test suite independently,
+- cross-checks completion criteria from `02-planning.md` against the implementation,
+- surfaces concerns about genome convention drift, sycophancy red flags, and regression risk,
+- escalates per a confidence × impact matrix.
+
+If the subagent invocation fails for any reason, the builder continues normal validation — the evaluator is opt-in advice, not a gate.
+
+Agent definitions are installed automatically by `reap install-skills` **and** `reap update`:
+- Claude Code → `~/.claude/agents/reap-*.md`
+- OpenCode → `~/.config/opencode/agent/reap-*.md`
+
+**Fitness phase + cruise mode** (gen-067): the evaluator also runs during the fitness phase. After receiving its reply, the builder persists the verdict on the generation state via `reap run validation --phase report-evaluator --severity <high|low|none> --summary "..."`. High-severity concerns recorded during validation **automatically abort cruise mode** when the next fitness phase runs — `cruiseCount` is cleared from `config.yml`, the cruise prompt is replaced with a supervised fallback, and the human reviews the concern before composing fitness feedback. Cruise can be resumed manually with `reap cruise <N>` once the concern is resolved. Low-severity concerns surface in the prompt's "Prior Evaluator Concerns" section without aborting cruise.
+
 ## Project Structure
 
 ```
@@ -290,6 +316,7 @@ strictEdit: false # Restrict code changes to REAP lifecycle
 strictMerge: false # Restrict direct git pull/push/merge
 agentClient: claude-code # AI agent client
 # cruiseCount: 1/5             # Present = cruise mode (current/total)
+# evaluator: true              # Opt-in: launch reap-evaluate during validation
 ```
 
 Key settings:
@@ -298,6 +325,7 @@ Key settings:
 - **`strictEdit`**: Restricts code changes to the implementation stage within the planned scope.
 - **`strictMerge`**: Restricts direct git pull/push/merge — use `/reap.pull`, `/reap.push`, `/reap.merge` instead.
 - **`agentClient`**: Determines which adapter is used for skill deployment.
+- **`evaluator`**: Opt-in independent reviewer. When `true`, the validation stage launches the `reap-evaluate` subagent as an advisor (read-only, qualitative-only). Default `false` keeps validation byte-identical to pre-gen-066 behaviour. See [Evaluator Agent](#evaluator-agent-opt-in) above.
 
 ## Upgrading from v0.15 [↗](https://reap.cc/docs/migration-guide)
 

package/RELEASE_NOTICE.md

@@ -1,5 +1,11 @@
 # Release Notices
 
+## v0.16.6
+### en
+Evaluator Agent integration (opt-in `evaluator: true`): independent `reap-evaluate` subagent runs during validation and fitness phases as an advisor. High-severity concerns automatically abort cruise mode for human review.
+### ko
+Evaluator Agent 통합 (opt-in `evaluator: true`): `reap-evaluate` 서브에이전트가 validation/fitness 단계에서 독립 검토자로 실행됨 (advisor 모델). high-severity concern 발생 시 cruise mode 자동 중단.
+
 ## v0.16.4
 ### en
 Restore missing npm metadata (license, author, repository, homepage, keywords). Fix GitHub Releases showing empty release notes.

package/dist/cli/index.js

@@ -8984,6 +8984,116 @@ function buildBasePrompt(knowledge, paths, state, cruiseCount, clarityResult, co
   return lines.join(`
 `);
 }
+function buildEvaluatorPrompt(knowledge, paths, state, opts) {
+  const lines = [];
+  const isMerge = state?.type === "merge";
+  lines.push(`## Evaluator Invocation — ${opts.stage} stage`);
+  lines.push("");
+  if (opts.stage === "validation") {
+    lines.push("You are being invoked during the **validation** stage of a REAP generation. " + "The builder (reap-evolve) has completed implementation and run validation checks. " + "Your role is independent verification: confirm the build is sound, tests pass, " + "completion criteria from 02-planning.md are met, and the code changes follow " + "genome conventions. Surface any concern the builder may have missed.");
+  } else {
+    lines.push("You are being invoked during the **fitness** phase of completion. " + "Assess the generation against the 6 fitness dimensions in your agent definition " + "(goal achievement, code quality, regression safety, artifact quality, vision alignment, " + "cross-generation coherence). Cross-reference with vision goals and recent lineage.");
+  }
+  lines.push("");
+  lines.push("## Current Generation");
+  if (state) {
+    lines.push(`- ID: ${state.id}`);
+    lines.push(`- Type: ${state.type}`);
+    lines.push(`- Goal: ${state.goal}`);
+    lines.push(`- Stage: ${state.stage}`);
+    if (isMerge) {
+      lines.push(`- Parents: ${state.parents.join(", ")}`);
+      if (state.commonAncestor)
+        lines.push(`- Common Ancestor: ${state.commonAncestor}`);
+    }
+  } else {
+    lines.push("- (no active generation — abort and report to the user)");
+  }
+  lines.push("");
+  lines.push("## Artifacts to Read");
+  lines.push("");
+  if (isMerge) {
+    lines.push(`- ${paths.artifact("01-detect.md")}`);
+    lines.push(`- ${paths.artifact("02-mate.md")}`);
+    lines.push(`- ${paths.artifact("03-merge.md")}`);
+    lines.push(`- ${paths.artifact("04-reconcile.md")}`);
+    lines.push(`- ${paths.artifact("05-validation.md")}`);
+  } else {
+    lines.push(`- ${paths.artifact("01-learning.md")}`);
+    lines.push(`- ${paths.artifact("02-planning.md")}`);
+    lines.push(`- ${paths.artifact("03-implementation.md")}`);
+    lines.push(`- ${paths.artifact("04-validation.md")}`);
+  }
+  lines.push("");
+  if (knowledge.visionGoals) {
+    lines.push("## Vision Goals (current)");
+    lines.push(knowledge.visionGoals);
+    lines.push("");
+  }
+  if (knowledge.memoryShortterm || knowledge.memoryMidterm) {
+    lines.push("## Memory");
+    if (knowledge.memoryShortterm) {
+      lines.push("### Shortterm (1-2 sessions)");
+      lines.push(knowledge.memoryShortterm);
+      lines.push("");
+    }
+    if (knowledge.memoryMidterm) {
+      lines.push("### Midterm (multi-generation)");
+      lines.push(knowledge.memoryMidterm);
+      lines.push("");
+    }
+  }
+  lines.push("## Verification Tasks");
+  lines.push("");
+  if (opts.stage === "validation") {
+    lines.push("1. Run `npm run typecheck` (or the project's typecheck command) and record the result.");
+    lines.push("2. Run `npm run build` (or the project's build command) and record the result.");
+    lines.push("3. Run the full test suite (unit/e2e/scenario as defined by the project) and record each.");
+    lines.push("4. Read 02-planning.md and verify each completion criterion against the implementation + validation artifacts.");
+    lines.push("5. Read `git diff` against the parent commit and check for genome convention compliance.");
+    lines.push('6. Check 04-validation.md for sycophancy red flags ("it will probably pass", "it passed before").');
+    lines.push("");
+  } else {
+    lines.push("1. Read all completed artifacts and validate goal achievement.");
+    lines.push("2. Cross-reference the implementation against genome conventions (application.md, evolution.md).");
+    lines.push("3. Run the test suite and confirm no regression.");
+    lines.push("4. Compare this generation's contribution against vision/goals.md and recent lineage.");
+    lines.push("5. Assess each of the 6 fitness dimensions qualitatively (no scores).");
+    lines.push("");
+  }
+  lines.push("## Output Format");
+  lines.push("");
+  lines.push("Apply the escalation matrix from your agent definition:");
+  lines.push("");
+  lines.push("| Confidence | Impact | Action |");
+  lines.push("| --- | --- | --- |");
+  lines.push("| High | Low | Direct judgment |");
+  lines.push("| High | High | Escalate with judgment |");
+  lines.push("| Low | Any | Escalate without judgment (facts only) |");
+  lines.push("");
+  lines.push("Structure your reply as:");
+  lines.push("- **Summary** (1-2 sentences)");
+  lines.push("- **Verification results** (typecheck / build / tests / criteria)");
+  lines.push("- **Concerns** (if any — call out the severity)");
+  lines.push("- **Recommendation** (per the matrix)");
+  lines.push("");
+  lines.push("## HARD-GATE — Evaluator Constraints");
+  lines.push("");
+  lines.push("- You MUST NOT write, edit, or create any source files. Use Read/Grep/Glob/Bash only.");
+  lines.push("- You MUST NOT run git commands that modify state (`git commit`, `git push`, `git checkout`, `git reset`).");
+  lines.push("- You MUST NOT run `reap run` commands — the lifecycle is the builder's responsibility.");
+  lines.push("- You MUST NOT produce numerical scores, ratings, or percentages (Goodhart's Law).");
+  lines.push("- Your verdict is an **advisor recommendation**, not a binding judgment. The builder owns the lifecycle verdict; the human owns final fitness.");
+  lines.push("");
+  lines.push("## If You Cannot Proceed");
+  lines.push("");
+  lines.push("If you cannot complete verification (missing tools, broken state, ambiguous artifacts), " + "report the obstacle in your reply with enough context for the human to act. Do not block " + "the builder's lifecycle — the builder will continue validation if you cannot.");
+  lines.push("");
+  lines.push("## Project");
+  lines.push(`Path: ${paths.root}`);
+  return lines.join(`
+`);
+}
 
