Add no-code Test Plan and Workflow Design modes to reflective-spec-plan

Implement the bounded, in-place repair licensed by the corrected
evidence/promotion gate (cbb0b6c): operationalize test design and
workflow design as modes *inside* reflective-spec-plan, not as a new
skill and not as Prompt Sources pointers (install copies only SKILL.md,
so a pointer would not reach installed users).

- reflective-spec-plan: Test Plan Mode (requirement-to-test matrix,
  negative/edge/regression/adversarial, false-positive guards) and
  Workflow Design Mode (formalization ladder, control model, state/
  artifacts, transition ownership, recovery, verification) + an
  Implementation Decision Gate (Claim/Evidence/Unknowns/Counterargument/
  Decision/Falsifier). Description and body now match.
- reflective-dispatch: route table + explicit test-design and
  workflow-design boundaries (executable work routes to
  reflective-implement; comparison to research; review to review).
- README, skill-map, trigger cheatsheets: reflect the two new modes.
- benchmark + route eval (route-001/002) updated for the new routing.
- Regenerated stale index.json.

Guardrails preserve the runtime non-goal: a workflow spec is not proof
of persistence, replay, recovery, cancellation, or idempotency.

Verified: project-knowledge, governance, links, lint, 52 tests, and the
routing paraphrase eval (100% consistency) all pass.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

Author: johnteee

reflective-prompt-library/README.md

@@ -51,7 +51,7 @@ The prompt files are the source material. The `skills/` directory is the operati
 | --- | --- |
 | Route a task to the smallest useful reflective workflow | `skills/reflective-dispatch/SKILL.md` |
 | Clarify goal, assumptions, scope, acceptance, falsifiability | `skills/reflective-brief/SKILL.md` |
-| Write spec, usage-first design, task slices, or a no-code Test Plan | `skills/reflective-spec-plan/SKILL.md` |
+| Write spec, usage-first design, task slices, a no-code Test Plan, or a no-code agent workflow design | `skills/reflective-spec-plan/SKILL.md` |
 | Implement code with verification and traceability | `skills/reflective-implement/SKILL.md` |
 | Challenge unnecessary code, dependencies, abstractions, or scope before implementation | `skills/reflective-minimality/SKILL.md` |
 | Review code, plans, specs, and AI outputs | `skills/reflective-review/SKILL.md` |

reflective-prompt-library/README.zh-TW.md

@@ -34,7 +34,7 @@ Doing the right thing > doing things right.
 | --- | --- |
 | 分流任務到最小適用 workflow | `skills/reflective-dispatch/SKILL.md` |
 | 釐清目標、假設、範圍、驗收、可證偽 | `skills/reflective-brief/SKILL.md` |
-| 寫 spec、usage-first 設計、task slices，或不寫程式碼的 Test Plan | `skills/reflective-spec-plan/SKILL.md` |
+| 寫 spec、usage-first 設計、task slices、不寫程式碼的 Test Plan，或不寫 runtime code 的 agent workflow design | `skills/reflective-spec-plan/SKILL.md` |
 | 實作程式碼，附驗證與可追蹤性 | `skills/reflective-implement/SKILL.md` |
 | 實作前挑戰不必要 code、dependency、抽象層或 scope | `skills/reflective-minimality/SKILL.md` |
 | 審查 code / plan / spec / AI output | `skills/reflective-review/SKILL.md` |

reflective-prompt-library/index.json

@@ -1,6 +1,6 @@
 {
   "version": "1.0",
-  "generated_at": "2026-06-20T01:31:58.655885",
+  "generated_at": "2026-06-20T02:33:36.899079",
   "repo_root": ".",
   "categories": {
     "prompt-library": {
@@ -1421,8 +1421,8 @@
           "text": "Conclusion"
         }
       ],
-      "line_count": 300,
-      "char_count": 11962
+      "line_count": 303,
+      "char_count": 12363
     },
     {
       "path": "reflective-prompt-library/plans/external-adoption-case-studies-2026-06-20.md",
@@ -1442,7 +1442,6 @@
           "agentic-sop-workflow-reflection-2026-06-13.md",
           "../04-agent/sop-compiler.md",
           "../skills/reflective-review/SKILL.md",
-          "../PROJECT_KNOWLEDGE.md",
           "../skills/reflective-research/SKILL.md",
           "../skills/reflective-minimality/SKILL.md",
           "../skills/reflective-dispatch/SKILL.md"
@@ -1473,6 +1472,14 @@
           "level": 2,
           "text": "Signal Accounting (do not miscount)"
         },
+        {
+          "level": 2,
+          "text": "Decision-Rule Correction"
+        },
+        {
+          "level": 3,
+          "text": "Applied Counterexample: Test Plan Routing"
+        },
         {
           "level": 2,
           "text": "Methodology Layer vs Operationalization Layer"
@@ -1494,8 +1501,8 @@
           "text": "State Ledger"
         }
       ],
-      "line_count": 116,
-      "char_count": 7057
+      "line_count": 161,
+      "char_count": 9690
     },
     {
       "path": "reflective-prompt-library/plans/ROUTING_CONTRACT.md",
@@ -2848,7 +2855,7 @@
         }
       ],
       "line_count": 78,
-      "char_count": 3928
+      "char_count": 4051
     },
     {
       "path": "reflective-prompt-library/skills/SKILL_TRIGGER_CHEATSHEET.md",
@@ -2911,8 +2918,8 @@
           "text": "Fast Routing Rule"
         }
       ],
-      "line_count": 154,
-      "char_count": 5032
+      "line_count": 158,
+      "char_count": 5367
     },
     {
       "path": "reflective-prompt-library/skills/SKILL_TRIGGER_CHEATSHEET.zh-TW.md",
@@ -2977,8 +2984,8 @@
           "text": "快速順序"
         }
       ],
-      "line_count": 138,
-      "char_count": 2296
+      "line_count": 142,
+      "char_count": 2495
     },
     {
       "path": "reflective-prompt-library/05-domain/creative-template.md",
@@ -3426,8 +3433,8 @@
           "text": "Guardrails"
         }
       ],
-      "line_count": 111,
-      "char_count": 4685
+      "line_count": 115,
+      "char_count": 5356
     },
     {
       "path": "reflective-prompt-library/skills/reflective-minimality/SKILL.md",
@@ -3975,8 +3982,8 @@
           "text": "Prompt Sources"
         }
       ],
-      "line_count": 148,
-      "char_count": 5173
+      "line_count": 156,
+      "char_count": 6005
     },
     {
       "path": "reflective-prompt-library/skills/reflective-handoff-retro/SKILL.md",
@@ -4289,11 +4296,11 @@
       "category": "skills",
       "subcategory": "reflective-spec-plan",
       "name": "reflective-spec-plan",
-      "description": "Use this to turn a requirement into a practical spec, usage-first design, and small reviewable task plan. Trigger when the user asks for a spec, PRD, tickets, implementation plan, workflow plan, or wants to avoid building the wrong thing.",
+      "description": "Use this to turn a requirement into a practical spec, usage-first design, reviewable task plan, no-code Test Plan, or agent workflow design. Trigger when the user asks for a spec, PRD, tickets, implementation plan, workflow plan, stateful or resumable workflow specification, rigorous test design before implementation, or wants to avoid building the wrong thing.",
       "license": "MIT",
       "frontmatter": {
         "name": "reflective-spec-plan",
-        "description": "Use this to turn a requirement into a practical spec, usage-first design, and small reviewable task plan. Trigger when the user asks for a spec, PRD, tickets, implementation plan, workflow plan, or wants to avoid building the wrong thing.",
+        "description": "Use this to turn a requirement into a practical spec, usage-first design, reviewable task plan, no-code Test Plan, or agent workflow design. Trigger when the user asks for a spec, PRD, tickets, implementation plan, workflow plan, stateful or resumable workflow specification, rigorous test design before implementation, or wants to avoid building the wrong thing.",
         "license": "MIT",
         "risk_level": "low",
         "human_review_required": "false",
@@ -4308,7 +4315,10 @@
         "spec.md",
         "usage.md",
         "task-plan.md",
+        "test-plan.md",
+        "workflow-spec.md",
         "02-engineering/spec-writer.md",
+        "02-engineering/test-designer.md",
         "02-engineering/usage-first.md",
         "02-engineering/task-slicer.md",
         "04-agent/workflow-engine.md",
@@ -4328,10 +4338,26 @@
           "level": 2,
           "text": "Module Contract"
         },
+        {
+          "level": 2,
+          "text": "Implementation Decision Gate"
+        },
         {
           "level": 2,
           "text": "Workflow"
         },
+        {
+          "level": 2,
+          "text": "Test Plan Mode"
+        },
+        {
+          "level": 3,
+          "text": "TEST-001: <name>"
+        },
+        {
+          "level": 2,
+          "text": "Workflow Design Mode"
+        },
         {
           "level": 2,
           "text": "Artifact Output"
@@ -4353,8 +4379,8 @@
           "text": "Prompt Sources"
         }
       ],
-      "line_count": 135,
-      "char_count": 3909
+      "line_count": 230,
+      "char_count": 10292
     }
   ],
   "total_files": 79

reflective-prompt-library/plans/QUALITY_GATES_SUMMARY.md

@@ -18,7 +18,7 @@ This document summarizes the Phase 1 quality gates implemented for TeaPrompt bas
 - Checks for required frontmatter fields (name, description)
 
 **Results:**
-- Scanned 96 markdown files
+- Scanned 103 files
 - 0 errors found
 - All links and references are valid
 
@@ -63,9 +63,9 @@ python3 reflective-prompt-library/plans/generate_index.py
 - Provides actionable suggestions
 
 **Results:**
-- Scanned 96 files
+- Scanned 103 files
 - 0 errors, 0 warnings
-- 75 files with suggestions (mostly non-critical)
+- 81 files with suggestions (mostly non-critical)
 - All 9 skills pass validation
 
 **Usage:**
@@ -80,14 +80,14 @@ python3 reflective-prompt-library/plans/lint_skills.py
 **What it does:**
 - Implements ROUTE-001 from code-followups-plan.md
 - Loads intent groups, adversarial sets, thresholds, trace fields, and expected workflows from `route-001-paraphrase-eval.yaml`
-- Applies a small deterministic boundary classifier for ambiguous review/risk/planning phrasing
+- Applies a small deterministic boundary classifier for ambiguous review/risk/planning and workflow-design phrasing
 - Tests routing consistency across paraphrased intents
 - Validates that same intent routes to same workflow
 - Measures confidence, fallback behavior, low-confidence trace coverage, and silent downgrade incidents
 - Generates consistency reports
 
 **Results:**
-- Tested 15 groups with 113 paraphrases, including 4 adversarial boundary groups, the minimality gate, and both sides of the Test Plan / executable-test boundary
+- Tested 16 groups with 128 paraphrases, including 4 adversarial boundary groups, the minimality gate, and the Test Plan, executable-test, and workflow-design boundaries
 - Overall consistency: 100.0% (passes Phase-1 threshold >=70% and aspirational target >=95%)
 - Low-confidence route trace coverage: 100.0%
 - Adversarial groups now pass through boundary rules rather than isolated synonym matching
@@ -111,10 +111,10 @@ python3 reflective-prompt-library/plans/route_paraphrase_eval.py
 - Keeps a lower Phase-1 bar and separate aspirational target to avoid over-claiming
 
 **Results:**
-- Tested 9 holdout groups with 37 paraphrases
+- Tested 11 holdout groups with 46 paraphrases
 - Overall consistency: 100.0% (passes Phase-1 threshold >=80% and aspirational target >=90%)
 - Low-confidence route trace coverage: 100.0%
-- Review phrasing around patches/regressions and clarification phrasing around unknown desired outcome now pass through boundary concepts
+- Review, clarification, workflow-design, workflow-implementation, research, and orchestration-selection phrasing now pass through boundary concepts
 - Output: `plans/route-002-results.json`
 
 **Usage:**
@@ -176,16 +176,16 @@ python3 reflective-prompt-library/plans/validate_governance.py
 **File:** `reflective-prompt-library/plans/benchmark_tasks.py`
 
 **What it does:**
-- Defines 21 golden tasks for validation
-- Covers all major workflows (7 different skills)
-- Balanced difficulty distribution (5 easy, 9 medium, 6 hard)
-- Diverse categories (12 different categories)
+- Defines 23 golden tasks for validation
+- Covers all major benchmarked workflows (8 different skills)
+- Balanced difficulty distribution (5 easy, 11 medium, 7 hard)
+- Diverse categories (15 different categories)
 - Clear acceptance criteria for each task
 - Expected workflow mapping
 
 **Results:**
-- 21 benchmark tasks created
-- Tasks cover: implementation, planning, review, research, risk, handoff, debugging, refactoring, retrospective, runtime governance, scaffold provenance, SOP compiler planning, and minimality
+- 23 benchmark tasks created
+- Tasks cover: implementation, planning, review, research, risk, handoff, debugging, refactoring, retrospective, runtime governance, scaffold provenance, SOP compiler planning, minimality, test planning, and workflow design
 - Output: `plans/benchmark-tasks.json`
 
 **Usage:**
@@ -207,17 +207,17 @@ The implementation aligns with research findings:
 
 | Metric | Value | Status |
 |--------|-------|--------|
-| Files validated | 96 | ✅ All pass |
+| Files validated | 103 | ✅ All pass |
 | Links validated | 0 broken | ✅ Perfect |
 | Skills with governance | 9/9 | ✅ Complete |
-| Benchmark tasks | 21 | ✅ Ready |
+| Benchmark tasks | 23 | ✅ Ready |
 | Routing consistency | 100.0% | ✅ Passes ROUTE-001 expanded boundary eval |
 | Holdout routing consistency | 100.0% | ✅ Passes ROUTE-002 holdout eval |
 | Linting errors | 0 | ✅ Clean |
 
 ### Routing Consistency Tracking
 
-The current routing consistency measurement is 100.0% on ROUTE-001 across 15 groups and 113 paraphrases. The eval fixture is now the single source of truth for thresholds, expected workflows, trace fields, and adversarial sets, while `route_paraphrase_eval.py` is responsible for loading the fixture and measuring the current router. This passes the Phase-1 threshold (>=70%) and aspirational target (>=95%) for the expanded seeded boundary set, including the minimality gate and the Test Plan / executable-test boundary. The ROUTING_CONTRACT.md keeps both bars explicit so future evals can distinguish minimum acceptance from aspirational quality.
+The current routing consistency measurement is 100.0% on ROUTE-001 across 16 groups and 128 paraphrases. The eval fixture is now the single source of truth for thresholds, expected workflows, trace fields, and adversarial sets, while `route_paraphrase_eval.py` is responsible for loading the fixture and measuring the current router. This passes the Phase-1 threshold (>=70%) and aspirational target (>=95%) for the expanded seeded boundary set, including the minimality gate and the Test Plan, executable-test, and workflow-design boundaries. The ROUTING_CONTRACT.md keeps both bars explicit so future evals can distinguish minimum acceptance from aspirational quality.
 
 ### Critical Reflection
 
@@ -227,19 +227,22 @@ ROUTE-001 should not be interpreted as proof that routing is solved. The useful
 - research vs local review
 - clarification vs planning
 - handoff vs dispatch
+- workflow design vs executable orchestration
+- workflow research or review vs first-draft design
+- workflow formalization selection vs committing to a workflow
 
 The latest router improvement uses concept-level boundary rules for these cases instead of raw synonym accumulation. Future improvements should add unseen holdout cases or baseline-vs-skill evaluation before claiming general routing quality.
 
 ### Holdout Tracking
 
-ROUTE-002 currently measures unseen phrasing separately from ROUTE-001. It passes the holdout aspirational target with 100.0% consistency after adding concept-level review and clarification boundaries. This should still be treated as a seeded holdout, not proof of broad semantic routing; the next useful step is adding fresh holdout cases before further router tuning.
+ROUTE-002 currently measures unseen phrasing separately from ROUTE-001. It passes the holdout aspirational target with 100.0% consistency after adding concept-level review, clarification, workflow-design, and orchestration-selection boundaries. This should still be treated as a seeded holdout, not proof of broad semantic routing; the next useful step is adding fresh holdout cases before further router tuning.
 
 ## Next Steps (Phase 2)
 
 Based on the research and current implementation, suggested next steps:
 
 1. **Expand ROUTE-002 with fresh holdout cases** - Add unseen cases before further router tuning
-2. **Run benchmark tests** - Execute the 20 tasks with/without skills to measure effectiveness
+2. **Run benchmark tests** - Execute the 23 tasks with/without skills to measure effectiveness
 3. **Add CI/CD** - Integrate validation scripts into automated pipeline
 4. **Collect feedback** - Use CONTRIBUTING.md process to gather community input
 5. **Iterate on routing** - Use paraphrase eval results to improve keyword matching

reflective-prompt-library/plans/benchmark_tasks.py

@@ -358,6 +358,21 @@ def define_tasks(self) -> List[BenchmarkTask]:
                     "Adversarial or anti-cheating checks are included when relevant",
                     "Unknowns and false-positive guards are recorded without changing production code"
                 ]
+            ),
+            BenchmarkTask(
+                id="B023",
+                name="No-code agent workflow design",
+                description="Design a resumable human-in-the-loop agent workflow without implementing a runner",
+                expected_workflow="reflective-spec-plan",
+                difficulty="hard",
+                category="workflow-design",
+                acceptance_criteria=[
+                    "The lowest sufficient formalization level is selected and justified",
+                    "Fixed workflow, dynamic agent, and selected orchestration patterns are distinguished",
+                    "State, artifacts, checkpoints, and transition control ownership are explicit",
+                    "Side effects define authority, idempotency, retry caps, compensation, and hard stops",
+                    "Observability and scenario tests separate specified behavior from unproven runtime guarantees"
+                ]
             )
         ]
 

reflective-prompt-library/plans/route-001-paraphrase-eval.yaml

@@ -34,6 +34,8 @@ intent_groups:
       - Patch the parser so it handles empty input
       - Build this feature in the code
       - Make the code change directly
+      - Implement a LangGraph workflow with persistence
+      - Build the runner from this workflow design
 
   - intent: spec_planning
     expected_workflow: reflective-spec-plan
@@ -53,6 +55,23 @@ intent_groups:
       - Create a roadmap for the workflow
       - Prepare a release plan for this feature
 
+  - intent: workflow_design_only
+    expected_workflow: reflective-spec-plan
+    expected_rigor: medium
+    expected_enhancements:
+      - state_and_artifact_contracts
+      - transition_ownership
+      - recovery_and_observability
+    phrases:
+      - Design a resumable agent workflow with persisted state, gates, recovery, and human review
+      - Turn this SOP into a gated workflow specification without implementing a runner
+      - Create a workflow design with sequential parallel routing and evaluator loops
+      - Define the state model transitions checkpoints and rollback for this process
+      - Plan a manager-and-handoff workflow without writing code
+      - Map this process into stages artifacts gates and stop conditions
+      - Produce an orchestration plan that separates deterministic and model-decided control
+      - Specify a human-in-the-loop workflow with observable failure recovery
+
   - intent: test_plan_only
     expected_workflow: reflective-spec-plan
     expected_rigor: medium
@@ -101,6 +120,7 @@ intent_groups:
       - Look over my code in detail
       - Make sure it actually works
       - Review this like it is going to production
+      - Review this workflow design for missing recovery paths
 
   - intent: research_task
     expected_workflow: reflective-research
@@ -119,6 +139,8 @@ intent_groups:
       - Check source-backed guidance for this package
       - Find the current official recommendation
       - Research whether this approach is still valid
+      - Research current workflow orchestration patterns
+      - Compare official framework docs for durable workflow design
 
   - intent: risk_gating
     expected_workflow: reflective-risk
@@ -185,6 +207,8 @@ intent_groups:
       - Dispatch this to the best skill
       - Decide whether this needs a workflow or just a prompt
       - What skill is appropriate here?
+      - Choose between a deterministic workflow and a dynamic agent for this process
+      - Decide whether this needs a prompt workflow or full agent
 
   - intent: minimality_gate
     expected_workflow: reflective-minimality