@guanlin4924/devflow-pi 0.1.86 → 0.1.88

package/plugins/devflow-core/skills/workflow-orchestrator/scripts/review_run.py

@@ -7,6 +7,7 @@ pretend that a source excerpt alone proves runtime behaviour.
 from __future__ import annotations
 
 import hashlib
+import hmac
 import json
 import os
 from pathlib import Path
@@ -36,6 +37,29 @@ def canonical(value: Any) -> str:
     return json.dumps(value, ensure_ascii=False, sort_keys=True, separators=(",", ":"))
 
 
+def verified_review_binding(request: dict) -> tuple[bool, str]:
+    """A completed reviewer task, not the caller, owns the submitted report."""
+    binding = request.get("core_review_binding")
+    secret = os.environ.get("DEVFLOW_REVIEW_BINDING_KEY", "")
+    if not isinstance(binding, dict) or not secret:
+        return False, "missing Core-bound independent reviewer task"
+    fields = ("agent", "feature", "contract_id", "task_call_id", "report_sha256", "signature")
+    if any(not isinstance(binding.get(field), str) or not binding[field] for field in fields):
+        return False, "incomplete Core reviewer task binding"
+    target = request.get("target")
+    if binding["agent"] != "devflow-code-reviewer" or not isinstance(target, dict) or binding["feature"] != target.get("feature"):
+        return False, "reviewer identity or feature does not match the report"
+    unsigned_report = {key: value for key, value in request.items() if key != "core_review_binding"}
+    report_hash = digest(canonical(unsigned_report).encode())
+    if not hmac.compare_digest(report_hash, binding["report_sha256"]):
+        return False, "review report differs from the completed reviewer task"
+    unsigned_binding = {key: binding[key] for key in fields if key != "signature"}
+    expected = hmac.new(secret.encode(), canonical(unsigned_binding).encode(), hashlib.sha256).hexdigest()
+    if not hmac.compare_digest(expected, binding["signature"]):
+        return False, "invalid Core reviewer task signature"
+    return True, ""
+
+
 def safe_file(root: Path, relative: str) -> Path:
     if not isinstance(relative, str) or not relative or Path(relative).is_absolute() or "\x00" in relative:
         raise ValueError("path must be a nonempty project-relative path")
@@ -136,6 +160,10 @@ def db_open(root: Path) -> sqlite3.Connection:
         id INTEGER PRIMARY KEY, chain_id TEXT, status TEXT NOT NULL, detail_json TEXT NOT NULL,
         created_at TEXT NOT NULL DEFAULT CURRENT_TIMESTAMP
       );
+      CREATE TABLE IF NOT EXISTS reviewer_tasks (
+        task_call_id TEXT PRIMARY KEY, chain_id TEXT NOT NULL, round INTEGER NOT NULL,
+        report_sha256 TEXT NOT NULL
+      );
     """)
     return db
 
@@ -221,6 +249,35 @@ def verify_core_result(root: Path, feature: str, evidence: Any, expected_status:
         return False, f"unreadable Core acceptance result: {exc}"
 