 // src/core/dump-state-sync.ts
 var DUMP_COMMANDS = new Set([
@@ -10604,7 +10714,7 @@ async function execute6() {
 }
 
 // src/cli/commands/run/index.ts
-var import_yaml17 = __toESM(require_dist(), 1);
+var import_yaml18 = __toESM(require_dist(), 1);
 init_fs();
 init_integrity();
 
@@ -11865,6 +11975,10 @@ async function execute10(paths, phase) {
   }
 }
 
+// src/cli/commands/run/validation.ts
+var import_yaml11 = __toESM(require_dist(), 1);
+init_fs();
+
 // src/core/artifact-check.ts
 init_fs();
 var PRE_VALIDATION_NORMAL = {
@@ -11899,7 +12013,7 @@ function isUnfilled(content) {
 }
 
 // src/cli/commands/run/validation.ts
-async function execute11(paths, phase) {
+async function execute11(paths, phase, extra) {
   const gm = new GenerationManager(paths);
   const state = await gm.current();
   if (!state)
@@ -11908,6 +12022,64 @@ async function execute11(paths, phase) {
     emitError("validation", `Current stage is '${state.stage}', not 'validation'.`);
   const isMerge = state.type === "merge";
   const s = state;
+  if (phase === "report-evaluator") {
+    let severity;
+    let summary;
+    if (extra) {
+      try {
+        const parsed = JSON.parse(extra);
+        severity = parsed.severity;
+        summary = parsed.summary;
+      } catch {
+        emitError("validation", "report-evaluator: failed to parse options. Expected --severity and --summary.");
+      }
+    }
+    if (!severity) {
+      emitError("validation", "report-evaluator requires --severity <high|low|none>.");
+    }
+    const sev = severity.toLowerCase();
+    if (sev === "none") {
+      emitOutput({
+        status: "ok",
+        command: "validation",
+        phase: "report-evaluator",
+        completed: ["gate", "noop"],
+        context: { id: s.id, severity: "none" },
+        message: "Evaluator reported no concern — state unchanged."
+      });
+      return;
+    }
+    if (sev !== "high" && sev !== "low") {
+      emitError("validation", `report-evaluator: invalid severity '${severity}'. Use high, low, or none.`);
+    }
+    if (!summary || summary.trim().length === 0) {
+      emitError("validation", 'report-evaluator requires --summary "<one-line description>".');
+    }
+    const concern = {
+      stage: "validation",
+      severity: sev,
+      summary: summary.trim(),
+      recordedAt: new Date().toISOString()
+    };
+    if (!s.evaluatorConcerns)
+      s.evaluatorConcerns = [];
+    s.evaluatorConcerns.push(concern);
+    await gm.save(s);
+    emitOutput({
+      status: "ok",
+      command: "validation",
+      phase: "report-evaluator",
+      completed: ["gate", "concern-recorded"],
+      context: {
+        id: s.id,
+        severity: concern.severity,
+        summary: concern.summary,
+        total: s.evaluatorConcerns.length
+      },
+      message: `Evaluator concern recorded (severity=${concern.severity}). Total: ${s.evaluatorConcerns.length}.`
+    });
+    return;
+  }
   if (!phase || phase === "work") {
     verifyTransition("validation", s, "validation:entry");
     await copyArtifactTemplate("validation", paths.artifact, isMerge);
@@ -11944,48 +12116,92 @@ async function execute11(paths, phase) {
       });
       return;
     }
+    const basePromptLines = [
+      "## Validation Stage",
+      "",
+      "### HARD-GATE:",
+      "- Do NOT declare 'pass' without running the validation commands.",
+      "- Do NOT reuse results from a previous run — execute them FRESH.",
+      "- 'It will probably pass' is NOT validation.",
+      "",
+      "### Steps:",
+      "1. **TypeCheck**: Run `npm run typecheck` (or project's typecheck command). Record result.",
+      "2. **Build**: Run `npm run build` (or project's build command). Record result.",
+      "3. **Tests**: Run ALL test commands the project has (e.g., e2e scripts). Record each result.",
+      "4. **Completion Criteria**: Verify EACH criterion from 02-planning.md one by one.",
+      "5. **Minor Fix** (trivial issues only, under 5 minutes): Fix and re-run the failed command.",
+      "6. **Verdict**: Determine pass / partial / fail.",
+      "",
+      "### Red Flags (sycophancy prevention):",
+      "- 'It will probably pass' → Run it.",
+      "- 'It passed before' → Run it again.",
+      "- 'It\\'s trivial, no need to test' → Test it anyway.",
+      "",
+      "### Verdict Criteria:",
+      "- **pass**: All checks pass, all completion criteria met.",
+      "- **partial**: Minor issues remain but core functionality works. Document what's incomplete.",
+      "- **fail**: Critical failures. Must regress to implementation.",
+      "",
+      `### Artifact: Write \`.reap/life/${isMerge ? "05" : "04"}-validation.md\` progressively (after each command).`,
+      "",
+      "If pass/partial: reap run validation --phase complete",
+      "If fail: reap run back to regress"
+    ];
+    const configContent = await readTextFile(paths.config);
+    const config = configContent ? import_yaml11.default.parse(configContent) : null;
+    const evaluatorEnabled = config?.evaluator === true;
+    const context = {
+      id: s.id,
+      goal: s.goal,
+      type: s.type,
