intentdna 1.5.20 → 1.6.0

package/dist/templates/flutter-rewrite.dna.yaml

@@ -239,28 +239,29 @@ roles:
       - "DO NOT run tests. DO NOT edit app code. Analysis only."
 
   analysis_reviewer:
-    description: "Reviews analyzer output quality. Read-only."
+    description: "Reviews analyzer output quality and writes durable diagnosis review verdict."
     tool_permissions:
-      allow: [Read, Grep, Glob]
-      deny: [Bash, Edit, Write, NotebookEdit]
+      allow: [Read, Grep, Glob, Write]
+      deny: [Bash, Edit, NotebookEdit]
     scope:
       read: ["**/*"]
-      write: []
+      write: [".dna/specs/diagnosis-*.review.json"]
     instructions:
       - "REQUIRED FIRST: Read all context files listed in SKILL.md"
       - "Verify analyzer spec: v1 references exist? Classifications sound?"
       - "Check missing: any failing test not covered?"
       - "Verify diagnosis remains evidence-only: no fix prescriptions, implementation order, call-site instructions, or Notes for Surgeon"
-      - "Output a structured verdict block with verdict, artifact_reviewed, sections_to_fix, evidence_paths, confidence, and summary"
+      - "Output a structured verdict block with verdict, artifact_reviewed, sections_to_fix, evidence_paths, confidence, summary, and updated_at"
+      - "Write the same structured verdict JSON to .dna/specs/diagnosis-$ARGUMENTS.review.json before finishing"
 
   fix_reviewer:
-    description: "Reviews surgeon changes against approved diagnosis, v1 evidence, and v2 architecture. Read-only."
+    description: "Reviews surgeon changes against approved diagnosis, v1 evidence, and v2 architecture. Writes only the structured review verdict artifact."
     tool_permissions:
-      allow: [Read, Grep, Glob, Bash]
-      deny: [Edit, Write, NotebookEdit]
+      allow: [Read, Grep, Glob, Bash, Write]
+      deny: [Edit, NotebookEdit]
     scope:
       read: ["**/*"]
-      write: []
+      write: [".dna/reviews/**"]
     instructions:
       - "Read diagnosis spec, behavior doc, and git diff before judging"
       - "Verify changes are minimal and limited to the approved diagnosis classifications"
@@ -283,6 +284,26 @@ roles:
       - "Classify red tests by type (compile/logic/widget/hung)"
       - "Output progress delta, not test content"
 
+controllers:
+  fix-loop:
+    name: Fix Loop
+    description: "Run diagnosis once if needed, then loop fix/review/verify until PASS or stop condition"
+    kind: flutter_rewrite_fix_loop
+    max_rounds: 10
+    progress_artifact_path: ".dna/fix-progress/$ARGUMENTS.json"
+    diagnosis_workflow: diagnosis
+    fix_workflow: fix
+    diagnosis_artifact_path: ".dna/specs/diagnosis-$ARGUMENTS.md"
+    diagnosis_review_artifact_path: ".dna/specs/diagnosis-$ARGUMENTS.review.json"
+    review_artifact_path: ".dna/reviews/$ARGUMENTS-review.json"
+    roles:
+      analyzer: analyzer
+      analysis_reviewer: analysis_reviewer
+      surgeon: surgeon
+      fix_reviewer: fix_reviewer
+      test_runner: test_runner
+    stop_format: zh_three_part
+
 workflows:
   behavior-lock:
     name: Behavior Lock
@@ -411,10 +432,13 @@ workflows:
             "sections_to_fix": ["section ids or titles; empty when approved"],
             "evidence_paths": ["paths that justify the verdict"],
             "confidence": "high" | "medium" | "low",
-            "summary": "brief reason"
+            "summary": "brief reason",
+            "updated_at": "ISO-8601 timestamp"
           }
           ```
 