 
+def verify_core_impact(root: Path, feature: str, evidence: dict, impact: str, trigger: str) -> tuple[bool, str]:
+    """A failed TC alone does not establish the reviewer's causal/impact label."""
+    try:
+        suite = read_json(safe_file(root, f".planning/requirements/{feature}/tests.json"))
+        node = next((n for n in suite.get("nodes", []) if isinstance(n, dict) and n.get("id") == evidence.get("tc_id")), None)
+        data = node.get("data", {}) if node else {}
+        if data.get("review_impact") == impact and data.get("review_trigger") == trigger:
+            return True, ""
+        return False, "Core frozen TC does not confirm the claimed failure trigger and business impact"
+    except (OSError, ValueError, TypeError, json.JSONDecodeError) as exc:
+        return False, f"Core frozen impact cannot be verified: {exc}"
+
+
+def declared_change_paths(root: Path, feature: str) -> tuple[set[str], str, str]:
+    relative = f".planning/features/{feature}/变更清单.md"
+    source = safe_file(root, relative)
+    declarations = re.findall(r"^\s*-\s*\*\*文件\*\*\s*[：:]\s*(.+)$", source.read_text(encoding="utf-8"), re.MULTILINE)
+    paths = set()
+    for declaration in declarations:
+        matches = re.findall(r"`([^`]+)`", declaration)
+        if len(matches) != 1 or declaration.strip() != f"`{matches[0]}`":
+            raise ValueError("ambiguous change-list file declaration")
+        safe_file(root, matches[0])
+        paths.add(matches[0])
+    if not paths:
+        raise ValueError("change list has no unambiguous source file declarations")
+    return paths, relative, digest(source.read_bytes())
+
+
 def verify_coverage(root: Path, feature: str, required: list[str], coverage_evidence: Any) -> tuple[set[str], list[str]]:
     if not isinstance(coverage_evidence, dict):
         return set(), ["Each review surface and scenario needs a distinct Core TC binding"]
@@ -241,7 +298,10 @@ def verify_coverage(root: Path, feature: str, required: list[str], coverage_evid
         suite = read_json(safe_file(root, f".planning/requirements/{feature}/tests.json"))
         node = next((n for n in suite.get("nodes", []) if isinstance(n, dict) and n.get("id") == ref["tc_id"]), None)
         types = {"scenario:normal": {"api", "page"}, "scenario:unauthorized": {"permission"}, "scenario:failure_rollback": {"resilience", "boundary"}, "scenario:compatibility": {"migration"}}
-        if not node or node.get("kind") != "test_case" or node.get("status") != "confirmed" or node.get("data", {}).get("test_type") not in types.get(tag, set()):
+        data = node.get("data", {}) if node else {}
+        explicit = tag in data.get("review_coverage_tags", []) if isinstance(data.get("review_coverage_tags"), list) else False
+        typed = data.get("test_type") in types.get(tag, set())
+        if not node or node.get("kind") != "test_case" or node.get("status") != "confirmed" or not (explicit or typed):
             issues.append(f"{tag}: frozen TC type cannot prove this coverage obligation; request human confirmation")
             continue
         covered.add(tag)
@@ -256,10 +316,139 @@ def suite_has_failures(root: Path, feature: str) -> bool:
         return True
 
 
+def gate_status(root: Path, feature: str, contract_id: str | None = None) -> dict:
+    """Read only the Core ledger; prose reports never confer approval."""
+    root = root.resolve(strict=True)
+    if not contract_id:
+        return {"status": "human_confirmation", "reason": "Current Core review contract identity is required"}
+    try:
+        contract = read_json(safe_file(root, f".planning/contracts/{feature}/{contract_id}.json"))
+        if contract.get("origin") != "core_execution_contract" or contract.get("step") != "code-review" or contract.get("feature") != feature or contract.get("contract_id") != contract_id:
+            return {"status": "core_contract_conflict", "reason": "Review contract identity is not a frozen Core code-review contract"}
+        sys.path.insert(0, str(Path(__file__).resolve().parent))
+        from review_preflight import load_state
+        state = load_state(safe_file(root, ".planning/STATE.md"))
+        graph = read_json(safe_file(root, f".planning/requirements/{feature}/governance.json"))
+        workflow = safe_file(root, contract["workflow_path"])
+        if (state.get("current_feature") != feature or state.get("current_step") != "code-review"
+            or state.get("revision") != contract.get("state_revision")
+            or graph.get("revision") != contract.get("graph_revision")
+            or digest(workflow.read_bytes()) != contract.get("workflow_fingerprint")):
+            return {"status": "stale_evidence", "reason": "Review contract differs from current Core state, graph, or workflow"}
+    except (OSError, ValueError, KeyError, TypeError, ImportError, json.JSONDecodeError) as exc:
+        return {"status": "human_confirmation", "reason": f"Current Core review contract cannot be verified: {exc}"}
+    database = root / ".planning" / "review-runs" / "review-runs.sqlite"
+    if not database.is_file() or database.is_symlink():
+        return {"status": "human_confirmation", "reason": "No Core review round exists"}
+    db = sqlite3.connect(f"file:{database}?mode=ro", uri=True)
+    try:
+        rows = db.execute("SELECT result_json FROM rounds ORDER BY created_at DESC, rowid DESC").fetchall()
+    finally:
+        db.close()
+    for (raw,) in rows:
+        result = json.loads(raw)
+        task = result.get("reviewer_task", {})
+        if task.get("feature") != feature:
+            continue
+        if task.get("contract_id") != contract_id:
+            continue
+        report_path = next((item for item in contract.get("allowed_paths", []) if isinstance(item, str) and item.endswith("/REVIEW.md")), None)
+        if not report_path:
+            return {"status": "core_contract_conflict", "reason": "Core contract has no REVIEW.md output"}
+        try:
+            if safe_file(root, report_path).read_text(encoding="utf-8") != review_markdown(result):
+                return {"status": "core_contract_conflict", "reason": "REVIEW.md differs from Core review round"}
+        except (OSError, ValueError):
+            return {"status": "core_contract_conflict", "reason": "Core-derived REVIEW.md is missing"}
+        if task.get("agent") != "devflow-code-reviewer":
+            return {"status": "core_contract_conflict", "reason": "Review round lacks independent reviewer"}
+        fixed = result.get("fixed_versions", {})
+        index_generation = fixed.get("code_index_generation")
+        if index_generation is not None:
+            try:
+                snapshot = read_json(safe_file(root, ".planning/code-index/SNAPSHOT.json"))
+                if snapshot.get("generation") != index_generation:
+                    return {"status": "stale_evidence", "reason": "Code-index generation changed after review", "chain_id": result.get("chain_id")}
+            except (OSError, ValueError, json.JSONDecodeError):
+                return {"status": "stale_evidence", "reason": "Code-index snapshot unavailable after review", "chain_id": result.get("chain_id")}
+        for relative, expected in fixed.get("source", {}).items():
+            try:
+                if digest(safe_file(root, relative).read_bytes()) != expected:
+                    return {"status": "stale_evidence", "reason": f"Source changed after review: {relative}", "chain_id": result.get("chain_id")}
+            except (OSError, ValueError):
+                return {"status": "stale_evidence", "reason": f"Source unavailable after review: {relative}", "chain_id": result.get("chain_id")}
+        for authority in fixed.get("authority", []):
+            try:
+                if digest(safe_file(root, authority["authority_path"]).read_bytes()) != authority["authority_sha256"]:
+                    return {"status": "stale_evidence", "reason": "Authority changed after review", "chain_id": result.get("chain_id")}
+            except (OSError, ValueError, KeyError):
+                return {"status": "stale_evidence", "reason": "Authority unavailable after review", "chain_id": result.get("chain_id")}
+        for relative, expected in fixed.get("core_results", {}).items():
+            try:
+                if digest(safe_file(root, relative).read_bytes()) != expected:
+                    return {"status": "stale_evidence", "reason": f"Core test evidence changed after review: {relative}", "chain_id": result.get("chain_id")}
+            except (OSError, ValueError):
+                return {"status": "stale_evidence", "reason": f"Core test evidence unavailable: {relative}", "chain_id": result.get("chain_id")}
+        plan = fixed.get("change_plan", {})
+        try:
+            if not plan or digest(safe_file(root, plan["path"]).read_bytes()) != plan["sha256"]:
+                return {"status": "stale_evidence", "reason": "Core change-list scope changed after review", "chain_id": result.get("chain_id")}
+        except (OSError, ValueError, KeyError, TypeError):
+            return {"status": "stale_evidence", "reason": "Core change-list scope unavailable", "chain_id": result.get("chain_id")}
+        status = result.get("status")
+        verdict = "pass" if status == "clean" and result.get("automatic_release_allowed") else ("terminate_task" if status == "terminated_p0" else "repair")
+        return {"status": verdict, "chain_id": result.get("chain_id"), "round": result.get("round"), "evidence_chain": {"reviewer_task": task, "fixed_versions": fixed, "coverage": result.get("coverage")}, "open_findings": [f for f in result.get("findings", []) if f.get("state") != "resolved"], "next_action": "advance" if verdict == "pass" else ("stop_current_task" if verdict == "terminate_task" else "repair_and_re_review")}
+    return {"status": "human_confirmation", "reason": "No Core review round exists for this feature and contract"}
+
+
+def review_markdown(result: dict) -> str:
+    lines = ["# Code Review", "", "> Core-derived view; the review ledger is authoritative.", "", f"- Chain: `{result['chain_id']}`", f"- Round: {result['round']}", f"- Decision: `{result['status']}`", f"- Reviewer task: `{result['reviewer_task']['task_call_id']}`", "", "## Open findings", ""]
+    for finding in result.get("findings", []):
+        if finding.get("state") != "resolved":
+            source = finding.get("source", {})
+            lines.append(f"- `{finding['finding_id']}` {finding['severity']}: {source.get('path')}:{source.get('line')} — {finding.get('actual', '')}")
+    if lines[-1] == "":
+        lines.append("- None")
+    lines.extend(["", "## Next action", "", "Advance to the next workflow step." if result.get("status") == "clean" else "Repair findings and request a new independent review.", ""])
+    return "\n".join(lines)
+
+
+def render_review(root: Path, feature: str, contract_id: str) -> dict:
+    root = root.resolve(strict=True)
+    contract = read_json(safe_file(root, f".planning/contracts/{feature}/{contract_id}.json"))
+    path = next((item for item in contract.get("allowed_paths", []) if isinstance(item, str) and item.endswith("/REVIEW.md")), None)
+    if not path:
+        return {"status": "core_contract_conflict", "reason": "Core contract has no REVIEW.md output"}
+    database = root / ".planning/review-runs/review-runs.sqlite"
+    if not database.is_file():
+        return {"status": "human_confirmation", "reason": "No Core review round exists"}
+    db = sqlite3.connect(f"file:{database}?mode=ro", uri=True)
+    try:
+        rows = db.execute("SELECT result_json FROM rounds ORDER BY created_at DESC, rowid DESC").fetchall()
+    finally:
+        db.close()
+    for (raw,) in rows:
+        result = json.loads(raw)
+        task = result.get("reviewer_task", {})
+        if task.get("feature") == feature and task.get("contract_id") == contract_id:
+            destination = root / path
+            if destination.is_symlink() or not destination.resolve().is_relative_to(root):
+                return {"status": "core_contract_conflict", "reason": "Unsafe review output path"}
+            destination.parent.mkdir(parents=True, exist_ok=True)
+            temporary = destination.with_name(destination.name + ".core-tmp")
+            if temporary.exists() or temporary.is_symlink():
+                return {"status": "core_contract_conflict", "reason": "Review output temporary path is occupied"}
+            temporary.write_text(review_markdown(result), encoding="utf-8")
+            os.replace(temporary, destination)
+            return {"status": "rendered", "path": path, "chain_id": result["chain_id"], "round": result["round"]}
+    return {"status": "human_confirmation", "reason": "No Core review round exists for this contract"}
+
+
 def run(root: Path, request: dict) -> dict:
     root = root.resolve(strict=True)
     if not isinstance(request, dict):
         raise ValueError("request must be an object")
+    bound, binding_issue = verified_review_binding(request)
     if "round" in request or "findings" in request or "severity" in request:
         raise ValueError("Core assigns rounds and findings; caller-supplied authority fields are forbidden")
     target = request.get("target")
@@ -275,6 +464,9 @@ def run(root: Path, request: dict) -> dict:
         raise ValueError("review_plan.questions must be nonempty")
     if not all(isinstance(q, str) and q.strip() for q in plan["questions"]):
         raise ValueError("review questions must be strings")
+    review_questions = request.get("review_questions", [])
+    if not isinstance(review_questions, list) or not all(isinstance(q, str) and q.strip() for q in review_questions):
+        raise ValueError("review_questions must be a list of nonempty strings")
     surfaces, scenarios = plan.get("surfaces"), plan.get("required_scenarios")
     if not isinstance(surfaces, list) or not set(surfaces) <= SURFACES or not isinstance(scenarios, list) or not set(scenarios) <= SCENARIOS:
         raise ValueError("review plan uses an unsupported surface or scenario")
@@ -284,6 +476,14 @@ def run(root: Path, request: dict) -> dict:
     if not isinstance(expectations, list):
         raise ValueError("expectations must be a list")
     accepted, questions = validate_authority(root, target["feature"], expectations)
+    try:
+        declared_paths, change_plan_path, change_plan_hash = declared_change_paths(root, target["feature"])
+        missing_paths = declared_paths - set(paths)
+        if missing_paths:
+            questions.append("Review target omits change-list files: " + ", ".join(sorted(missing_paths)))
+    except (OSError, ValueError, UnicodeError) as exc:
+        declared_paths, change_plan_path, change_plan_hash = set(), None, None
+        questions.append(f"Core change-list review scope cannot be verified: {exc}")
     chain_id = request.get("chain_id")
     if chain_id is not None and (not isinstance(chain_id, str) or not chain_id.startswith("rev_")):
         raise ValueError("invalid chain_id")
@@ -311,11 +511,14 @@ def run(root: Path, request: dict) -> dict:
                 raise ValueError("three completed review rounds exhausted")
         else:
             round_no = 1
-        if questions:
-            result = response("awaiting_human_confirmation", chain_id, round_no, questions=questions)
+        if questions or review_questions:
+            result = response("awaiting_human_confirmation", chain_id, round_no, questions=[*questions, *review_questions])
             db.execute("INSERT INTO events(chain_id,status,detail_json) VALUES(?,?,?)", (chain_id, result["status"], canonical(result)))
             db.commit()
             return result
+        binding = request.get("core_review_binding") if bound else None
+        if bound and db.execute("SELECT 1 FROM reviewer_tasks WHERE task_call_id=?", (binding["task_call_id"],)).fetchone():
+            raise ValueError("reviewer task was already consumed by another review round")
         if not chain_id:
             # A new ID cannot erase an unresolved P0 for the same target and authority.
             for old_id, raw in db.execute("SELECT c.chain_id,r.result_json FROM chains c JOIN rounds r ON r.chain_id=c.chain_id AND r.round=(SELECT MAX(round) FROM rounds WHERE chain_id=c.chain_id) WHERE c.target_key=? AND c.authority_key=?", (target_key, authority_key)):
@@ -370,6 +573,10 @@ def run(root: Path, request: dict) -> dict:
                 valid, reason = verify_core_result(root, target["feature"], item.get("runtime_evidence"), "failed", authority_node_id)
                 if not valid:
                     evidence_issues.append(f"{evidence['path']}:{evidence['line']}: {reason}")
+                else:
+                    causal, reason = verify_core_impact(root, target["feature"], item["runtime_evidence"], item["impact"], item.get("trigger"))
+                    if not causal:
+                        evidence_issues.append(f"{evidence['path']}:{evidence['line']}: {reason}; human confirmation required")
             fingerprint = digest(canonical([item["expectation_id"], evidence["path"], evidence["line"], item["impact"], item.get("trigger")]).encode())[:20]
             findings.append({"finding_id": "F-" + fingerprint, "expectation_id": item["expectation_id"], "severity": IMPACTS[item["impact"]], "severity_source": "reviewer_claim", "impact": item["impact"], "source": evidence, "actual": item.get("actual"), "trigger": item.get("trigger"), "reproduction": item.get("reproduction"), "improvement": item.get("improvement"), "confidence": "source_and_core_tc_verified_causality_unverified" if item.get("runtime_evidence") else "source_verified_inference", "epistemic_status": "reviewer_interpretation", "verified_boundaries": ["approved_core_graph_expectation", "source_line_and_hash", "core_tc_identity_and_status"] if item.get("runtime_evidence") else ["approved_core_graph_expectation", "source_line_and_hash"]})
         gaps = sorted(set(required) - covered)
@@ -390,13 +597,12 @@ def run(root: Path, request: dict) -> dict:
                     evidence_issues.append(f"{old_id}: source unchanged; cannot resolve by removing a report observation")
                 else:
                     regression = request.get("regression_evidence", {}).get(old_id)
-                    try:
-                        if not isinstance(regression, dict) or digest(safe_file(root, regression.get("path")).read_bytes()) != regression.get("sha256"):
-                            evidence_issues.append(f"{old_id}: missing verified regression evidence")
-                        else:
-                            findings.append({**old, "state": "resolved"})
-                    except (OSError, ValueError, TypeError):
-                        evidence_issues.append(f"{old_id}: missing verified regression evidence")
+                    authority_node_id = next((e["node_id"] for e in accepted if e["id"] == old.get("expectation_id")), None)
+                    valid, reason = verify_core_result(root, target["feature"], regression, "passed", authority_node_id)
+                    if not valid:
+                        evidence_issues.append(f"{old_id}: missing Core passed regression TC: {reason}")
+                    else:
+                        findings.append({**old, "state": "resolved"})
         for rel, start_hash in versions.items():
             if digest(safe_file(root, rel).read_bytes()) != start_hash:
                 result = response("stale_evidence", chain_id, round_no, questions=[f"Source changed during review: {rel}"])
@@ -424,33 +630,24 @@ def run(root: Path, request: dict) -> dict:
             return result
         open_findings = [f for f in findings if f["state"] != "resolved"]
         summary = {p: sum(f["severity"] == p for f in open_findings) for p in ("P0", "P1", "P2", "P3")}
-        if open_findings:
-            # Source and TC hashes establish identity, not that the reviewer's
-            # causal story or impact category is true. Until Core has a
-            # verifiable independent-review attestation, these are proposals,
-            # not completed rounds that can terminate a task after three tries.
-            result = response(
-                "awaiting_human_confirmation", chain_id, round_no,
-                questions=["Independent reviewer identity and business causality/impact require confirmation"],
-                proposed_findings=open_findings,
-                proposed_risk_summary=summary,
-                proposed_improvement_plan=[{"finding_id": f["finding_id"], "severity": f["severity"], "action": f["improvement"]} for f in open_findings],
-                coverage={"covered": sorted(covered), "gaps": gaps},
-            )
+        if not bound:
+            result = response("awaiting_human_confirmation", chain_id, round_no, questions=[binding_issue], proposed_findings=open_findings, proposed_risk_summary=summary, coverage={"covered": sorted(covered), "gaps": gaps})
             db.execute("INSERT INTO events(chain_id,status,detail_json) VALUES(?,?,?)", (chain_id, result["status"], canonical(result)))
             db.commit()
             return result
-        if not summary["P0"] and not summary["P1"]:
+        if not open_findings:
             reasons = [*evidence_issues]
             if suite_has_failures(root, target["feature"]):
                 reasons.append("The same Core acceptance run contains failed or blocked TCs; clean cannot be inferred from selected passed cases")
-            reasons.append("No Core-verifiable independent reviewer identity or causal semantic attestation is available; clean needs human confirmation")
-            result = response("awaiting_human_confirmation", chain_id, round_no, questions=reasons, coverage={"covered": sorted(covered), "gaps": gaps})
-            db.execute("INSERT INTO events(chain_id,status,detail_json) VALUES(?,?,?)", (chain_id, result["status"], canonical(result)))
-            db.commit()
-            return result
-        status = "terminated_p0" if round_no == 3 and summary["P0"] else ("findings" if summary["P0"] or summary["P1"] else "clean")
-        result = response(status, chain_id, round_no, fixed_versions={"source": versions, "authority": accepted, "code_index_generation": index_discovery.get("generation")}, index_discovery=index_discovery, coverage={"covered": sorted(covered), "gaps": gaps, "index_gap": index_discovery.get("reason") if index_discovery["status"] != "located" else None}, findings=findings, risk_summary=summary, improvement_plan=[{"finding_id": f["finding_id"], "severity": f["severity"], "action": f["improvement"]} for f in sorted(open_findings, key=lambda x: x["severity"])], all_p0_resolved=summary["P0"] == 0, severity_source="reviewer_claim", semantic_causality_verified=False, automatic_release_allowed=False, termination_scope="review_chain_only" if status == "terminated_p0" else None, history=[{"round": r, "status": json.loads(raw)["status"]} for r, raw in db.execute("SELECT round,result_json FROM rounds WHERE chain_id=? ORDER BY round", (chain_id,))])
+            if reasons:
+                result = response("awaiting_human_confirmation", chain_id, round_no, questions=reasons, coverage={"covered": sorted(covered), "gaps": gaps})
+                db.execute("INSERT INTO events(chain_id,status,detail_json) VALUES(?,?,?)", (chain_id, result["status"], canonical(result)))
+                db.commit()
+                return result
+        status = "terminated_p0" if round_no == 3 and summary["P0"] else ("findings" if open_findings else "clean")
+        core_results = {rel: digest(safe_file(root, rel).read_bytes()) for rel in (f".planning/requirements/{target['feature']}/tests.json", f".planning/test-runs/{target['feature']}/acceptance-results.json")}
+        result = response(status, chain_id, round_no, fixed_versions={"source": versions, "authority": accepted, "core_results": core_results, "change_plan": {"path": change_plan_path, "sha256": change_plan_hash, "declared_paths": sorted(declared_paths)}, "code_index_generation": index_discovery.get("generation")}, index_discovery=index_discovery, coverage={"covered": sorted(covered), "gaps": gaps, "index_gap": index_discovery.get("reason") if index_discovery["status"] != "located" else None}, findings=findings, risk_summary=summary, improvement_plan=[{"finding_id": f["finding_id"], "severity": f["severity"], "action": f["improvement"]} for f in sorted(open_findings, key=lambda x: x["severity"])], all_p0_resolved=summary["P0"] == 0, severity_source="independent_reviewer_core_evidence_bound", semantic_causality_verified=True, automatic_release_allowed=status == "clean", termination_scope="current_task_only" if status == "terminated_p0" else None, reviewer_task={key: binding[key] for key in ("agent", "feature", "contract_id", "task_call_id", "report_sha256")}, history=[{"round": r, "status": json.loads(raw)["status"]} for r, raw in db.execute("SELECT round,result_json FROM rounds WHERE chain_id=? ORDER BY round", (chain_id,))])
+        db.execute("INSERT INTO reviewer_tasks(task_call_id,chain_id,round,report_sha256) VALUES(?,?,?,?)", (binding["task_call_id"], chain_id, round_no, binding["report_sha256"]))
         db.execute("INSERT INTO rounds(chain_id,round,result_json) VALUES(?,?,?)", (chain_id, round_no, canonical(result)))
         if status == "terminated_p0":
             db.execute("UPDATE chains SET terminal=? WHERE chain_id=?", (status, chain_id))
@@ -466,7 +663,12 @@ def run(root: Path, request: dict) -> dict:
 if __name__ == "__main__":
     try:
         project = Path(sys.argv[1] if len(sys.argv) > 1 else os.getcwd())
-        print(canonical(run(project, json.load(sys.stdin))))
+        if len(sys.argv) > 3 and sys.argv[2] == "--gate":
+            print(canonical(gate_status(project.resolve(strict=True), sys.argv[3], sys.argv[4] if len(sys.argv) > 4 else None)))
+        elif len(sys.argv) > 4 and sys.argv[2] == "--render":
+            print(canonical(render_review(project, sys.argv[3], sys.argv[4])))
+        else:
+            print(canonical(run(project, json.load(sys.stdin))))
     except (ValueError, OSError, json.JSONDecodeError, sqlite3.Error, StaleEvidence) as error:
         print(canonical({"status": "error", "error": str(error)}))
         sys.exit(2)

package/plugins/devflow-core/skills/workflow-orchestrator/scripts/validate_state.rb

@@ -14,6 +14,7 @@ require_relative "checker_registry"
 require_relative "requirement_coverage"
 require_relative "workflow_gate_contract"
 require_relative "parallel_run"
+require_relative "checkpoint_meta"
 
 module WorkflowStateValidator
   REQUIRED_STATE_FIELDS = %w[
@@ -29,6 +30,32 @@ module WorkflowStateValidator
   TERMINAL_ARTIFACT_STATUSES = %w[approved completed passed].freeze
   NON_TERMINAL_ARTIFACT_STATUSES = %w[planned generated pending_rerun stale failed].freeze
   TERMINAL_WORKFLOW_STATUSES = %w[completed failed cancelled].freeze
+  # Core quality checks are advisory by default. Keep only structural and
+  # write-safety failures hard; AI review decides whether business-quality
+  # drift, stale metadata, or incomplete generated evidence should pause work.
+  SOFT_GATE_MODE = ENV.fetch("DEVFLOW_SOFT_GATES", "1") != "0"
+  HARD_VALIDATION_CODES = %w[
+    invalid_control_path
+    missing_state_file
+    missing_state_front_matter
+    invalid_state_yaml
+    missing_state_field
+    invalid_workflow_config
+    workflow_step_unknown
+    phase_ownership_conflict
+  ].freeze
+
+  def self.hard_validation_code?(code)
+    value = code.to_s
+    return true if HARD_VALIDATION_CODES.include?(value)
+    return true if value.start_with?("artifact_registry_", "parallel_", "PARALLEL_")
+    return true if value.start_with?("duplicate_", "unknown_", "missing_tracked_step")
+    return true if value.start_with?("step_in_", "completed_", "terminal_", "non_completed_")
+    return true if value.start_with?("waiting_checkpoint_")
+    return true if %w[missing_required_output output_not_registered].include?(value)
+
+    false
+  end
 
   module_function
 
@@ -89,7 +116,7 @@ module WorkflowStateValidator
     end
     validate_dependencies(issues, step_by_id, completed_ids)
     validate_completed_outputs(issues, root, state, step_by_id, completed_ids, artifact_by_path, strict_artifact_status, workflow)
-    validate_waiting_checkpoint(issues, root, state, step_by_id, artifact_by_path, strict_artifact_status)
+    validate_waiting_checkpoint(issues, root, state, workflow, step_by_id, artifact_by_path, strict_artifact_status)
     validate_current_in_progress_output_drift(issues, root, state, step_by_id, artifact_by_path)
     graph_gate = workflow.dig("governance", "requirement_graph_gate").to_s
     governed_steps = %w[requirement-baseline api-contract architecture-feasibility integration-verify change-plan execute run-acceptance-tests]
@@ -112,6 +139,15 @@ module WorkflowStateValidator
       end
     end
 
+    if SOFT_GATE_MODE
+      issues.each do |entry|
+        next unless entry["severity"] == "error"
+        next if hard_validation_code?(entry["code"])
+
+        entry["severity"] = "warning"
+        entry["soft_gate"] = true
+      end
+    end
     ok = issues.none? { |entry| entry["severity"] == "error" }
     result(ok, state_path, workflow_path, registry_path, issues)
   end
@@ -357,15 +393,26 @@ unless %w[prepared claimed consumed superseded failed].include?(status)
 
     declared.each do |step_id, checker_ids|
       Array(checker_ids).each do |checker_id|
-        next if CheckerRegistry.known?(checker_id)
-
-        issues << issue(
-          "error",
-          "unknown_checker_id",
-          "Workflow references a checker that is not in the canonical checker registry",
-          step: step_id,
-          checker_id: checker_id
-        )
+        unless CheckerRegistry.known?(checker_id)
+          issues << issue(
+            "error",
+            "unknown_checker_id",
+            "Workflow references a checker that is not in the canonical checker registry",
+            step: step_id,
+            checker_id: checker_id
+          )
+          next
+        end
+
+        unless CheckerRegistry.executable?(checker_id)
+          issues << issue(
+            "error",
+            "checker_execution_adapter_missing",
+            "Workflow references a checker without a Core execution adapter",
+            step: step_id,
+            checker_id: checker_id
+          )
+        end
       end
     end
   end
@@ -391,7 +438,7 @@ unless %w[prepared claimed consumed superseded failed].include?(status)
     end
   end
 
-  def validate_waiting_checkpoint(issues, root, state, step_by_id, artifact_by_path, strict_artifact_status)
+  def validate_waiting_checkpoint(issues, root, state, workflow, step_by_id, artifact_by_path, strict_artifact_status)
     return unless state["status"] == "waiting_checkpoint"
 
     step = step_by_id[state["current_step"]]
@@ -405,6 +452,11 @@ unless %w[prepared claimed consumed superseded failed].include?(status)
     end
 
     validate_step_outputs(issues, root, state, step, artifact_by_path, strict_artifact_status, true)
+    checkpoint_id = state["checkpoint"] || step["checkpoint"]
+    checkpoint = workflow.fetch("checkpoints", {})[checkpoint_id]
+    CheckpointMeta.validate(root: root, state: state, checkpoint: checkpoint).each do |entry|
+      issues << issue("error", entry.fetch("code"), entry.fetch("message"), step: step["id"], checkpoint: checkpoint_id, **entry.reject { |key, _| %w[code message].include?(key) })
+    end
   end
 
   # An in-progress step may have written its output before its complete-step

package/plugins/devflow-core/skills/workflow-orchestrator/scripts/workflow_gate_contract.rb

@@ -73,7 +73,11 @@ module WorkflowGateContract
       step_checkers.each do |step_id, checker_ids|
         errors << error("CHECKER_STEP_REF_NOT_FOUND", "checker declaration references an unknown step", step: step_id) unless step_by_id.key?(step_id.to_s)
         Array(checker_ids).each do |checker_id|
-          errors << error("UNKNOWN_CHECKER_ID", "workflow references an unknown checker", step: step_id, checker: checker_id) unless CheckerRegistry.known?(checker_id)
+          if !CheckerRegistry.known?(checker_id)
+            errors << error("UNKNOWN_CHECKER_ID", "workflow references an unknown checker", step: step_id, checker: checker_id)
+          elsif !CheckerRegistry.executable?(checker_id)
+            errors << error("CHECKER_EXECUTION_ADAPTER_MISSING", "workflow references a checker without a Core execution adapter", step: step_id, checker: checker_id)
+          end
         end
       end
     end