+      artifactPath: paths.artifact(isMerge ? "05-validation.md" : "04-validation.md"),
+      evaluator: { enabled: evaluatorEnabled }
+    };
+    const promptLines = [...basePromptLines];
+    if (evaluatorEnabled) {
+      const knowledge = await loadReapKnowledge(paths);
+      const evaluatorPrompt = buildEvaluatorPrompt(knowledge, paths, s, { stage: "validation" });
+      context.evaluator.prompt = evaluatorPrompt;
+      promptLines.push("");
+      promptLines.push("### Evaluator Subagent Invocation (opt-in via `evaluator: true`)");
+      promptLines.push("");
+      promptLines.push("Before declaring your verdict, launch an independent reviewer using the Agent tool:");
+      promptLines.push("");
+      promptLines.push("- subagent_type: `reap-evaluate`");
+      promptLines.push("- description: independent validation review");
+      promptLines.push("- prompt: the `evaluator.prompt` value from the context above");
+      promptLines.push("");
+      promptLines.push("**Advisor model** — the evaluator's assessment is a recommendation, not a verdict:");
+      promptLines.push("- You (the builder) decide the final pass/partial/fail verdict.");
+      promptLines.push("- Surface every evaluator concern to the user in your validation report, even if you disagree.");
+      promptLines.push("- If the evaluator escalates a high-impact concern, lean toward `partial` and document the concern in 04-validation.md.");
+      promptLines.push("");
+      promptLines.push("**Persist the verdict for the fitness phase (gen-067)**:");
+      promptLines.push("");
+      promptLines.push("After receiving the evaluator's reply, record the outcome on the generation state so the");
+      promptLines.push("subsequent fitness phase can act on it (cruise mode auto-abort, evaluator concern surfacing):");
+      promptLines.push("");
+      promptLines.push("- High-impact escalation:");
+      promptLines.push('  `reap run validation --phase report-evaluator --severity high --summary "<one-line description>"`');
+      promptLines.push("- Low-impact concern (informational):");
+      promptLines.push('  `reap run validation --phase report-evaluator --severity low --summary "<one-line description>"`');
+      promptLines.push("- Clean review (no concern):");
+      promptLines.push('  `reap run validation --phase report-evaluator --severity none --summary ""`');
+      promptLines.push("");
+      promptLines.push("This call does NOT advance the lifecycle — it only appends to `state.evaluatorConcerns`.");
+      promptLines.push("");
+      promptLines.push("**Fallback** — if the evaluator subagent fails (tool unavailable, model error, malformed reply):");
+      promptLines.push("- Tell the user the evaluator could not run and why.");
+      promptLines.push("- Continue normal validation. The evaluator is opt-in advice, not a gate.");
+      promptLines.push("- Skip the `report-evaluator` CLI call (no concern was generated).");
+    }
     emitOutput({
       status: "prompt",
       command: "validation",
       phase: "work",
       completed: ["gate", "artifact-check"],
-      context: {
-        id: s.id,
-        goal: s.goal,
-        type: s.type,
-        artifactPath: paths.artifact(isMerge ? "05-validation.md" : "04-validation.md")
-      },
-      prompt: [
-        "## Validation Stage",
-        "",
-        "### HARD-GATE:",
-        "- Do NOT declare 'pass' without running the validation commands.",
-        "- Do NOT reuse results from a previous run — execute them FRESH.",
-        "- 'It will probably pass' is NOT validation.",
-        "",
-        "### Steps:",
-        "1. **TypeCheck**: Run `npm run typecheck` (or project's typecheck command). Record result.",
-        "2. **Build**: Run `npm run build` (or project's build command). Record result.",
-        "3. **Tests**: Run ALL test commands the project has (e.g., e2e scripts). Record each result.",
-        "4. **Completion Criteria**: Verify EACH criterion from 02-planning.md one by one.",
-        "5. **Minor Fix** (trivial issues only, under 5 minutes): Fix and re-run the failed command.",
-        "6. **Verdict**: Determine pass / partial / fail.",
-        "",
-        "### Red Flags (sycophancy prevention):",
-        "- 'It will probably pass' → Run it.",
-        "- 'It passed before' → Run it again.",
-        "- 'It\\'s trivial, no need to test' → Test it anyway.",
-        "",
-        "### Verdict Criteria:",
-        "- **pass**: All checks pass, all completion criteria met.",
-        "- **partial**: Minor issues remain but core functionality works. Document what's incomplete.",
-        "- **fail**: Critical failures. Must regress to implementation.",
-        "",
-        `### Artifact: Write \`.reap/life/${isMerge ? "05" : "04"}-validation.md\` progressively (after each command).`,
-        "",
-        "If pass/partial: reap run validation --phase complete",
-        "If fail: reap run back to regress"
-      ].join(`
+      context,
+      prompt: promptLines.join(`
 `),
       nextCommand: "reap run validation --phase complete"
     });
@@ -12013,13 +12229,13 @@ init_fs();
 
