cclaw-cli 0.22.0 → 0.23.1

package/dist/cli.d.ts

@@ -24,6 +24,8 @@ interface ParsedArgs {
     evalJudge?: boolean;
     evalJson?: boolean;
     evalNoWrite?: boolean;
+    evalUpdateBaseline?: boolean;
+    evalConfirm?: boolean;
     showHelp?: boolean;
     showVersion?: boolean;
 }

package/dist/cli.js

@@ -14,6 +14,7 @@ import { createDefaultConfig, createProfileConfig } from "./config.js";
 import { detectHarnesses } from "./init-detect.js";
 import { HARNESS_ADAPTERS } from "./harness-adapters.js";
 import { runEval } from "./eval/runner.js";
+import { writeBaselinesFromReport } from "./eval/baseline.js";
 import { writeJsonReport, writeMarkdownReport } from "./eval/report.js";
 import { EVAL_TIERS } from "./eval/types.js";
 import { FLOW_STAGES } from "./types.js";
@@ -53,15 +54,17 @@ Commands:
              Flags: --name=<feature>    Feature slug (default: inferred from 00-idea.md).
                     --skip-retro       Bypass mandatory retro gate (requires --retro-reason).
                     --retro-reason=<t> Reason for bypassing retro gate.
-  eval       Run cclaw evals against .cclaw/evals/corpus (Phase 7, Wave 7.0 foundations).
-             Flags: --stage=<id>        Limit to one flow stage (${FLOW_STAGES.join("|")}).
-                    --tier=<A|B|C>      Fidelity tier (A=single-shot, B=tools, C=workflow).
-                    --schema-only       Run only structural verifiers (Wave 7.1).
-                    --rules             Run structural + rule verifiers (Wave 7.2).
-                    --judge             Include LLM judging (Wave 7.3; requires API key).
-                    --dry-run           Validate config + corpus, print summary, do not execute.
-                    --json              Emit machine-readable JSON on stdout.
-                    --no-write          Skip writing the report to .cclaw/evals/reports/.
+  eval       Run cclaw evals against .cclaw/evals/corpus (Phase 7: structural verifier + baselines).
+             Flags: --stage=<id>         Limit to one flow stage (${FLOW_STAGES.join("|")}).
+                    --tier=<A|B|C>       Fidelity tier (A=single-shot, B=tools, C=workflow).
+                    --schema-only        Run only structural verifiers (default).
+                    --rules              Run structural + rule verifiers (not wired yet).
+                    --judge              Include LLM judging (not wired yet; requires API key).
+                    --dry-run            Validate config + corpus, print summary, do not execute.
+                    --json               Emit machine-readable JSON on stdout.
+                    --no-write           Skip writing the report to .cclaw/evals/reports/.
+                    --update-baseline    Overwrite baselines from the current run (requires --confirm).
+                    --confirm            Acknowledge --update-baseline (prevents accidental resets).
   upgrade    Refresh generated files in .cclaw without modifying user artifacts.
   uninstall  Remove .cclaw runtime and the generated harness shim files.
 
@@ -453,6 +456,14 @@ function parseArgs(argv) {
             parsed.evalNoWrite = true;
             continue;
         }
+        if (flag === "--update-baseline") {
+            parsed.evalUpdateBaseline = true;
+            continue;
+        }
+        if (flag === "--confirm") {
+            parsed.evalConfirm = true;
+            continue;
+        }
     }
     // `--json` is shared between doctor and eval. Disambiguate by command.
     if (parsed.command === "eval" && parsed.doctorJson === true) {
@@ -592,22 +603,42 @@ async function runCommand(parsed, ctx) {
             }
             return 0;
         }
+        if (parsed.evalUpdateBaseline === true && parsed.evalConfirm !== true) {
+            error(ctx, "--update-baseline requires --confirm to prevent accidental baseline resets.");
+            return 1;
+        }
+        if (parsed.evalUpdateBaseline === true) {
+            if (result.summary.failed > 0) {
+                error(ctx, `Refusing to update baselines: ${result.summary.failed} case(s) currently failing. Fix structural checks first.`);
+                return 1;
+            }
+            const written = await writeBaselinesFromReport(ctx.cwd, result);
+            for (const file of written) {
+                info(ctx, `Baseline written: ${path.relative(ctx.cwd, file)}`);
+            }
+        }
         if (parsed.evalNoWrite !== true) {
             const jsonPath = await writeJsonReport(ctx.cwd, result);
             const mdPath = await writeMarkdownReport(ctx.cwd, result);
             info(ctx, `Report written: ${path.relative(ctx.cwd, jsonPath)}`);
             info(ctx, `Report written: ${path.relative(ctx.cwd, mdPath)}`);
         }
+        const regressionCount = result.baselineDelta?.criticalFailures ?? 0;
         if (parsed.evalJson === true) {
             ctx.stdout.write(`${JSON.stringify(result, null, 2)}\n`);
         }
         else {
+            const regressionNote = regressionCount > 0 ? `, ${regressionCount} regression(s)` : "";
             ctx.stdout.write(`cclaw eval: ${result.summary.totalCases} case(s), ` +
                 `${result.summary.passed} passed, ` +
                 `${result.summary.failed} failed, ` +
-                `${result.summary.skipped} skipped (Wave 7.0 skeleton — verifiers land in Wave 7.1+)\n`);
+                `${result.summary.skipped} skipped${regressionNote}\n`);
         }
-        return result.summary.failed > 0 ? 1 : 0;
+        if (result.summary.failed > 0)
+            return 1;
+        if (regressionCount > 0)
+            return 1;
+        return 0;
     }
     if (command === "archive") {
         const archived = await archiveRun(ctx.cwd, parsed.archiveName, {