+          Write the same verdict JSON to .dna/specs/diagnosis-$ARGUMENTS.review.json before finishing.
+
           APPROVE → diagnosis complete.
           REQUEST_REANALYSIS → sections_to_fix must be passed back to analyze. docs/behavior/blocked_items.md is historical context only and must not be treated as the current run verdict.
         handoff:
@@ -423,7 +447,11 @@ workflows:
               path: ".dna/specs/diagnosis-$ARGUMENTS.md"
               description: "Diagnosis spec from analyzer"
           produces:
+            - type: file
+              path: ".dna/specs/diagnosis-$ARGUMENTS.review.json"
+              description: "Durable diagnosis review verdict (APPROVE or REQUEST_REANALYSIS)"
             - type: summary
+              required: false
               description: "Review verdict (APPROVE or REQUEST_REANALYSIS)"
 
   fix:
@@ -474,9 +502,9 @@ workflows:
         depends_on: [fix_bugs]
         description: "Review surgeon changes before verification"
         prompt: |
-          Read diagnosis spec, behavior doc, and git diff.
+          Read diagnosis spec and behavior doc before reviewing the surgeon's handoff changes.
 
-          Review only the surgeon's changes:
+          Review only the surgeon's handoff changes:
           1. Are changes limited to BUG / UNIMPLEMENTED / INFRA / TEST_BUG items in the approved diagnosis?
           2. Is REMOVED handling absent unless explicitly approved by user?
           3. Does implementation preserve v1 behavior and follow v2 architecture?
@@ -487,14 +515,26 @@ workflows:
           ```json
           {
             "verdict": "APPROVE" | "REQUEST_CHANGES",
+            "module": "$ARGUMENTS",
+            "round": 1,
+            "run_id": "controller run id or manual-run",
             "issues": ["specific issue with evidence path; empty when approved"],
             "evidence_paths": ["paths that justify the verdict"],
             "failure_reason": "why changes need another round; empty when approved",
             "next_round_focus": "specific priority/classification/test file for the next fix invocation; empty when approved",
-            "summary": "brief reason"
+            "summary": "brief reason",
+            "updated_at": "ISO-8601 timestamp"
           }
           ```
 
+          Write the same verdict JSON to .dna/reviews/$ARGUMENTS-review.json before finishing.
+
+          JSON field rules:
+          - module is required and must equal $ARGUMENTS.
+          - round is required and must be the provided current_round, or 1 for manual fix invocation.
+          - run_id is required and must equal the provided controller_run_id, or "manual-run" for manual fix invocation.
+          - updated_at is a required ISO timestamp string.
+
           REQUEST_CHANGES → include failure_reason and concrete next_round_focus for the next fix invocation. The next_round_focus must name the priority/classification and specific test file or evidence path to address first.
         handoff:
           consumes:
@@ -508,7 +548,8 @@ workflows:
               path: "docs/behavior/$ARGUMENTS.md"
               description: "Behavior document"
           produces:
-            - type: summary
+            - type: file
+              path: ".dna/reviews/$ARGUMENTS-review.json"
               description: "Review verdict (APPROVE or REQUEST_CHANGES)"
 
       - id: verify_report
@@ -522,37 +563,101 @@ workflows:
 
           If the review verdict is APPROVE:
           - Run full module test suite for $ARGUMENTS.
+          - Read diagnosis classification summary.
+          - Read current test summary.
           - Compare with diagnosis spec baseline.
+          - Compare skipped tests against remaining UNIMPLEMENTED diagnosis items.
+
+          Hard PASS rules:
+          - 0 failed is not sufficient for PASS.
+          - PASS is forbidden if any approved diagnosis UNIMPLEMENTED item remains skipped.
+          - Diagnosis-backed UNIMPLEMENTED skipped tests count as Remaining, not Blocked.
+          - UNIMPLEMENTED skipped tests count as Remaining.
+          - Before writing Module complete: yes, read diagnosis classification summary, read current test summary, compare skipped tests against remaining UNIMPLEMENTED diagnosis items, and if any UNIMPLEMENTED skipped remains, Module complete must be no.
+
+          Verdict mapping:
+          - review REQUEST_CHANGES -> Verdict: BLOCKED, Failure reason: BLOCKED_BY_REVIEW.
+          - 0 failed + remaining UNIMPLEMENTED skipped > 0 -> Verdict: CONTINUE, Module complete: no, Next Round Focus: next diagnosis-backed UNIMPLEMENTED test/evidence path.
+          - progress contradicts diagnosis/test evidence -> Verdict: BLOCKED, Failure reason: PROGRESS_DIAGNOSIS_CONFLICT.
+          - new red caused by current round -> Verdict: REQUEST_CHANGES.
+          - new red appears to be newly exposed baseline gap AND remaining count decreases -> Verdict: CONTINUE.
+          - ambiguous new red -> Verdict: BLOCKED, Failure reason: AMBIGUOUS_NEW_RED.
+          - 0 failed + no remaining diagnosis-backed UNIMPLEMENTED skipped + no unresolved actionable BUG / UNIMPLEMENTED items + only TEST_BUG / INFRA / REMOVED / explicitly user-approved blocked items remain -> Verdict: PASS.
 