 // src/core/archive.ts
 init_fs();
-var import_yaml12 = __toESM(require_dist(), 1);
+var import_yaml13 = __toESM(require_dist(), 1);
 import { join as join16 } from "path";
 import { readdir as readdir13, rm as rm4, unlink } from "fs/promises";
 
 // src/core/compression.ts
 init_fs();
-var import_yaml11 = __toESM(require_dist(), 1);
+var import_yaml12 = __toESM(require_dist(), 1);
 import { readdir as readdir12, rm as rm3 } from "fs/promises";
 import { join as join15 } from "path";
 var LEVEL1_THRESHOLD = 20;
@@ -12052,7 +12268,7 @@ async function compressLineage(lineageDir) {
     let meta = { id: dirName, type: "unknown", goal: "", parents: [] };
     if (metaContent) {
       try {
-        meta = import_yaml11.default.parse(metaContent);
+        meta = import_yaml12.default.parse(metaContent);
       } catch {}
     }
     const completion = await readTextFile(join15(dirPath, "05-completion.md")) ?? "";
@@ -12107,7 +12323,7 @@ async function compressToEpoch(lineageDir) {
     let meta = { id: file.replace(".md", ""), type: "unknown", goal: "", parents: [] };
     if (fmMatch) {
       try {
-        const parsed = import_yaml11.default.parse(fmMatch[1]);
+        const parsed = import_yaml12.default.parse(fmMatch[1]);
         meta = { ...meta, ...parsed };
       } catch {}
     }
@@ -12124,7 +12340,7 @@ async function compressToEpoch(lineageDir) {
     if (!metaContent)
       continue;
     try {
-      const m = import_yaml11.default.parse(metaContent);
+      const m = import_yaml12.default.parse(metaContent);
       if (m.parents)
         m.parents.forEach((p) => dirParents.add(p));
     } catch {}
@@ -12137,7 +12353,7 @@ async function compressToEpoch(lineageDir) {
     const fmMatch = content.match(/^---\n([\s\S]*?)\n---/);
     if (fmMatch) {
       try {
-        const parsed = import_yaml11.default.parse(fmMatch[1]);
+        const parsed = import_yaml12.default.parse(fmMatch[1]);
         const lastId = parsed.lastGeneration;
         if (lastId)
           dirParents.add(lastId);
@@ -12219,7 +12435,7 @@ async function writeArchive(paths, state, archiveDir, extraMeta) {
     timeline: state.timeline,
     ...extraMeta
   };
-  await writeTextFile(join16(archiveDir, "meta.yml"), import_yaml12.default.stringify(meta));
+  await writeTextFile(join16(archiveDir, "meta.yml"), import_yaml13.default.stringify(meta));
   for (const entry of lifeEntries) {
     if (entry === "backlog")
       continue;
@@ -12544,7 +12760,7 @@ function tokenize(text) {
 
 // src/core/cruise.ts
 init_fs();
-var import_yaml13 = __toESM(require_dist(), 1);
+var import_yaml14 = __toESM(require_dist(), 1);
 function parseCruiseCount(config) {
   if (!config.cruiseCount)
     return null;
@@ -12557,30 +12773,38 @@ async function advanceCruise(configPath) {
   const content = await readTextFile(configPath);
   if (!content)
     return false;
-  const config = import_yaml13.default.parse(content);
+  const config = import_yaml14.default.parse(content);
   const cruise = parseCruiseCount(config);
   if (!cruise)
     return false;
   if (cruise.current >= cruise.total) {
     delete config.cruiseCount;
-    await writeTextFile(configPath, import_yaml13.default.stringify(config));
+    await writeTextFile(configPath, import_yaml14.default.stringify(config));
     return false;
   }
   config.cruiseCount = `${cruise.current + 1}/${cruise.total}`;
-  await writeTextFile(configPath, import_yaml13.default.stringify(config));
+  await writeTextFile(configPath, import_yaml14.default.stringify(config));
   return true;
 }
+async function clearCruise(configPath) {
+  const content = await readTextFile(configPath);
+  if (!content)
+    return;
+  const config = import_yaml14.default.parse(content);
+  delete config.cruiseCount;
+  await writeTextFile(configPath, import_yaml14.default.stringify(config));
+}
 async function setCruise(configPath, total) {
   const content = await readTextFile(configPath);
   if (!content)
     return;
-  const config = import_yaml13.default.parse(content);
+  const config = import_yaml14.default.parse(content);
   config.cruiseCount = `1/${total}`;
-  await writeTextFile(configPath, import_yaml13.default.stringify(config));
+  await writeTextFile(configPath, import_yaml14.default.stringify(config));
 }
 