package/dist/constants.d.ts

@@ -5,10 +5,10 @@ export declare const CCLAW_VERSION = "0.1.1";
 export declare const FLOW_VERSION = "1.0.0";
 export declare const DEFAULT_HARNESSES: HarnessId[];
 /**
- * Evals subtree. Wave 7.0 scaffolds the directory layout and a default config.yaml;
- * verifiers and LLM wiring arrive in Waves 7.1–7.5. Keeping this separate from the
- * main REQUIRED_DIRS list makes it explicit that the evals runtime is additive and
- * does not affect non-eval cclaw behavior.
+ * Evals subtree. Scaffolds the directory layout and a default config.yaml; the
+ * structural verifier, rule verifiers, and LLM wiring layer on incrementally.
+ * Keeping this separate from the main REQUIRED_DIRS list makes it explicit that
+ * the evals runtime is additive and does not affect non-eval cclaw behavior.
  */
 export declare const EVALS_ROOT = ".cclaw/evals";
 export declare const EVALS_CONFIG_PATH = ".cclaw/evals/config.yaml";

package/dist/constants.js

@@ -9,10 +9,10 @@ export const DEFAULT_HARNESSES = [
     "codex"
 ];
 /**
- * Evals subtree. Wave 7.0 scaffolds the directory layout and a default config.yaml;
- * verifiers and LLM wiring arrive in Waves 7.1–7.5. Keeping this separate from the
- * main REQUIRED_DIRS list makes it explicit that the evals runtime is additive and
- * does not affect non-eval cclaw behavior.
+ * Evals subtree. Scaffolds the directory layout and a default config.yaml; the
+ * structural verifier, rule verifiers, and LLM wiring layer on incrementally.
+ * Keeping this separate from the main REQUIRED_DIRS list makes it explicit that
+ * the evals runtime is additive and does not affect non-eval cclaw behavior.
  */
 export const EVALS_ROOT = `${RUNTIME_ROOT}/evals`;
 export const EVALS_CONFIG_PATH = `${EVALS_ROOT}/config.yaml`;

package/dist/content/eval-scaffold.d.ts

@@ -4,8 +4,8 @@
  * scaffold is intentionally minimal: a usable default config plus short
  * READMEs that point at `docs/evals.md` for authoring guidance.
  */
-export declare const EVAL_CONFIG_YAML = "# cclaw eval config\n# See docs/evals.md for the full schema and Wave 7.1\u20137.6 rollout plan.\n#\n# All values can be overridden at runtime with CCLAW_EVAL_* environment\n# variables (env wins). Secrets like CCLAW_EVAL_API_KEY never live here.\nprovider: zai\nbaseUrl: https://api.z.ai/api/coding/paas/v4\nmodel: glm-5.1\n\n# Default fidelity tier when --tier is not supplied.\n#   A = single-shot API call (cheap, Wave 7.3)\n#   B = SDK with tool use     (realistic, Wave 7.4)\n#   C = multi-stage workflow  (end-to-end, Wave 7.5)\ndefaultTier: A\n\n# Per-call timeout and retry budget.\ntimeoutMs: 120000\nmaxRetries: 2\n\n# Optional hard-stop on estimated USD spend per day. Leave unset for no cap.\n# dailyUsdCap: 5\n\n# Regression thresholds used by CI (Wave 7.3+).\nregression:\n  # Fail when overall score drops by more than this fraction (e.g. -0.15 = 15%).\n  failIfDeltaBelow: -0.15\n  # Fail when any single critical rubric drops below this absolute score.\n  failIfCriticalBelow: 3.0\n";
-export declare const EVAL_CORPUS_README = "# Eval Corpus\n\nSeed cases live in `./<stage>/<id>.yaml`, one file per case.\nSee `docs/evals.md` for the schema; authoring begins in Wave 7.1.\n\nMinimal shape:\n\n```yaml\nid: brainstorm-01\nstage: brainstorm\ninput_prompt: |\n  One short paragraph describing the user's task.\ncontext_files: []\nexpected:\n  # verifier-specific hints; optional in Wave 7.0\n```\n\nWave 7.1 will add 3 cases per stage (24 total). Wave 7.2 will expand to 5 per\nstage (40 total). Wave 7.4/7.5 may add `context_files` pulled from real\nprojects to exercise Tier B/C sandboxes.\n";
-export declare const EVAL_RUBRICS_README = "# Eval Rubrics\n\nLLM-judge rubrics land in Wave 7.3. Each rubric is a short list of checks\nscored on a `1\u20135` scale with a rationale:\n\n```yaml\nstage: brainstorm\nchecks:\n  - id: distinctness\n    prompt: \"Are the proposed directions genuinely distinct (not rephrasings)?\"\n    scale: \"1-5 where 5=fully distinct approaches\"\n    weight: 1.0\n```\n\nRubric authoring happens when Tier A runs start producing artifacts, so we\nscore the *right* properties rather than retrofitting generic quality checks.\nSee `docs/evals.md` for the full schema.\n";
-export declare const EVAL_BASELINES_README = "# Eval Baselines\n\nFrozen score snapshots used by regression gates. Baselines are committed to\ngit and updated explicitly via `cclaw eval --update-baseline --confirm`\n(wired in Wave 7.1).\n\nEach baseline file is a JSON document keyed by stage and case id. Do not edit\nby hand; CI will flag baseline churn.\n";
+export declare const EVAL_CONFIG_YAML = "# cclaw eval config\n# See docs/evals.md for the full schema and rollout plan.\n#\n# All values can be overridden at runtime with CCLAW_EVAL_* environment\n# variables (env wins). Secrets like CCLAW_EVAL_API_KEY never live here.\nprovider: zai\nbaseUrl: https://api.z.ai/api/coding/paas/v4\nmodel: glm-5.1\n\n# Default fidelity tier when --tier is not supplied.\n#   A = single-shot API call (cheap)\n#   B = SDK with tool use     (realistic)\n#   C = multi-stage workflow  (end-to-end)\ndefaultTier: A\n\n# Per-call timeout and retry budget.\ntimeoutMs: 120000\nmaxRetries: 2\n\n# Optional hard-stop on estimated USD spend per day. Leave unset for no cap.\n# dailyUsdCap: 5\n\n# Regression thresholds used by CI.\nregression:\n  # Fail when overall score drops by more than this fraction (e.g. -0.15 = 15%).\n  failIfDeltaBelow: -0.15\n  # Fail when any single critical rubric drops below this absolute score.\n  failIfCriticalBelow: 3.0\n";
+export declare const EVAL_CORPUS_README = "# Eval Corpus\n\nSeed cases live in `./<stage>/<id>.yaml`, one file per case.\nSee `docs/evals.md` for the schema.\n\nMinimal shape:\n\n```yaml\nid: brainstorm-01\nstage: brainstorm\ninput_prompt: |\n  One short paragraph describing the user's task.\ncontext_files: []\nexpected:\n  # verifier-specific hints; optional\n```\n\nStart with 3 structural cases per stage (24 total), then expand to 5 per\nstage (40 total) once rule verifiers land. Tier B/C runs may add\n`context_files` pulled from real projects to exercise the sandbox.\n";
+export declare const EVAL_RUBRICS_README = "# Eval Rubrics\n\nLLM-judge rubrics. Each rubric is a short list of checks scored on a\n`1\u20135` scale with a rationale:\n\n```yaml\nstage: brainstorm\nchecks:\n  - id: distinctness\n    prompt: \"Are the proposed directions genuinely distinct (not rephrasings)?\"\n    scale: \"1-5 where 5=fully distinct approaches\"\n    weight: 1.0\n```\n\nRubric authoring happens when Tier A runs start producing artifacts, so we\nscore the *right* properties rather than retrofitting generic quality checks.\nSee `docs/evals.md` for the full schema.\n";
+export declare const EVAL_BASELINES_README = "# Eval Baselines\n\nFrozen score snapshots used by regression gates. Baselines are committed to\ngit and updated explicitly via `cclaw eval --update-baseline --confirm`.\n\nEach baseline file is a JSON document keyed by stage and case id. Do not edit\nby hand; CI will flag baseline churn.\n";
 export declare const EVAL_REPORTS_README = "# Eval Reports\n\nGenerated reports (JSON + Markdown) land here. This directory is gitignored.\nRun `cclaw eval --dry-run` to preview configuration without producing a\nreport.\n";

package/dist/content/eval-scaffold.js

@@ -5,7 +5,7 @@
  * READMEs that point at `docs/evals.md` for authoring guidance.
  */
 export const EVAL_CONFIG_YAML = `# cclaw eval config
-# See docs/evals.md for the full schema and Wave 7.1–7.6 rollout plan.
+# See docs/evals.md for the full schema and rollout plan.
 #
 # All values can be overridden at runtime with CCLAW_EVAL_* environment
 # variables (env wins). Secrets like CCLAW_EVAL_API_KEY never live here.
@@ -14,9 +14,9 @@ baseUrl: https://api.z.ai/api/coding/paas/v4
 model: glm-5.1
 
 # Default fidelity tier when --tier is not supplied.
-#   A = single-shot API call (cheap, Wave 7.3)
-#   B = SDK with tool use     (realistic, Wave 7.4)
-#   C = multi-stage workflow  (end-to-end, Wave 7.5)
+#   A = single-shot API call (cheap)
+#   B = SDK with tool use     (realistic)
+#   C = multi-stage workflow  (end-to-end)
 defaultTier: A
 
 # Per-call timeout and retry budget.
@@ -26,7 +26,7 @@ maxRetries: 2
 # Optional hard-stop on estimated USD spend per day. Leave unset for no cap.
 # dailyUsdCap: 5
 
-# Regression thresholds used by CI (Wave 7.3+).
+# Regression thresholds used by CI.
 regression:
   # Fail when overall score drops by more than this fraction (e.g. -0.15 = 15%).
   failIfDeltaBelow: -0.15
@@ -36,7 +36,7 @@ regression:
 export const EVAL_CORPUS_README = `# Eval Corpus
 
 Seed cases live in \`./<stage>/<id>.yaml\`, one file per case.
-See \`docs/evals.md\` for the schema; authoring begins in Wave 7.1.
+See \`docs/evals.md\` for the schema.
 
 Minimal shape:
 
@@ -47,17 +47,17 @@ input_prompt: |
   One short paragraph describing the user's task.
 context_files: []
 expected:
-  # verifier-specific hints; optional in Wave 7.0
+  # verifier-specific hints; optional
 \`\`\`
 
-Wave 7.1 will add 3 cases per stage (24 total). Wave 7.2 will expand to 5 per
-stage (40 total). Wave 7.4/7.5 may add \`context_files\` pulled from real
-projects to exercise Tier B/C sandboxes.
+Start with 3 structural cases per stage (24 total), then expand to 5 per
+stage (40 total) once rule verifiers land. Tier B/C runs may add
+\`context_files\` pulled from real projects to exercise the sandbox.
 `;
 export const EVAL_RUBRICS_README = `# Eval Rubrics
 
-LLM-judge rubrics land in Wave 7.3. Each rubric is a short list of checks
-scored on a \`1–5\` scale with a rationale:
+LLM-judge rubrics. Each rubric is a short list of checks scored on a
+\`1–5\` scale with a rationale:
 
 \`\`\`yaml
 stage: brainstorm
@@ -75,8 +75,7 @@ See \`docs/evals.md\` for the full schema.
 export const EVAL_BASELINES_README = `# Eval Baselines
 
 Frozen score snapshots used by regression gates. Baselines are committed to
-git and updated explicitly via \`cclaw eval --update-baseline --confirm\`
-(wired in Wave 7.1).
+git and updated explicitly via \`cclaw eval --update-baseline --confirm\`.
 
 Each baseline file is a JSON document keyed by stage and case id. Do not edit
 by hand; CI will flag baseline churn.

package/dist/content/examples.js

@@ -277,23 +277,23 @@ T-1 ──▶ T-2 ──▶ T-3
 
 Parallel opportunity: T-1 is a prerequisite for both T-2 and T-3 (T-3 also needs T-2).
 
-## Dependency Waves
+## Dependency Batches
 
-#### Wave 1 (foundation)
+#### Batch 1 (foundation)
 - Task IDs: T-1
 - Verification gate: schema tests pass, dedupe key fixtures validated
 
-#### Wave 2 (core logic)
+#### Batch 2 (core logic)
 - Task IDs: T-2
-- Depends on: Wave 1 (T-1 complete)
+- Depends on: Batch 1 (T-1 complete)
 - Verification gate: integration test proves publish-to-outbox path
 
-#### Wave 3 (integration)
+#### Batch 3 (integration)
 - Task IDs: T-3
-- Depends on: Wave 2 (T-2 complete)
+- Depends on: Batch 2 (T-2 complete)
 - Verification gate: e2e tests pass for delivery, dedupe, and degraded mode
 
-Execution rule: complete and verify each wave before starting the next wave.
+Execution rule: complete and verify each batch before starting the next batch.
 
 ## Task List
 
@@ -313,10 +313,10 @@ Execution rule: complete and verify each wave before starting the next wave.
 
 ## Risk Assessment
 
-| Task/Wave | Risk | Likelihood | Impact | Mitigation |
+| Task/Batch | Risk | Likelihood | Impact | Mitigation |
 | --- | --- | --- | --- | --- |
-| T-3 (Wave 3) | SSE reconnect logic complex | Medium | High | Spike reconnect in isolation before integrating with feed UI |
-| Wave 2 → 3 | Publisher API contract may shift | Low | Medium | Pin contract in T-1 schema; T-2 integration test validates |
+| T-3 (Batch 3) | SSE reconnect logic complex | Medium | High | Spike reconnect in isolation before integrating with feed UI |
+| Batch 2 → 3 | Publisher API contract may shift | Low | Medium | Pin contract in T-1 schema; T-2 integration test validates |
 
 ## WAIT_FOR_CONFIRM
 - Status: pending
@@ -682,7 +682,7 @@ const STAGE_EXAMPLE_SECTION_HEADINGS = {
         "Approval block"
     ],
     plan: [
-        "Dependency graph + dependency waves",
+        "Dependency graph + dependency batches",
         "Task list with effort + minutes estimate per task",
         "Acceptance mapping (every AC → task IDs)",
         "No-Placeholder scan row + WAIT_FOR_CONFIRM marker"

package/dist/content/hooks.js

@@ -296,7 +296,7 @@ if [ "$SUGGESTIONS_ENABLED" = "true" ] && [ "$STAGE_MUTED" != "true" ]; then
     scope) STAGE_SUGGESTION="Suggestion: lock explicit in-scope/out-of-scope boundaries and choose one scope mode." ;;
     design) STAGE_SUGGESTION="Suggestion: map failure modes per new codepath and confirm architecture boundaries before moving forward." ;;
     spec) STAGE_SUGGESTION="Suggestion: ensure every acceptance criterion is measurable and mapped to a concrete test." ;;
-    plan) STAGE_SUGGESTION="Suggestion: group tasks into dependency waves and keep WAIT_FOR_CONFIRM pending until approval." ;;
+    plan) STAGE_SUGGESTION="Suggestion: group tasks into dependency batches and keep WAIT_FOR_CONFIRM pending until approval." ;;
     tdd) STAGE_SUGGESTION="Suggestion: execute RED → GREEN → REFACTOR for each selected slice and capture evidence per cycle." ;;
     review) STAGE_SUGGESTION="Suggestion: run Layer 1 before Layer 2 and reconcile findings into 07-review-army.json." ;;
     ship) STAGE_SUGGESTION="Suggestion: verify preflight + rollback plan before selecting exactly one finalization mode." ;;

package/dist/content/skills.d.ts

@@ -1,8 +1,8 @@
 import type { FlowStage } from "../types.js";
 /**
- * Long-form Wave Execution walkthrough. Rendered once into
- * \`.cclaw/references/stages/tdd-wave-walkthrough.md\` by the installer.
+ * Long-form Batch Execution walkthrough. Rendered once into
+ * \`.cclaw/references/stages/tdd-batch-walkthrough.md\` by the installer.
  */
-export declare const TDD_WAVE_WALKTHROUGH_MARKDOWN = "# TDD \u2014 Wave Execution Walkthrough\n\nDetailed RED / GREEN / REFACTOR transcript for a 3-task wave. Illustrative\nonly \u2014 do not copy the command names blindly, match them to your stack.\n\n## Wave 1 example tasks\n\n| Task ID | Description | AC | Verification |\n|---|---|---|---|\n| T-1 `[~3m]` | Add `User.emailNormalized` column | AC-1 | `npm test -- users/schema` |\n| T-2 `[~4m]` | Normalize on write in `UserRepo.save` | AC-1 | `npm test -- users/repo` |\n| T-3 `[~3m]` | Reject duplicates in `UserService.signup` | AC-2 | `npm test -- users/service` |\n\n## Execution transcript\n\n### T-1 \u2014 RED\n\n> Run: `npm test -- users/schema` \u2192 **FAIL** (missing column: `emailNormalized`). Captured the failure stack as RED evidence. No production code touched yet.\n\n### T-1 \u2014 GREEN\n\n> Added the column in the schema module. Re-ran `npm test -- users/schema` \u2192 **PASS**. Ran the full suite `npm test` \u2192 **PASS**. Captured both outputs as GREEN evidence.\n\n### T-1 \u2014 REFACTOR\n\n> Extracted the column definition into a shared `NormalizedEmail` type used by T-2/T-3. Re-ran `npm test` \u2192 **PASS**. Captured REFACTOR note: \"Extracted NormalizedEmail type to keep T-2/T-3 DRY; zero behavior change, all tests still green.\"\n\n### T-2 \u2014 RED / GREEN / REFACTOR\n\nWrite the repo test that expects normalised writes, watch it fail (RED), implement normalisation inside `UserRepo.save` only (GREEN), then refactor the normaliser out of the repo into a helper shared with T-3 (REFACTOR).\n\n### T-3 \u2014 RED / GREEN / REFACTOR\n\nWrite the service-level duplicate test that expects a rejection, watch it fail (RED), add the duplicate check in `UserService.signup` (GREEN), refactor the error message into a named constant (REFACTOR).\n\n## Wave gate check\n\nAfter T-3 REFACTOR, before declaring Wave 1 done:\n\n1. Run the full suite (`npm test`) one final time \u2192 **PASS** captured as wave-exit evidence.\n2. Verify the TDD artifact contains RED, GREEN, and REFACTOR evidence for T-1, T-2, **and** T-3. No partial waves.\n3. Only now mark Wave 1 complete. Wave 2 cannot start until this step.\n\n## When to stop mid-wave (do NOT push through)\n\n- A RED test fails for a reason you did not predict (e.g. an unrelated flaky test) \u2192 **pause**, diagnose, log an operational-self-improvement entry, and decide with the user before proceeding.\n- A GREEN step would require touching code outside the task's acceptance criterion \u2192 **pause**, the task is scoped wrong; adjust the plan or open a follow-up task.\n- The same RED failure reappears after a GREEN change \u2192 **escalate** per the 3-attempts rule; do not keep patching.\n";
+export declare const TDD_BATCH_WALKTHROUGH_MARKDOWN = "# TDD \u2014 Batch Execution Walkthrough\n\nDetailed RED / GREEN / REFACTOR transcript for a 3-task batch. Illustrative\nonly \u2014 do not copy the command names blindly, match them to your stack.\n\n## Batch 1 example tasks\n\n| Task ID | Description | AC | Verification |\n|---|---|---|---|\n| T-1 `[~3m]` | Add `User.emailNormalized` column | AC-1 | `npm test -- users/schema` |\n| T-2 `[~4m]` | Normalize on write in `UserRepo.save` | AC-1 | `npm test -- users/repo` |\n| T-3 `[~3m]` | Reject duplicates in `UserService.signup` | AC-2 | `npm test -- users/service` |\n\n## Execution transcript\n\n### T-1 \u2014 RED\n\n> Run: `npm test -- users/schema` \u2192 **FAIL** (missing column: `emailNormalized`). Captured the failure stack as RED evidence. No production code touched yet.\n\n### T-1 \u2014 GREEN\n\n> Added the column in the schema module. Re-ran `npm test -- users/schema` \u2192 **PASS**. Ran the full suite `npm test` \u2192 **PASS**. Captured both outputs as GREEN evidence.\n\n### T-1 \u2014 REFACTOR\n\n> Extracted the column definition into a shared `NormalizedEmail` type used by T-2/T-3. Re-ran `npm test` \u2192 **PASS**. Captured REFACTOR note: \"Extracted NormalizedEmail type to keep T-2/T-3 DRY; zero behavior change, all tests still green.\"\n\n### T-2 \u2014 RED / GREEN / REFACTOR\n\nWrite the repo test that expects normalised writes, watch it fail (RED), implement normalisation inside `UserRepo.save` only (GREEN), then refactor the normaliser out of the repo into a helper shared with T-3 (REFACTOR).\n\n### T-3 \u2014 RED / GREEN / REFACTOR\n\nWrite the service-level duplicate test that expects a rejection, watch it fail (RED), add the duplicate check in `UserService.signup` (GREEN), refactor the error message into a named constant (REFACTOR).\n\n## Batch gate check\n\nAfter T-3 REFACTOR, before declaring Batch 1 done:\n\n1. Run the full suite (`npm test`) one final time \u2192 **PASS** captured as batch-exit evidence.\n2. Verify the TDD artifact contains RED, GREEN, and REFACTOR evidence for T-1, T-2, **and** T-3. No partial batches.\n3. Only now mark Batch 1 complete. Batch 2 cannot start until this step.\n\n## When to stop mid-batch (do NOT push through)\n\n- A RED test fails for a reason you did not predict (e.g. an unrelated flaky test) \u2192 **pause**, diagnose, log an operational-self-improvement entry, and decide with the user before proceeding.\n- A GREEN step would require touching code outside the task's acceptance criterion \u2192 **pause**, the task is scoped wrong; adjust the plan or open a follow-up task.\n- The same RED failure reappears after a GREEN change \u2192 **escalate** per the 3-attempts rule; do not keep patching.\n";
 export declare function stageSkillFolder(stage: FlowStage): string;
 export declare function stageSkillMarkdown(stage: FlowStage): string;

package/dist/content/skills.js

@@ -103,19 +103,19 @@ Reference utility skill:
 \`.cclaw/skills/verification-before-completion/SKILL.md\`
 `;
 }
-function waveExecutionModeBlock(stage) {
+function batchExecutionModeBlock(stage) {
     const schema = stageSchema(stage);
-    if (!schema.waveExecutionAllowed)
+    if (!schema.batchExecutionAllowed)
         return "";
-    return `## Wave Execution Mode
+    return `## Batch Execution Mode
 
-Execute the current dependency wave task-by-task (RED -> GREEN -> REFACTOR).
+Execute the current dependency batch task-by-task (RED -> GREEN -> REFACTOR).
 Stop on BLOCKED status or when user input is required.
-Apply concise turn announces: one announce per wave boundary (or when risk/plan
+Apply concise turn announces: one announce per batch boundary (or when risk/plan
 changes materially), then execute tasks without repetitive boilerplate.
 
 Detailed walkthrough:
-\`.cclaw/${STAGE_EXAMPLES_REFERENCE_DIR}/tdd-wave-walkthrough.md\`
+\`.cclaw/${STAGE_EXAMPLES_REFERENCE_DIR}/tdd-batch-walkthrough.md\`
 `;
 }
 function crossStageTraceBlock(stage) {
@@ -190,7 +190,7 @@ function stageSpecificSeeAlso(stage) {
         ],
         tdd: [
             `- \`${RUNTIME_ROOT}/skills/debugging/SKILL.md\``,
-            `- \`${RUNTIME_ROOT}/references/stages/tdd-wave-walkthrough.md\``
+            `- \`${RUNTIME_ROOT}/references/stages/tdd-batch-walkthrough.md\``
         ],
         review: [
             `- \`${RUNTIME_ROOT}/skills/security/SKILL.md\``,
@@ -239,15 +239,15 @@ function quickStartBlock(stage) {
 `;
 }
 /**
- * Long-form Wave Execution walkthrough. Rendered once into
- * \`.cclaw/references/stages/tdd-wave-walkthrough.md\` by the installer.
+ * Long-form Batch Execution walkthrough. Rendered once into
+ * \`.cclaw/references/stages/tdd-batch-walkthrough.md\` by the installer.
  */
-export const TDD_WAVE_WALKTHROUGH_MARKDOWN = `# TDD — Wave Execution Walkthrough
+export const TDD_BATCH_WALKTHROUGH_MARKDOWN = `# TDD — Batch Execution Walkthrough
 
-Detailed RED / GREEN / REFACTOR transcript for a 3-task wave. Illustrative
+Detailed RED / GREEN / REFACTOR transcript for a 3-task batch. Illustrative
 only — do not copy the command names blindly, match them to your stack.
 
-## Wave 1 example tasks
+## Batch 1 example tasks
 
 | Task ID | Description | AC | Verification |
 |---|---|---|---|
@@ -277,15 +277,15 @@ Write the repo test that expects normalised writes, watch it fail (RED), impleme
 
 Write the service-level duplicate test that expects a rejection, watch it fail (RED), add the duplicate check in \`UserService.signup\` (GREEN), refactor the error message into a named constant (REFACTOR).
 
-## Wave gate check
+## Batch gate check
 
-After T-3 REFACTOR, before declaring Wave 1 done:
+After T-3 REFACTOR, before declaring Batch 1 done:
 
-1. Run the full suite (\`npm test\`) one final time → **PASS** captured as wave-exit evidence.
-2. Verify the TDD artifact contains RED, GREEN, and REFACTOR evidence for T-1, T-2, **and** T-3. No partial waves.
-3. Only now mark Wave 1 complete. Wave 2 cannot start until this step.
+1. Run the full suite (\`npm test\`) one final time → **PASS** captured as batch-exit evidence.
+2. Verify the TDD artifact contains RED, GREEN, and REFACTOR evidence for T-1, T-2, **and** T-3. No partial batches.
+3. Only now mark Batch 1 complete. Batch 2 cannot start until this step.
 
-## When to stop mid-wave (do NOT push through)
+## When to stop mid-batch (do NOT push through)
 
 - A RED test fails for a reason you did not predict (e.g. an unrelated flaky test) → **pause**, diagnose, log an operational-self-improvement entry, and decide with the user before proceeding.
 - A GREEN step would require touching code outside the task's acceptance criterion → **pause**, the task is scoped wrong; adjust the plan or open a follow-up task.
@@ -362,7 +362,7 @@ ${schema.interactionProtocol.map((item, i) => `${i + 1}. ${item}`).join("\n")}
 Shared decision/ask-user protocol:
 \`${DECISION_PROTOCOL_PATH}\`
 
-${waveExecutionModeBlock(stage)}
+${batchExecutionModeBlock(stage)}
 ## Required Gates
 ${gateList}
 

package/dist/content/stage-schema.js

@@ -36,7 +36,7 @@ const REQUIRED_GATE_IDS = {
     ],
     plan: [
         "plan_tasks_sliced_2_5_min",
-        "plan_dependency_waves_defined",
+        "plan_dependency_batches_defined",
         "plan_acceptance_mapped",
         "plan_wait_for_confirm"
     ],
@@ -64,7 +64,7 @@ const REQUIRED_ARTIFACT_SECTIONS = {
     scope: ["Scope Mode", "In Scope / Out of Scope", "Completion Dashboard", "Scope Summary"],
     design: ["Architecture Boundaries", "Architecture Diagram", "Failure Mode Table", "Completion Dashboard"],
     spec: ["Acceptance Criteria", "Edge Cases", "Testability Map", "Approval"],
-    plan: ["Task List", "Dependency Waves", "Acceptance Mapping", "WAIT_FOR_CONFIRM"],
+    plan: ["Task List", "Dependency Batches", "Acceptance Mapping", "WAIT_FOR_CONFIRM"],
     tdd: ["RED Evidence", "GREEN Evidence", "REFACTOR Notes", "Traceability"],
     review: ["Layer 1 Verdict", "Review Army Contract", "Severity Summary", "Final Verdict"],
     ship: ["Preflight Results", "Release Notes", "Rollback Plan", "Finalization"]

package/dist/content/stages/plan.js

@@ -22,7 +22,7 @@ export const PLAN = {
     checklist: [
         "Read upstream — load spec, design, and scope artifacts. Cross-reference acceptance criteria.",
         "Build dependency graph — identify task ordering, parallel opportunities, and blocking dependencies.",
-        "Group tasks into dependency waves — wave N+1 cannot start until wave N has verification evidence.",
+        "Group tasks into dependency batches — batch N+1 cannot start until batch N has verification evidence.",
         "Slice into vertical tasks — each task targets 2-5 minutes, produces one testable outcome, and touches one coherent area.",
         "Attach verification — every task has an acceptance criterion mapping and a concrete verification command.",
         "Map scope Locked Decisions — every D-XX from scope is referenced by at least one plan task (or explicitly marked deferred with reason).",
@@ -33,7 +33,7 @@ export const PLAN = {
     interactionProtocol: [
         "Plan in read-only mode relative to implementation.",
         "Split work into small vertical slices (target 2-5 minute tasks).",
-        "Publish explicit dependency waves with entry and exit checks for each wave.",
+        "Publish explicit dependency batches with entry and exit checks for each batch.",
         "Attach verification step to every task.",
         "Preserve locked scope boundaries: no silent scope reduction language in task rows.",
         "Enforce WAIT_FOR_CONFIRM: present the plan summary with options (A) Approve / (B) Revise / (C) Reject.",
@@ -41,7 +41,7 @@ export const PLAN = {
     ],
     process: [
         "Build dependency graph and ordered slices.",
-        "Group slices into execution waves and define gate criteria per wave.",
+        "Group slices into execution batches and define gate criteria per batch.",
         "Define each task with acceptance mapping and verification commands.",
         "Trace every locked decision (D-XX) to plan tasks or explicit defer rationale.",
         "Record checkpoints and blockers.",
@@ -50,7 +50,7 @@ export const PLAN = {
     requiredGates: [
         { id: "plan_tasks_sliced_2_5_min", description: "Tasks are small, executable slices." },
         { id: "plan_dependency_graph_written", description: "Dependency graph and order are explicit." },
-        { id: "plan_dependency_waves_defined", description: "Tasks are grouped into executable waves with gate checks." },
+        { id: "plan_dependency_batches_defined", description: "Tasks are grouped into executable batches with gate checks." },
         { id: "plan_verification_steps_defined", description: "Each task has verification guidance." },
         { id: "plan_acceptance_mapped", description: "Each task maps to a spec acceptance criterion." },
         { id: "plan_wait_for_confirm", description: "Execution blocked until explicit user confirmation." }
@@ -60,7 +60,7 @@ export const PLAN = {
         "Task list includes acceptance mapping.",
         "Locked decision coverage table present with D-XX trace links.",
         "Dependency graph documented.",
-        "Dependency waves documented with wave-by-wave verification gates.",
+        "Dependency batches documented with batch-by-batch verification gates.",
         "WAIT_FOR_CONFIRM status recorded."
     ],
     inputs: ["approved spec", "codebase context", "delivery constraints"],
@@ -69,11 +69,11 @@ export const PLAN = {
         "current architecture",
         "known technical debt and dependencies"
     ],
-    outputs: ["task graph", "dependency wave plan", "ordered plan", "explicit confirmation checkpoint"],
+    outputs: ["task graph", "dependency batch plan", "ordered plan", "explicit confirmation checkpoint"],
     blockers: [
         "tasks too broad",
         "dependency uncertainty unresolved",
-        "wave boundaries are unclear",
+        "batch boundaries are unclear",
         "locked decisions from scope are not mapped to tasks",
         "no explicit confirmation"
     ],
@@ -91,13 +91,13 @@ export const PLAN = {
         "Using placeholder tokens or scope-reduction phrases (`v1`, `for now`, `later`) in task definitions",
         "No dependency graph",
         "No WAIT_FOR_CONFIRM marker",
-        "No explicit dependency waves",
+        "No explicit dependency batches",
         "Tasks exceed one coherent outcome",
         "No acceptance mapping",
         "Locked decisions are missing or not mapped",
         "Scope-reduction language appears without explicit approved defer decision"
     ],
-    policyNeedles: ["WAIT_FOR_CONFIRM", "Task Graph", "Dependency Waves", "Acceptance Mapping", "verification steps", "Locked Decision Coverage"],
+    policyNeedles: ["WAIT_FOR_CONFIRM", "Task Graph", "Dependency Batches", "Acceptance Mapping", "verification steps", "Locked Decision Coverage"],
     artifactFile: "05-plan.md",
     next: "tdd",
     reviewSections: [
@@ -113,13 +113,13 @@ export const PLAN = {
             stopGate: true
         },
         {
-            title: "Wave Completeness Audit",
+            title: "Batch Completeness Audit",
             evaluationPoints: [
-                "Does every task belong to exactly one wave?",
-                "Does each wave have a verification gate?",
-                "Are wave dependencies explicit and acyclic?",
+                "Does every task belong to exactly one batch?",
+                "Does each batch have a verification gate?",
+                "Are batch dependencies explicit and acyclic?",
                 "Is the acceptance mapping complete — every spec criterion covered?",
-                "Are there hidden dependencies between tasks in different waves?"
+                "Are there hidden dependencies between tasks in different batches?"
             ],
             stopGate: true
         },
@@ -129,7 +129,7 @@ export const PLAN = {
                 "Does every task carry an explicit minutes estimate (e.g. `[~3m]`) and does every estimate fit the 2-to-5-minute budget? Estimates >5 minutes must be split.",
                 "Are all file paths, test commands, and verification commands copy-pasteable as written — no `TODO`, `TBD`, `FIXME`, `<fill-in>`, `<your-*-here>`, `xxx`, or ellipsis standing in for omitted args?",
                 "Does every acceptance-criterion reference resolve to a real R# / AC-### in the spec (not a blank link)?",
-                "If an estimate is genuinely uncertain (first-time integration, unfamiliar library), is the uncertainty named explicitly and scheduled as a spike task in wave 0, rather than hidden behind a large estimate?"
+                "If an estimate is genuinely uncertain (first-time integration, unfamiliar library), is the uncertainty named explicitly and scheduled as a spike task in batch 0, rather than hidden behind a large estimate?"
             ],
             stopGate: true
         }
@@ -142,12 +142,12 @@ export const PLAN = {
     },
     artifactValidation: [
         { section: "Dependency Graph", required: true, validationRule: "Ordering and parallel opportunities explicit. No circular dependencies." },
-        { section: "Dependency Waves", required: true, validationRule: "Every task belongs to a wave. Each wave has an exit gate and dependency statement." },
+        { section: "Dependency Batches", required: true, validationRule: "Every task belongs to a batch. Each batch has an exit gate and dependency statement." },
         { section: "Task List", required: true, validationRule: "Each task row includes ID, description, acceptance criterion, verification command, and effort estimate (S/M/L). Every task must also carry a minutes estimate within the 2-5 minute budget." },
         { section: "Acceptance Mapping", required: true, validationRule: "Every spec criterion is covered by at least one task." },
         { section: "Locked Decision Coverage", required: false, validationRule: "Every locked decision ID (D-XX) from scope is listed with linked task IDs or explicit defer rationale." },
-        { section: "Risk Assessment", required: false, validationRule: "If present: per-task or per-wave risk identification with likelihood, impact, and mitigation strategy." },
-        { section: "Boundary Map", required: false, validationRule: "If present: per-wave or per-task interface contracts listing what each task produces (exports) and consumes (imports) from other tasks." },
+        { section: "Risk Assessment", required: false, validationRule: "If present: per-task or per-batch risk identification with likelihood, impact, and mitigation strategy." },
+        { section: "Boundary Map", required: false, validationRule: "If present: per-batch or per-task interface contracts listing what each task produces (exports) and consumes (imports) from other tasks." },
         { section: "WAIT_FOR_CONFIRM", required: true, validationRule: "Explicit marker present. Status: pending until user approves." },
         { section: "No-Placeholder Scan", required: false, validationRule: "Confirmation that a text scan for `TODO`, `TBD`, `FIXME`, `<fill-in>`, `<your-*-here>`, `xxx`, or bare ellipses has zero hits in the task list. A placeholder is a deferred decision masquerading as a plan." },
         { section: "No Scope Reduction Language Scan", required: false, validationRule: "Confirmation that scope-reduction phrases (`v1`, `for now`, `later`, `temporary`, `placeholder`) are absent from task rows when locked decisions exist." }

package/dist/content/stages/schema-types.d.ts

@@ -90,8 +90,8 @@ export interface StageSchema {
     completionStatus: string[];
     crossStageTrace: CrossStageTrace;
     artifactValidation: ArtifactValidation[];
-    /** When true, stage skill includes wave auto-execute guidance (tdd). */
-    waveExecutionAllowed?: boolean;
+    /** When true, stage skill includes batch auto-execute guidance (tdd). */
+    batchExecutionAllowed?: boolean;
     /** Sections that remain required even when the trivial-change escape hatch is active (design only). */
     trivialOverrideSections?: string[];
     /** Agent names that MUST be dispatched (or waived) before stage transition — derived from mandatory auto-subagent rows. */

package/dist/content/stages/tdd.js

@@ -179,5 +179,5 @@ export const TDD = {
         { section: "Test Pyramid Shape", required: false, validationRule: "If present: per-slice count of Small/Medium/Large tests added, to let reviewers verify the suite is not drifting top-heavy." },
         { section: "Prove-It Reproduction", required: false, validationRule: "Required for bug-fix slices: original failing reproduction test (RED without fix), passing output with fix (GREEN), and a note confirming the test fails again if the fix is reverted." }
     ],
-    waveExecutionAllowed: true
+    batchExecutionAllowed: true
 };

package/dist/content/subagents.js

@@ -129,7 +129,7 @@ If you catch yourself writing “read PLAN.md Task 3” or “implement the next
 | Status | Meaning | Controller action |
 |---|---|---|
 | DONE | Implementation complete; tests orchestrated per prompt; no known material risks | Proceed to reviewers |
-| DONE_WITH_CONCERNS | Shippable but with documented tradeoffs/risks | Proceed with reviewer + explicit notes; do not “hand-wave” concerns |
+| DONE_WITH_CONCERNS | Shippable but with documented tradeoffs/risks | Proceed with reviewer + explicit notes; do not dismiss concerns |
 | NEEDS_CONTEXT | Missing authoritative information only the parent/user can supply | Parent gathers context, then re-dispatch implementer with augmented prompt |
 | BLOCKED | Hard stop (permissions, tool failure, conflicting requirements, unsafe state) | Parent escalates to user; do not stack speculative guesses |