           Output report:
           - Verdict: PASS / CONTINUE / REQUEST_CHANGES / BLOCKED
           - Review: APPROVE / REQUEST_CHANGES
+          - Module complete: yes/no
           - Fixed: N tests now green
           - New red: M tests that regressed
-          - Blocked: K tests skipped (see blocked_items.md)
-          - Remaining: R tests still failing
+          - Blocked: K tests skipped for TEST_BUG / INFRA / REMOVED / user-approved blocked reasons
+          - Remaining: R tests still failing or diagnosis-backed UNIMPLEMENTED skipped
+          - Remaining UNIMPLEMENTED skipped: U diagnosis-backed skipped tests still remaining
           - Failure reason: why progress stopped, if any
           - Next round focus: the priority/classification/test file to continue with, if verdict is CONTINUE or REQUEST_CHANGES
 
           Always write or update .dna/fix-progress/$ARGUMENTS.md before finishing, even when review blocks verification. Include these sections:
-          - Current verdict: Verdict, Review, Last fix commit, Updated at
-          - Round summary: Fixed, New red, Blocked, Remaining, Failure reason
+          - Current verdict: Verdict, Review, Module complete, Last fix commit, Updated at
+          - Round summary: Fixed, New red, Blocked, Remaining, Remaining UNIMPLEMENTED skipped, Failure reason
           - Next Round Focus: non-empty when Verdict is CONTINUE or REQUEST_CHANGES
           - Review Issues To Address First: reviewer issues when Review is REQUEST_CHANGES
           - Failed approaches / Do not repeat: approaches that failed or caused regressions
           - Evidence: commands run and relevant paths
 
-          If Verdict is PASS, mark the module complete in .dna/fix-progress/$ARGUMENTS.md.
+          Always write or update .dna/fix-progress/$ARGUMENTS.json before finishing, even when review blocks verification. The JSON must match this contract exactly:
+          ```json
+          {
+            "verdict": "PASS | CONTINUE | REQUEST_CHANGES | BLOCKED",
+            "review": "APPROVE | REQUEST_CHANGES",
+            "module": "$ARGUMENTS",
+            "round": 1,
+            "run_id": "controller run id or manual-run",
+            "diagnosis_artifact": ".dna/specs/diagnosis-$ARGUMENTS.md",
+            "review_artifact": ".dna/reviews/$ARGUMENTS-review.json",
+            "git_head": null,
+            "module_complete": false,
+            "fixed_count": 0,
+            "new_red_count": 0,
+            "blocked_count": 0,
+            "remaining_count": 0,
+            "remaining_unimplemented_skipped_count": 0,
+            "next_round_focus": "",
+            "failure_reason": "",
+            "last_fix_commit": null,
+            "evidence_paths": [],
+            "updated_at": "ISO-8601 timestamp"
+          }
+          ```
+
+          JSON field rules:
+          - module is required and must equal $ARGUMENTS.
+          - round is required and must be the current fix-loop round number, or 1 for manual fix invocation.
+          - run_id is required and must identify the current controller run, or "manual-run" for manual fix invocation.
+          - diagnosis_artifact is required and must equal .dna/specs/diagnosis-$ARGUMENTS.md.
+          - review_artifact is required and must equal .dna/reviews/$ARGUMENTS-review.json.
+          - git_head is the current git HEAD after verification if available, otherwise null.
+          - verdict is required and must be PASS, CONTINUE, REQUEST_CHANGES, or BLOCKED.
+          - review is required and must be APPROVE or REQUEST_CHANGES.
+          - module_complete is a required boolean and must match the Markdown Module complete value.
+          - fixed_count, new_red_count, blocked_count, remaining_count, and remaining_unimplemented_skipped_count are required non-negative integers.
+          - next_round_focus is required and non-empty when verdict is CONTINUE or REQUEST_CHANGES; empty string is allowed when verdict is PASS.
+          - failure_reason is required and non-empty when verdict is BLOCKED or REQUEST_CHANGES; empty string is allowed when verdict is PASS or CONTINUE.
+          - last_fix_commit is the commit hash string if available, otherwise null.
+          - evidence_paths is a required array of command/test/diagnosis/review evidence paths.
+          - updated_at is a required ISO timestamp string.
+
+          If Verdict is PASS, mark Module complete: yes in .dna/fix-progress/$ARGUMENTS.md and set module_complete true in .dna/fix-progress/$ARGUMENTS.json.
 
