olympus-ai 4.4.13 → 4.4.15

package/resources/rules/construction/code-generation.md

@@ -30,7 +30,14 @@ This stage generates code for each unit of work through two integrated parts:
 
 **If an agent task fails**: Follow the Agent Task Failure Recovery procedure in `error-handling.md` — retry the delegation, never silently do the work yourself.
 
-**After agent completes**: The orchestrator reviews generated code, presents the completion message (Step 14), and manages the approval gate (Steps 15-16).
+**After agent completes**: The orchestrator MUST verify completion before proceeding:
+1. Read the plan file and count `[x]` vs `[ ]` checkboxes
+2. If ALL checkboxes are `[x]` AND code-summary.md exists → proceed to Step 14 (completion message)
+3. If ANY checkboxes remain `[ ]` OR code-summary.md is missing → **re-delegate** from the first unchecked step:
+   - Log the partial completion in audit.md ("Agent completed steps 1-{M} of {N}, re-delegating steps {M+1}-{N}")
+   - Send a new Task to the same agent type with the prompt: "Continue Part 2 from step {M+1}. Steps 1-{M} are already complete. Complete the remaining steps {M+1} through {N}." followed by the standard delegation prompt
+   - Repeat until all checkboxes are marked
+4. The orchestrator manages the approval gate (Steps 15-16)
 
 ### Mandatory Delegation Prompt Requirements
 
@@ -38,15 +45,22 @@ When delegating Part 2 to an agent, the Task tool prompt MUST include all of the
 
 ```
 You are executing Part 2 (Code Generation) for the AIDLC unit "{unit-name}".
+This plan has {N} steps. Your task is NOT complete until ALL {N} steps are marked [x].
 
 1. Read the complete code generation plan at:
    aidlc-docs/{workflow-id}/construction/plans/{unit-name}-code-generation-plan.md
 
-2. Execute each step in the plan exactly, in order. Do NOT skip steps or deviate.
+2. Count the total number of steps with checkboxes [ ]. This is your completion target.
 
-3. After completing each step, immediately mark its checkbox [x] in the plan file.
+3. Execute each step in the plan exactly, in order. Do NOT skip steps or deviate.
 
-4. After ALL steps are complete, create a code summary at:
+4. After completing each step, immediately mark its checkbox [x] in the plan file.
+
+5. After marking a checkbox, check: are there more [ ] checkboxes remaining in the plan?
+   - YES → Continue to the next step immediately. DO NOT STOP OR RETURN.
+   - NO → All steps done. Proceed to step 6.
+
+6. After ALL steps are marked [x], create a code summary at:
    aidlc-docs/{workflow-id}/construction/{unit-name}/code/code-summary.md
 
    The summary must include:
@@ -56,9 +70,13 @@ You are executing Part 2 (Code Generation) for the AIDLC unit "{unit-name}".
    - User stories implemented (reference story IDs)
    - Known gaps or deferred items
 
-Do not report completion until the plan checkboxes are updated and the code summary exists.
+⚠️ CRITICAL: Do NOT return after completing only some steps. Completing 2-3 steps out of {N} and stopping is a FAILURE. You must finish the ENTIRE plan — all {N} steps — before returning.
+
+Do not report completion until EVERY plan checkbox is [x] and the code summary file exists.
 ```
 
+**Note**: Replace `{N}` with the actual step count from the plan before delegating. The orchestrator must count the steps and include the concrete number.
+
 ## Orchestrator Execution Requirements
 
 When managing code generation, the orchestrator MUST leverage Olympus capabilities:
@@ -341,5 +359,8 @@ When generating UI code (web, mobile, desktop), ensure elements are automation-f
 - All steps in unit code generation plan marked [x]
 - All unit stories implemented according to plan
 - All code and tests generated (tests will be executed in Build & Test phase)
+- After code generation completes for a unit, proceed to the **Test Generation** stage
+  (see `resources/rules/construction/test-generation.md`) before moving to the next unit
+  or Build & Test.
 - Deployment artifacts generated
 - Complete unit ready for build and verification

package/resources/rules/construction/test-generation.md

@@ -0,0 +1,82 @@
+# Test Generation - Detailed Steps
+
+## Overview
+
+This stage generates and runs tests for the current unit after code generation completes.
+
+- Agent responsible: `qa-tester` (primary) or `olympian` for test writing
+- Output artifact: `aidlc-docs/{workflowId}/construction/{unitId}/testing/test-report.md`
+
+## Prerequisites
+
+- Code generation must be complete (`code-summary.md` must exist at `aidlc-docs/{workflowId}/construction/{unitId}/code/code-summary.md`)
+- Unit files in scope are read from `code-summary.md`
+- If `code-summary.md` does not exist, halt and report to orchestrator before proceeding
+
+## Step 1 — Framework Detection (Hybrid)
+
+- **1a**: The engine stores the detected framework in `test_framework` on `ConstructionUnitProgress`
+- **1b**: Agent independently verifies: read `package.json`, `vitest.config.*`, `jest.config.*` at project root
+- **1c**: If engine value and agent value disagree, agent value wins; log the discrepancy
+
+Known frameworks and their test commands:
+
+| Framework | Test Command |
+|-----------|-------------|
+| `vitest` | `npx vitest run` |
+| `jest` | `npx jest` |
+| `mocha` | `npx mocha` |
+| Unknown | Ask user before proceeding |
+
+## Step 2 — Determine Test Types (Auditable Criteria)
+
+Evaluate each criterion explicitly and record which test types apply in `test-report.md`:
+
+- **Unit tests**: Required for all pure functions, class methods, utilities. File naming: `*.test.ts` or `*.spec.ts` co-located with source.
+- **Integration tests**: Required when the unit touches 2 or more modules, external APIs, databases, or file I/O. Placed in `tests/integration/`.
+- **E2E tests**: Required only when the unit includes a user-facing entry point (HTTP endpoint, CLI command, UI page). Placed in `tests/e2e/`.
+
+## Step 3 — Generate Tests
+
+- Scope: only modify or create files listed in `code-summary.md`'s "Files created/modified" sections
+- Do NOT modify files from other units
+- Follow existing test file conventions in the project (import style, describe/it structure, mock patterns)
+- Use `data-testid` attributes for UI component tests
+
+## Step 4 — Run Tests
+
+- Execute the framework test command for the unit's files only (scope by file path filter where possible)
+- Capture: total count, passed count, failed count
+- Write results into `test-report.md`
+
+## Step 5 — Failure Handling
+
+- On first failure: attempt one automated fix per failing test (fix the test or the implementation; prefer fixing the test unless the implementation has a clear bug)
+- On second failure: attempt a second fix with a different strategy
+- After two failed attempts: escalate — write the failure details to `test-report.md` and set `tests_failed` count; do NOT attempt a third fix
+- Escalation message format: surface to the orchestrator with file path, test name, error message
+
+## Engine Gating Rules
+
+- The engine blocks unit completion if `tests_total === 0` (no tests detected)
+- The engine blocks unit completion if `tests_failed > 0`
+- Both blocks can be overridden by setting `allowFailures: true` in `TestGenerationOptions`
+- Override must be logged in `test-report.md` under the `## Override` section
+
+## Code Modification Scope
+
+- The agent may ONLY modify files listed in `code-summary.md` for this unit
+- `code-summary.md` is at: `aidlc-docs/{workflowId}/construction/{unitId}/code/code-summary.md`
+- If `code-summary.md` does not exist, halt and report to orchestrator before proceeding
+
+## Output Artifact
+
+- Path: `aidlc-docs/{workflowId}/construction/{unitId}/testing/test-report.md`
+- Must exist before the unit is marked complete
+
+## Completion Criteria
+
+- `test-report.md` written with actual counts (not placeholders)
+- `tests_total > 0`
+- `tests_failed === 0` (or override documented)
+- `ConstructionUnitProgress.stages['test-generation'].status === 'completed'`

package/resources/skills/continue/SKILL.md

@@ -138,6 +138,8 @@ If `current_phase === 'construction'`:
 - Check `construction_units` for the active unit
 - Determine which design stage is `in_progress` or `not_started`
 - Resume from that point
+- If a `construction_units` entry has `stages['test-generation'].status === 'in_progress'` or `test_generation_status === 'in_progress'`, resume at test-generation for that unit
+- Note: test-generation runs after code-generation for each unit; check `test_generation_status` in the unit progress
 
 ### 4d. Other Phases
 
@@ -163,6 +165,7 @@ Based on the resume point determined in Step 4, read the corresponding rule file
 | nfr-design | `~/.claude/olympus/rules/construction/nfr-design.md` |
 | infrastructure-design | `~/.claude/olympus/rules/construction/infrastructure-design.md` |
 | code-generation | `~/.claude/olympus/rules/construction/code-generation.md` |
+| test-generation | `~/.claude/olympus/rules/construction/test-generation.md` |
 
 ---
 
@@ -217,6 +220,7 @@ Wait for user response before proceeding.
 | infrastructure-design | `oracle-medium` |
 | code-generation (backend) | `olympian` or `olympian-high` |
 | code-generation (frontend) | `frontend-engineer` or `frontend-engineer-high` |
+| test-generation | `qa-tester` |
 | build-and-test | `qa-tester` |
 
 ### If user chose B (Review)