 // src/cli/commands/run/completion.ts
-var import_yaml14 = __toESM(require_dist(), 1);
+var import_yaml15 = __toESM(require_dist(), 1);
 async function execute12(paths, phase, feedback) {
   const gm = new GenerationManager(paths);
   const state = await gm.current();
@@ -12669,52 +12893,145 @@ async function execute12(paths, phase, feedback) {
     } else {
       verifyTransition("completion", s, "completion:fitness");
       const configContent = await readTextFile(paths.config);
-      const config = configContent ? import_yaml14.default.parse(configContent) : null;
+      const config = configContent ? import_yaml15.default.parse(configContent) : null;
       const cruise = config ? parseCruiseCount(config) : null;
+      const evaluatorEnabled = config?.evaluator === true;
       setTransitionNonces(s, "completion:fitness");
       await gm.save(s);
+      const highConcerns = (s.evaluatorConcerns ?? []).filter((c) => c.severity === "high");
+      let fitnessEvaluatorPrompt;
+      if (evaluatorEnabled) {
+        const knowledge = await loadReapKnowledge(paths);
+        fitnessEvaluatorPrompt = buildEvaluatorPrompt(knowledge, paths, s, { stage: "fitness" });
+      }
+      const evaluatorSection = [];
+      if (evaluatorEnabled) {
+        evaluatorSection.push("");
+        evaluatorSection.push("### Evaluator Subagent Invocation (opt-in via `evaluator: true`)");
+        evaluatorSection.push("");
+        evaluatorSection.push("Before composing the final fitness feedback, launch the independent reviewer using the Agent tool:");
+        evaluatorSection.push("");
+        evaluatorSection.push("- subagent_type: `reap-evaluate`");
+        evaluatorSection.push("- description: independent fitness assessment");
+        evaluatorSection.push("- prompt: the `evaluator.prompt` value from the context above");
+        evaluatorSection.push("");
+        evaluatorSection.push("The evaluator scores the generation along the 6 fitness dimensions (qualitative — no numeric scores).");
+        evaluatorSection.push("Surface every concern to the user in your fitness summary; the human owns the final feedback.");
+        evaluatorSection.push("");
+        evaluatorSection.push("**Fallback** — if the evaluator subagent fails, document the failure and continue. Fitness phase is not gated on the evaluator.");
+      }
+      const priorConcernsSection = [];
+      if ((s.evaluatorConcerns ?? []).length > 0) {
+        priorConcernsSection.push("");
+        priorConcernsSection.push("### Prior Evaluator Concerns (carried from earlier stages)");
+        priorConcernsSection.push("");
+        for (const c of s.evaluatorConcerns) {
+          priorConcernsSection.push(`- [${c.severity}] (${c.stage}) ${c.summary}`);
+        }
+        priorConcernsSection.push("");
+        priorConcernsSection.push("Surface each of the above to the user with your fitness summary.");
+      }
+      if (cruise && highConcerns.length > 0) {
+        await clearCruise(paths.config);
+        const fallbackPrompt = [
+          "## Cruise Aborted by Evaluator Concern",
+          "",
+          "Cruise mode has been **disengaged** because the evaluator raised a high-impact concern:",
+          ""
+        ];
+        for (const c of highConcerns) {
+          fallbackPrompt.push(`- [${c.severity}] (${c.stage}) ${c.summary}`);
+        }
+        fallbackPrompt.push("");
+        fallbackPrompt.push("The remaining cruise generations will NOT auto-start. This fitness phase now follows the supervised flow:");
+        fallbackPrompt.push("");
+        fallbackPrompt.push("1. Present the concern(s) above to the human together with a summary of this generation.");
+        fallbackPrompt.push("2. Wait for the human's decision (continue, override, abort generation, etc.).");
+        fallbackPrompt.push('3. Submit: `reap run completion --phase fitness --feedback "<human feedback>"`');
+        fallbackPrompt.push("");
+        fallbackPrompt.push("Cruise can be resumed manually with `reap cruise <N>` once the concern is resolved.");
+        fallbackPrompt.push(...evaluatorSection);
+        emitOutput({
+          status: "prompt",
+          command: "completion",
+          phase: "fitness",
+          completed: ["gate", "reflect", "cruise-aborted"],
+          context: {
+            id: s.id,
+            goal: s.goal,
+            cruiseMode: false,
+            cruiseAborted: true,
+            previousCruiseCount: config.cruiseCount,
+            evaluatorConcerns: s.evaluatorConcerns,
+            evaluator: evaluatorEnabled ? { enabled: true, prompt: fitnessEvaluatorPrompt } : { enabled: false }
+          },
+          prompt: fallbackPrompt.join(`
+`),
+          nextCommand: "reap run completion --phase fitness"
+        });
+        return;
+      }
       if (cruise) {
+        const cruisePrompt = [
+          "## Completion — Fitness Phase (Cruise Mode)",
+          "",
+          `Cruise: ${config.cruiseCount}`,
+          "",
+          "### Self-Assessment (not self-fitness, but metacognition):",
+          "1. Did this generation proceed as expected?",
+          "2. Are there uncertain areas or risks?",
+          "3. Are there items that need human confirmation?",
+          "",
+          'High confidence → auto-proceed: reap run completion --phase fitness --feedback "self-assessment: OK"',
+          "Uncertain/risky → stop cruise and request human feedback"
+        ];
+        cruisePrompt.push(...priorConcernsSection);
+        cruisePrompt.push(...evaluatorSection);
         emitOutput({
           status: "prompt",
           command: "completion",
           phase: "fitness",
           completed: ["gate", "reflect"],
-          context: { id: s.id, goal: s.goal, cruiseMode: true, cruiseCount: config.cruiseCount },
-          prompt: [
-            "## Completion — Fitness Phase (Cruise Mode)",
-            "",
-            `Cruise: ${config.cruiseCount}`,
-            "",
-            "### Self-Assessment (not self-fitness, but metacognition):",
-            "1. Did this generation proceed as expected?",
-            "2. Are there uncertain areas or risks?",
-            "3. Are there items that need human confirmation?",
-            "",
-            'High confidence → auto-proceed: reap run completion --phase fitness --feedback "self-assessment: OK"',
-            "Uncertain/risky → stop cruise and request human feedback"
-          ].join(`
+          context: {
+            id: s.id,
+            goal: s.goal,
+            cruiseMode: true,
+            cruiseCount: config.cruiseCount,
+            evaluatorConcerns: s.evaluatorConcerns ?? [],
+            evaluator: evaluatorEnabled ? { enabled: true, prompt: fitnessEvaluatorPrompt } : { enabled: false }
+          },
+          prompt: cruisePrompt.join(`
 `),
           nextCommand: "reap run completion --phase fitness"
         });
       } else {
+        const supervisedPrompt = [
+          "## Completion — Fitness Phase",
+          "",
+          "Collect feedback from the human.",
+          "",
+          "Present to the human:",
+          "1. Summary of what was done in this generation",
+          "2. What went well / areas for improvement",
+          "3. Suggested next direction",
+          "",
+          'Submit: reap run completion --phase fitness --feedback "human feedback here"'
+        ];
+        supervisedPrompt.push(...priorConcernsSection);
+        supervisedPrompt.push(...evaluatorSection);
         emitOutput({
           status: "prompt",
           command: "completion",
           phase: "fitness",
           completed: ["gate", "reflect"],
-          context: { id: s.id, goal: s.goal, cruiseMode: false },
-          prompt: [
-            "## Completion — Fitness Phase",
-            "",
-            "Collect feedback from the human.",
-            "",
-            "Present to the human:",
-            "1. Summary of what was done in this generation",
-            "2. What went well / areas for improvement",
-            "3. Suggested next direction",
-            "",
-            'Submit: reap run completion --phase fitness --feedback "human feedback here"'
-          ].join(`
+          context: {
+            id: s.id,
+            goal: s.goal,
+            cruiseMode: false,
+            evaluatorConcerns: s.evaluatorConcerns ?? [],
+            evaluator: evaluatorEnabled ? { enabled: true, prompt: fitnessEvaluatorPrompt } : { enabled: false }
+          },
+          prompt: supervisedPrompt.join(`
 `),
           nextCommand: "reap run completion --phase fitness"
         });
@@ -12726,7 +13043,7 @@ async function execute12(paths, phase, feedback) {
     const fitnessFeedback = s.fitnessFeedback;
     const visionGoals = await readTextFile(paths.visionGoals);
     const configContent = await readTextFile(paths.config);
-    const config = configContent ? import_yaml14.default.parse(configContent) : null;
+    const config = configContent ? import_yaml15.default.parse(configContent) : null;
     const maturity = detectMaturity(s.type, config?.cruiseCount);
     const generationCount = await gm.countLineage();
     setTransitionNonces(s, "completion:adapt");
@@ -12859,7 +13176,7 @@ async function execute12(paths, phase, feedback) {
 }
 
 // src/cli/commands/run/evolve.ts
-var import_yaml15 = __toESM(require_dist(), 1);
+var import_yaml16 = __toESM(require_dist(), 1);
 import { readdir as readdir14 } from "fs/promises";
 init_fs();
 async function collectClarityInput(paths, generationType) {
@@ -12896,7 +13213,7 @@ async function execute13(paths, _phase) {
   const gm = new GenerationManager(paths);
   const state = await gm.current();
   const configContent = await readTextFile(paths.config);
-  const config = configContent ? import_yaml15.default.parse(configContent) : null;
+  const config = configContent ? import_yaml16.default.parse(configContent) : null;
   const autoSubagent = config?.autoSubagent ?? true;
   const cruise = config ? parseCruiseCount(config) : null;
   const generationType = state?.type ?? "normal";
@@ -13972,19 +14289,19 @@ async function execute24(paths, _phase, extra) {
 
 // src/cli/commands/run/report.ts
 init_fs();
-var import_yaml16 = __toESM(require_dist(), 1);
+var import_yaml17 = __toESM(require_dist(), 1);
 async function execute25(paths, _phase) {
   let state = null;
   try {
     const raw = await readTextFile(paths.current);
     if (raw)
-      state = import_yaml16.default.parse(raw);
+      state = import_yaml17.default.parse(raw);
   } catch {}
   let projectName = "unknown";
   try {
     const raw = await readTextFile(paths.config);
     if (raw) {
-      const config = import_yaml16.default.parse(raw);
+      const config = import_yaml17.default.parse(raw);
       projectName = config.project ?? "unknown";
     }
   } catch {}
@@ -14081,13 +14398,16 @@ async function execute26(stage, options) {
   if (stage === "early-close") {
     extra = JSON.stringify({ reason: options.reason, sourceAction: options.sourceAction, deferTasks: options.deferTasks });
   }
+  if (stage === "validation" && options.phase === "report-evaluator") {
+    extra = JSON.stringify({ severity: options.severity, summary: options.summary });
+  }
   try {
     await handler(paths, options.phase, extra);
   } catch (err) {
     try {
       const configContent = await readTextFile(paths.config);
       if (configContent) {
-        const config = import_yaml17.default.parse(configContent);
+        const config = import_yaml18.default.parse(configContent);
         if (config.autoIssueReport !== false) {
           const cmd = `reap run ${stage}${options.phase ? ` --phase ${options.phase}` : ""}`;
           autoReport(cmd, err);
@@ -14230,7 +14550,7 @@ async function execute28(count) {
 }
 
 // src/cli/commands/install-skills.ts
-var import_yaml18 = __toESM(require_dist(), 1);
+var import_yaml19 = __toESM(require_dist(), 1);
 
 // src/adapters/claude-code/install.ts
 init_fs();
@@ -14280,18 +14600,35 @@ async function installSkills(_projectRoot) {
     message: `Cleaned ${cleaned.length} stale skills, installed ${installed} skill files to ${targetDir}`
   });
 }
-async function installAgents() {
-  const agentsDir = join20(homedir4(), ".claude", "agents");
-  await ensureDir(agentsDir);
-  const templateDir = __dirname2.includes("dist") ? join20(__dirname2, "..", "templates", "agents") : join20(__dirname2, "..", "..", "templates", "agents");
+var AGENT_PATTERN = /^reap-.+\.md$/;
+function agentsTemplateDir() {
+  return __dirname2.includes("dist") ? join20(__dirname2, "..", "templates", "agents") : join20(__dirname2, "..", "..", "templates", "agents");
+}
+async function installAgents(home = homedir4()) {
+  const targetDir = join20(home, ".claude", "agents");
+  await ensureDir(targetDir);
+  let cleaned = [];
   try {
-    const files = await readdir16(templateDir);
-    for (const file of files) {
-      if (file.endsWith(".md")) {
-        await cp2(join20(templateDir, file), join20(agentsDir, file));
-      }
+    const existing = await readdir16(targetDir);
+    cleaned = existing.filter((f) => AGENT_PATTERN.test(f));
+    for (const file of cleaned) {
+      await unlink2(join20(targetDir, file));
+    }
+  } catch {}
+  let installed = 0;
+  const files = [];
+  const templateDir = agentsTemplateDir();
+  try {
+    const sources = await readdir16(templateDir);
+    for (const file of sources) {
+      if (!file.endsWith(".md"))
+        continue;
+      await cp2(join20(templateDir, file), join20(targetDir, file));
+      files.push(file);
+      installed++;
     }
   } catch {}
+  return { cleaned, installed, files, targetDir };
 }
 async function installReapGuide() {
   const reapHome = join20(homedir4(), ".reap");
@@ -14354,6 +14691,7 @@ var claudeCodeAdapter = {
   },
   async registerSessionIntegration(_projectRoot) {
     await installSlashCommandsOnly();
+    await installAgents();
     await registerSessionHooks();
   }
 };
@@ -14559,11 +14897,45 @@ async function installSlashCommands(home = homedir5()) {
   } catch {}
   return { cleaned, installed, targetDir };
 }
+var AGENT_PATTERN2 = /^reap-.+\.md$/;
+function agentsTemplateDir2() {
+  return __dirname3.includes("dist") ? join21(__dirname3, "..", "templates", "agents") : join21(__dirname3, "..", "..", "templates", "agents");
+}
+function opencodeAgentsDir(home = homedir5()) {
+  return join21(home, ".config", "opencode", "agent");
+}
+async function installAgents2(home = homedir5()) {
+  const targetDir = opencodeAgentsDir(home);
+  await ensureDir(targetDir);
+  let cleaned = 0;
+  try {
+    const existing = await readdir17(targetDir);
+    for (const file of existing) {
+      if (AGENT_PATTERN2.test(file)) {
+        await unlink3(join21(targetDir, file));
+        cleaned++;
+      }
+    }
+  } catch {}
+  let installed = 0;
+  const srcDir = agentsTemplateDir2();
+  try {
+    const sources = await readdir17(srcDir);
+    for (const file of sources) {
+      if (!file.endsWith(".md"))
+        continue;
+      await cp3(join21(srcDir, file), join21(targetDir, file));
+      installed++;
+    }
+  } catch {}
+  return { cleaned, installed, targetDir };
+}
 async function installSkills2(projectRoot) {
   await installPluginFile(projectRoot);
   const opencodeJsonAction = await ensureOpencodeJson(projectRoot);
   await installReapGuide2();
   const slashCommands = await installSlashCommands();
+  const agents = await installAgents2();
   emitOutput({
     status: "ok",
     command: "install-skills",
@@ -14571,7 +14943,8 @@ async function installSkills2(projectRoot) {
       "install-plugin",
       "ensure-opencode-json",
       "install-reap-guide",
-      "install-slash-commands"
+      "install-slash-commands",
+      "install-agents"
     ],
     context: {
       agentClient: "opencode",
@@ -14581,15 +14954,21 @@ async function installSkills2(projectRoot) {
         cleaned: slashCommands.cleaned,
         installed: slashCommands.installed,
         targetDir: slashCommands.targetDir
+      },
+      agents: {
+        cleaned: agents.cleaned,
+        installed: agents.installed,
+        targetDir: agents.targetDir
       }
     },
-    message: `OpenCode integration installed (opencode.json: ${opencodeJsonAction}, ` + `slash commands: ${slashCommands.installed} installed, ` + `${slashCommands.cleaned} cleaned).`
+    message: `OpenCode integration installed (opencode.json: ${opencodeJsonAction}, ` + `slash commands: ${slashCommands.installed} installed, ` + `${slashCommands.cleaned} cleaned; ` + `agents: ${agents.installed} installed, ${agents.cleaned} cleaned).`
   });
 }
 async function registerSessionIntegration(projectRoot) {
   await installPluginFile(projectRoot);
   await ensureOpencodeJson(projectRoot);
   await installSlashCommands();
+  await installAgents2();
 }
 
 // src/adapters/opencode/index.ts
@@ -14632,7 +15011,7 @@ async function execute29() {
     const raw = await readTextFile(paths.config);
     if (raw) {
       try {
-        const cfg = import_yaml18.default.parse(raw);
+        const cfg = import_yaml19.default.parse(raw);
         agentClient = cfg.agentClient;
       } catch {}
     }
@@ -14642,7 +15021,7 @@ async function execute29() {
 }
 
 // src/cli/commands/fix.ts
-var import_yaml19 = __toESM(require_dist(), 1);
+var import_yaml20 = __toESM(require_dist(), 1);
 import { mkdir as mkdir3, stat as stat4 } from "fs/promises";
 import { join as join22, dirname as dirname8 } from "path";
 import { fileURLToPath as fileURLToPath6 } from "url";
@@ -14694,7 +15073,7 @@ async function fixProject(projectRoot) {
   if (currentContent !== null) {
     if (currentContent.trim()) {
       try {
-        const state = import_yaml19.default.parse(currentContent);
+        const state = import_yaml20.default.parse(currentContent);
         const allStages = [
           ...LIFECYCLE_STAGES,
           ...MERGE_STAGES
@@ -14761,7 +15140,7 @@ async function fixProject(projectRoot) {
   {
     const { ensureClaudeMd: ensureClaudeMd2 } = await Promise.resolve().then(() => (init_common(), exports_common));
     const configContent = await readTextFile(paths.config);
-    const projectName = configContent ? import_yaml19.default.parse(configContent)?.project ?? "my-project" : "my-project";
+    const projectName = configContent ? import_yaml20.default.parse(configContent)?.project ?? "my-project" : "my-project";
     const action = await ensureClaudeMd2(paths.root, projectName);
     if (action !== "skipped") {
       fixed.push(`CLAUDE.md ${action} with REAP section`);
@@ -14810,7 +15189,7 @@ async function execute30(check) {
 }
 
 // src/cli/commands/destroy.ts
-var import_yaml20 = __toESM(require_dist(), 1);
+var import_yaml21 = __toESM(require_dist(), 1);
 import { rm as rm6 } from "fs/promises";
 import { join as join23 } from "path";
 init_fs();
@@ -14821,7 +15200,7 @@ async function getProjectName(projectRoot) {
     const content = await readTextFile(paths.config);
     if (!content)
       return null;
-    const config = import_yaml20.default.parse(content);
+    const config = import_yaml21.default.parse(content);
     return config.project ?? null;
   } catch {
     return null;
@@ -15069,7 +15448,7 @@ async function execute32(options) {
 
 // src/cli/commands/check-version.ts
 init_integrity();
-var import_yaml21 = __toESM(require_dist(), 1);
+var import_yaml22 = __toESM(require_dist(), 1);
 import { execSync as execSync7 } from "child_process";
 import { readFileSync as readFileSync3 } from "fs";
 import { join as join26 } from "path";
@@ -15251,7 +15630,7 @@ async function execute33() {
     try {
       const configPath = join26(root, ".reap", "config.yml");
       const configContent = readFileSync3(configPath, "utf-8");
-      const config = import_yaml21.default.parse(configContent);
+      const config = import_yaml22.default.parse(configContent);
       const language = config?.language ?? "english";
       const notice = fetchReleaseNotice(result.to, language);
       if (notice)
@@ -15261,7 +15640,7 @@ async function execute33() {
 }
 
 // src/cli/commands/config.ts
-var import_yaml22 = __toESM(require_dist(), 1);
+var import_yaml23 = __toESM(require_dist(), 1);
 init_fs();
 init_integrity();
 async function execute34() {
@@ -15277,7 +15656,7 @@ async function execute34() {
   if (!configContent) {
     emitError("config", "Config file is empty.");
   }
-  const config = import_yaml22.default.parse(configContent);
+  const config = import_yaml23.default.parse(configContent);
   emitOutput({
     status: "ok",
     command: "config",
@@ -15296,7 +15675,7 @@ async function execute34() {
 }
 
 // src/cli/commands/update.ts
-var import_yaml23 = __toESM(require_dist(), 1);
+var import_yaml24 = __toESM(require_dist(), 1);
 import { readFileSync as readFileSync4 } from "fs";
 import { join as join27, dirname as dirname10 } from "path";
 import { fileURLToPath as fileURLToPath7 } from "url";
@@ -15355,7 +15734,7 @@ async function backfillConfig(paths) {
     return { added: [], removed: [] };
   let config;
   try {
-    config = import_yaml23.default.parse(content) ?? {};
+    config = import_yaml24.default.parse(content) ?? {};
   } catch {
     return { added: [], removed: [] };
   }
@@ -15381,7 +15760,7 @@ async function backfillConfig(paths) {
     }
   }
   if (added.length > 0 || removed.length > 0) {
-    await writeTextFile(paths.config, import_yaml23.default.stringify(config));
+    await writeTextFile(paths.config, import_yaml24.default.stringify(config));
   }
   return { added, removed };
 }
@@ -15432,7 +15811,7 @@ async function execute35(phase, postUpgrade) {
   let agentClient;
   if (configContent) {
     try {
-      const config = import_yaml23.default.parse(configContent);
+      const config = import_yaml24.default.parse(configContent);
       projectName = config.project ?? "my-project";
       agentClient = config.agentClient;
     } catch {}
@@ -15449,7 +15828,7 @@ async function execute35(phase, postUpgrade) {
     const raw = configContent ?? "";
     if (raw) {
       try {
-        const cfg = import_yaml23.default.parse(raw);
+        const cfg = import_yaml24.default.parse(raw);
         if (cfg?.language)
           language = cfg.language;
       } catch {}
@@ -15816,7 +16195,7 @@ ${lines.join(`
 }
 
 // src/cli/commands/load-context.ts
-var import_yaml24 = __toESM(require_dist(), 1);
+var import_yaml25 = __toESM(require_dist(), 1);
 import { join as join29 } from "path";
 init_fs();
 async function buildKnowledgeContext(cwd) {
@@ -15832,13 +16211,13 @@ async function buildKnowledgeContext(cwd) {
   let config = null;
   if (configContent) {
     try {
-      config = import_yaml24.default.parse(configContent);
+      config = import_yaml25.default.parse(configContent);
     } catch {}
   }
   let state = null;
   if (currentContent) {
     try {
-      state = import_yaml24.default.parse(currentContent);
+      state = import_yaml25.default.parse(currentContent);
     } catch {}
   }
   const sections = [];
@@ -15951,7 +16330,7 @@ program2.command("init [project-name]").description("Initialize a new reap proje
 program2.command("status").description("Show current project status").action(async () => {
   await execute6();
 });
-program2.command("run <stage>").description("Run a lifecycle stage (start, learning, planning, ...)").option("--phase <phase>", "Stage phase (work, complete, reflect, fitness, adapt, commit)").option("--goal <goal>", "Goal for start command").option("--type <type>", "Generation type (embryo, normal, merge)").option("--parents <parents>", "Parent generation IDs for merge (comma-separated)").option("--feedback <feedback>", "Fitness feedback text").option("--reason <reason>", "Reason for back regression or abort").option("--backlog <backlog>", "Backlog filename to consume for this generation (use --no-backlog to explicitly declare no relevant backlog)").option("--no-backlog", "Explicitly declare no backlog item is relevant to this generation (suppresses pending-backlog prompt). Negates --backlog.").option("--source-action <sourceAction>", "Source action for abort/early-close (rollback, stash, hold, none — rollback only for abort)").option("--save-backlog", "Save progress to backlog on abort").option("--defer-tasks <value>", "For early-close: auto-defer unchecked tasks to new backlog (true|false, default true)").action(async (stage, options) => {
+program2.command("run <stage>").description("Run a lifecycle stage (start, learning, planning, ...)").option("--phase <phase>", "Stage phase (work, complete, reflect, fitness, adapt, commit)").option("--goal <goal>", "Goal for start command").option("--type <type>", "Generation type (embryo, normal, merge)").option("--parents <parents>", "Parent generation IDs for merge (comma-separated)").option("--feedback <feedback>", "Fitness feedback text").option("--reason <reason>", "Reason for back regression or abort").option("--backlog <backlog>", "Backlog filename to consume for this generation (use --no-backlog to explicitly declare no relevant backlog)").option("--no-backlog", "Explicitly declare no backlog item is relevant to this generation (suppresses pending-backlog prompt). Negates --backlog.").option("--source-action <sourceAction>", "Source action for abort/early-close (rollback, stash, hold, none — rollback only for abort)").option("--save-backlog", "Save progress to backlog on abort").option("--defer-tasks <value>", "For early-close: auto-defer unchecked tasks to new backlog (true|false, default true)").option("--severity <severity>", "For validation --phase report-evaluator: evaluator concern severity (high|low|none — none is a no-op)").option("--summary <summary>", "For validation --phase report-evaluator: one-line evaluator concern summary").action(async (stage, options) => {
   await execute26(stage, options);
 });
 program2.command("make <resource>").description("Create a resource from template (backlog, hook)").option("--type <type>", "Resource type (backlog: genome-change/environment-change/task, hook: sh/md)").option("--title <title>", "Resource title").option("--body <body>", "Optional description body").option("--priority <priority>", "Priority (high, medium, low)").option("--event <event>", "Hook event (e.g. onLifeCompleted)").option("--name <name>", "Hook name").option("--condition <condition>", "Hook condition (default: always)").option("--order <order>", "Hook execution order (default: 50)").action(async (resource, options) => {

package/dist/templates/reap-guide.md

@@ -317,6 +317,10 @@ Switching clients is a config edit: change `agentClient` in `.reap/config.yml`,
 
 For OpenCode users, see `AGENTS.md` (auto-generated by `reap update`) for the project-level entry-point. `~/.reap/reap-guide.md` (installed by `reap install-skills`) is referenced from `AGENTS.md` and remains the canonical REAP usage guide.
 
+### Evaluator agent (opt-in)
+
+Both adapters install bundled agent definitions (`reap-evolve.md`, `reap-evaluate.md`) to the client's user-level agents directory (`~/.claude/agents/` for Claude Code, `~/.config/opencode/agent/` for OpenCode). The install runs on `reap install-skills` AND `reap update`, so user-level agents stay in sync with the bundled REAP version. Set `evaluator: true` in `.reap/config.yml` to launch `reap-evaluate` as an independent reviewer during the validation stage (advisor role — the builder owns the final verdict).
+
 ## Role Separation
 
 | Component        | Role                                                                |

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@c-d-cc/reap",
-  "version": "0.16.5",
+  "version": "0.16.6",
   "description": "Recursive Evolutionary Autonomous Pipeline — AI and humans evolve software across generations",
   "license": "MIT",
   "author": "HyeonIL Choi <hichoi@c-d.cc> (C to D)",