           Verdict meanings:
-          - PASS: all approved diagnosis items are fixed or intentionally skipped
+          - PASS: all actionable BUG / UNIMPLEMENTED items are fixed, with no diagnosis-backed UNIMPLEMENTED skipped tests remaining; only TEST_BUG / INFRA / REMOVED / explicitly user-approved blocked items may remain
           - CONTINUE: this round made progress and remaining approved items should continue in the next fix invocation
-          - REQUEST_CHANGES: review or verification found issues in this round's changes
-          - BLOCKED: user confirmation is needed or repeated no-progress prevents safe continuation
+          - REQUEST_CHANGES: verification found issues in this round's changes
+          - BLOCKED: review blocks verification, user confirmation is needed, evidence conflicts, or repeated no-progress prevents safe continuation
         handoff:
           consumes:
-            - type: summary
+            - type: file
               from: review_changes
+              path: ".dna/reviews/$ARGUMENTS-review.json"
               description: "Review verdict from fix_reviewer"
             - type: git_commit
               from: fix_bugs
@@ -564,5 +669,9 @@ workflows:
             - type: file
               path: ".dna/fix-progress/$ARGUMENTS.md"
               description: "Durable cumulative fix-progress artifact for the next fix invocation"
+            - type: file
+              path: ".dna/fix-progress/$ARGUMENTS.json"
+              description: "Machine-readable fix-progress verdict sidecar for controllers"
             - type: summary
+              required: false
               description: "Verification report with test delta"

package/dist/templates/full-pipeline.dna.yaml

@@ -142,6 +142,7 @@ workflow:
             description: "Implementation to test"
         produces:
           - type: test_result
+            name: test_results
             description: "Test results after fixes"
     - id: review
       role: reviewer
@@ -151,9 +152,11 @@ workflow:
         consumes:
           - type: test_result
             from: test
+            name: test_results
             description: "Test results to review"
         produces:
           - type: summary
+            required: false
             description: "Review verdict (PASS/FAIL)"
   transitions:
     - from: review

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "intentdna",
-  "version": "1.5.20",
+  "version": "1.6.0",
   "description": "Intent DNA — Declarative policy layer for AI agent behavior",
   "type": "module",
   "main": "dist/index.js",

package/spec/schema-spec.md

@@ -239,6 +239,8 @@ output_schema:
 
 Workflows define multi-step agent pipelines with dependency ordering, parallel execution, retry, and artifact handoff.
 
+Handoff artifacts compile into local session-scoped manifests. A required artifact must have a stable identity from `path`, `name`, or `artifact_id`; required `summary`/`state` artifacts without one are invalid. Runtime enforcement treats the manifest as the local fact source and uses workflow inputs such as `$ARGUMENTS` before resolving artifact keys. User-provided workflow `inputs` used in artifact path templates are constrained to safe single path components; trusted `resolved_variables` may still provide configured path prefixes. This blocks prompt/path/metadata-only satisfaction; cryptographic protection against direct manifest forgery is deferred to signing/registry work.
+
 ### Single Workflow
 ```yaml
 workflow:
@@ -266,6 +268,7 @@ workflow:
             description: "Implementation to review"
         produces:
           - type: summary
+            required: false
             description: "Review verdict"
 
   transitions: