@kontourai/flow-agents 0.1.1

package/evals/integration/test_hook_influence_cases.sh

@@ -0,0 +1,69 @@
+#!/usr/bin/env bash
+# test_hook_influence_cases.sh - behavioral hook-influence fixture contracts
+set -uo pipefail
+
+ROOT="$(cd "$(dirname "${BASH_SOURCE[0]}")/../.." && pwd)"
+source "$ROOT/evals/lib/node.sh"
+
+errors=0
+
+_pass() { echo "  ✓ $1"; }
+_fail() { echo "  ✗ $1"; errors=$((errors + 1)); }
+
+CASES="$ROOT/evals/fixtures/hook-influence/cases.json"
+VALIDATOR="$ROOT/scripts/validate-hook-influence-cases.js"
+
+if [[ -f "$CASES" ]]; then
+  _pass "hook influence case fixture exists"
+else
+  _fail "hook influence case fixture missing"
+fi
+
+if [[ -f "$VALIDATOR" ]]; then
+  _pass "hook influence case validator exists"
+else
+  _fail "hook influence case validator missing"
+fi
+
+if flow_agents_node "$VALIDATOR" "$CASES" > /tmp/flow-agents-hook-influence.out 2> /tmp/flow-agents-hook-influence.err; then
+  _pass "hook influence behavioral cases validate"
+else
+  _fail "hook influence behavioral cases failed validation: $(cat /tmp/flow-agents-hook-influence.out /tmp/flow-agents-hook-influence.err)"
+fi
+
+if rg -q '"tier": "adapter"' "$CASES" \
+  && rg -q '"tier": "live-acceptance"' "$CASES" \
+  && rg -q '"tier": "installed-command"' "$CASES" \
+  && rg -q '"tier": "documented-runtime-gap"' "$CASES" \
+  && rg -q '"tier": "design-target"' "$CASES"; then
+  _pass "hook influence cases distinguish adapter, live, installed-command, documented runtime-gap, and design-target evidence"
+else
+  _fail "hook influence cases do not distinguish required evidence tiers"
+fi
+
+if rg -q 'FLOW_AGENTS_ACCEPTANCE_CLAUDE_LLM=1' "$CASES" \
+  && rg -q '"runtime_scope": \["kiro-cli"\]' "$CASES" \
+  && rg -q 'codex exec exposes project hook guidance' "$CASES"; then
+  _pass "hook influence cases cover Claude, Kiro, and Codex runtime boundaries"
+else
+  _fail "hook influence cases miss a runtime boundary"
+fi
+
+if rg -q 'dev-builder-build-requires-pickup-probe-before-plan' "$CASES" \
+  && rg -q 'dev-builder-review-before-verify-after-execute' "$CASES" \
+  && rg -q 'dev-verify-fail-preserves-trace-before-rework' "$CASES" \
+  && rg -q 'codex-claude-strict-stop-adapter-contract' "$CASES"; then
+  _pass "hook influence cases cover #62 Builder Kit loop categories"
+else
+  _fail "hook influence cases miss a required #62 Builder Kit loop category"
+fi
+
+rm -f /tmp/flow-agents-hook-influence.out /tmp/flow-agents-hook-influence.err
+
+if [[ "$errors" -eq 0 ]]; then
+  echo "Hook influence case integration passed."
+  exit 0
+fi
+
+echo "Hook influence case integration failed: $errors issue(s)."
+exit 1

package/evals/integration/test_local_flow_kit_install.sh

@@ -0,0 +1,145 @@
+#!/usr/bin/env bash
+# test_local_flow_kit_install.sh — Exercise local Flow Kit install/list/status behavior.
+set -uo pipefail
+
+ROOT="$(cd "$(dirname "${BASH_SOURCE[0]}")/../.." && pwd)"
+source "$ROOT/evals/lib/node.sh"
+
+errors=0
+TMP_DIR="$(mktemp -d)"
+trap 'rm -rf "$TMP_DIR"' EXIT
+
+pass() { echo "  ✓ $1"; }
+fail() { echo "  ✗ $1"; errors=$((errors + 1)); }
+
+CLI="$ROOT/scripts/flow-kit.js"
+VALID_SRC="$ROOT/evals/fixtures/flow-kit-repository/valid-local-kit"
+INVALID_SRC="$ROOT/evals/fixtures/flow-kit-repository/invalid-missing-flow"
+DEST="$TMP_DIR/install-dest"
+REGISTRY="$DEST/kits/local/installed-kits.json"
+CATALOG_HASH_BEFORE="$(shasum -a 256 "$ROOT/kits/catalog.json" | awk '{print $1}')"
+mkdir -p "$DEST"
+
+echo "=== Local Flow Kit Install Checks ==="
+
+install_output="$TMP_DIR/install.out"
+if flow_agents_node "$CLI" install-local "$VALID_SRC" --dest "$DEST" >"$install_output" 2>&1; then
+  pass "valid local kit installs into temp destination"
+else
+  fail "valid local kit install failed"
+  sed -n '1,160p' "$install_output"
+fi
+
+if node - "$REGISTRY" <<'NODE'
+const fs = require("node:fs");
+const path = require("node:path");
+const registry = JSON.parse(fs.readFileSync(process.argv[2], "utf8"));
+const entry = registry.kits[0];
+for (const key of ["id", "source", "hash", "installed_at", "installed_path", "state"]) {
+  if (!(key in entry)) throw new Error(`missing metadata key: ${key}`);
+}
+if (entry.id !== "example-kit") throw new Error(`unexpected id: ${entry.id}`);
+if (!entry.hash.startsWith("sha256:")) throw new Error("hash should include sha256 prefix");
+if (Number.isNaN(Date.parse(entry.installed_at))) throw new Error("installed_at should be ISO parseable");
+if (!fs.existsSync(path.join(entry.installed_path, "kit.json"))) throw new Error("installed kit copy missing kit.json");
+console.log("ok");
+NODE
+then
+  pass "install records required provenance metadata"
+else
+  fail "install metadata is incomplete"
+fi
+
+registry_hash_before_invalid="$(shasum -a 256 "$REGISTRY" | awk '{print $1}')"
+invalid_output="$TMP_DIR/invalid.out"
+if flow_agents_node "$CLI" install-local "$INVALID_SRC" --dest "$DEST" >"$invalid_output" 2>&1; then
+  fail "invalid local kit install should fail"
+  sed -n '1,160p' "$invalid_output"
+elif rg -q 'Flow Kit repository validation failed' "$invalid_output" \
+  && [[ "$registry_hash_before_invalid" == "$(shasum -a 256 "$REGISTRY" | awk '{print $1}')" ]]; then
+  pass "invalid kit fails validation before registry mutation"
+else
+  fail "invalid kit failure did not preserve registry or diagnostic"
+  sed -n '1,160p' "$invalid_output"
+fi
+
+registry_hash_before_idempotent="$(shasum -a 256 "$REGISTRY" | awk '{print $1}')"
+idempotent_output="$TMP_DIR/idempotent.out"
+if flow_agents_node "$CLI" install-local "$VALID_SRC" --dest "$DEST" >"$idempotent_output" 2>&1 \
+  && rg -q "already installed" "$idempotent_output" \
+  && [[ "$registry_hash_before_idempotent" == "$(shasum -a 256 "$REGISTRY" | awk '{print $1}')" ]]; then
+  pass "same-source reinstall is idempotent"
+else
+  fail "same-source reinstall was not idempotent"
+  sed -n '1,160p' "$idempotent_output"
+fi
+
+CONFLICT_SRC="$TMP_DIR/conflict-source"
+cp -R "$VALID_SRC" "$CONFLICT_SRC"
+printf '\nconflict copy\n' >> "$CONFLICT_SRC/docs/README.md"
+conflict_output="$TMP_DIR/conflict.out"
+registry_hash_before_conflict="$(shasum -a 256 "$REGISTRY" | awk '{print $1}')"
+if flow_agents_node "$CLI" install-local "$CONFLICT_SRC" --dest "$DEST" >"$conflict_output" 2>&1; then
+  fail "different source with existing kit id should conflict"
+  sed -n '1,160p' "$conflict_output"
+elif rg -q 'conflict: kit' "$conflict_output" \
+  && [[ "$registry_hash_before_conflict" == "$(shasum -a 256 "$REGISTRY" | awk '{print $1}')" ]]; then
+  pass "different-source duplicate id conflicts without mutation"
+else
+  fail "duplicate id conflict did not preserve registry or diagnostic"
+  sed -n '1,160p' "$conflict_output"
+fi
+
+force_conflict_output="$TMP_DIR/force-conflict.out"
+if flow_agents_node "$CLI" install-local "$CONFLICT_SRC" --dest "$DEST" --force >"$force_conflict_output" 2>&1; then
+  fail "--force should not replace a different-source duplicate id"
+  sed -n '1,160p' "$force_conflict_output"
+elif rg -q 'conflict: kit' "$force_conflict_output" \
+  && [[ "$registry_hash_before_conflict" == "$(shasum -a 256 "$REGISTRY" | awk '{print $1}')" ]]; then
+  pass "--force preserves different-source duplicate id conflict"
+else
+  fail "--force duplicate id conflict did not preserve registry or diagnostic"
+  sed -n '1,160p' "$force_conflict_output"
+fi
+
+update_output="$TMP_DIR/update.out"
+if flow_agents_node "$CLI" install-local "$CONFLICT_SRC" --dest "$DEST" --update >"$update_output" 2>&1 \
+  && rg -q "updated local kit" "$update_output" \
+  && rg -q "$CONFLICT_SRC" "$REGISTRY"; then
+  pass "explicit update replaces duplicate id source"
+else
+  fail "explicit update did not replace duplicate id source"
+  sed -n '1,160p' "$update_output"
+fi
+
+list_output="$TMP_DIR/list.out"
+status_output="$TMP_DIR/status.out"
+if flow_agents_node "$CLI" list --dest "$DEST" >"$list_output" 2>&1 \
+  && flow_agents_node "$CLI" status --dest "$DEST" example-kit >"$status_output" 2>&1 \
+  && rg -q 'example-kit' "$list_output" \
+  && rg -q 'source=' "$list_output" \
+  && rg -q 'installed_at=' "$list_output" \
+  && rg -q '"id": "example-kit"' "$status_output" \
+  && rg -q '"state": "installed"' "$status_output"; then
+  pass "list and status expose installed kit provenance without mutation"
+else
+  fail "list/status output missing expected provenance"
+  sed -n '1,160p' "$list_output"
+  sed -n '1,160p' "$status_output"
+fi
+
+CATALOG_HASH_AFTER="$(shasum -a 256 "$ROOT/kits/catalog.json" | awk '{print $1}')"
+if [[ "$CATALOG_HASH_BEFORE" == "$CATALOG_HASH_AFTER" ]]; then
+  pass "local installs do not mutate source kits/catalog.json"
+else
+  fail "source kits/catalog.json changed during local install test"
+fi
+
+echo ""
+if [[ "$errors" -eq 0 ]]; then
+  echo "Local Flow Kit install checks passed."
+  exit 0
+fi
+
+echo "Local Flow Kit install checks failed: $errors issue(s)."
+exit 1

package/evals/integration/test_publish_change_helper.sh

@@ -0,0 +1,176 @@
+#!/usr/bin/env bash
+set -uo pipefail
+
+ROOT="$(cd "$(dirname "${BASH_SOURCE[0]}")/../.." && pwd)"
+SCRIPT="$ROOT/scripts/publish-change-helper.js"
+TMPDIR_EVAL="$(mktemp -d)"
+trap 'rm -rf "$TMPDIR_EVAL"' EXIT
+
+errors=0
+pass() { echo "  ✓ $1"; }
+fail() { echo "  ✗ $1"; errors=$((errors + 1)); }
+
+json_query() {
+  node -e 'const fs=require("fs"); let cur=JSON.parse(fs.readFileSync(process.argv[1],"utf8")); for (const part of process.argv[2].split(".")) cur=Array.isArray(cur) ? cur[Number(part)] : cur[part]; console.log(cur);' "$1" "$2"
+}
+
+echo "=== Publish Change Helper ==="
+
+BODY="$TMPDIR_EVAL/body.md"
+cat > "$BODY" <<'MD'
+Summary line.
+
+- first item
+- second item
+
+```sh
+printf 'hello\n'
+```
+
+Closes #60
+MD
+
+cat > "$TMPDIR_EVAL/render-input.json" <<JSON
+{
+  "provider": "github",
+  "change_provider": {"role": "ChangeProvider", "kind": "github"},
+  "title": "Workflow hardening",
+  "body_file": "$BODY",
+  "expected_closing_refs": ["kontourai/flow-agents#60"]
+}
+JSON
+
+node "$SCRIPT" render \
+  --input-json "$TMPDIR_EVAL/render-input.json" \
+  --body-out "$TMPDIR_EVAL/rendered-body.md" \
+  > "$TMPDIR_EVAL/rendered.json"
+status=$?
+[[ "$status" -eq 0 ]] && pass "renders change request successfully" || fail "renders change request successfully"
+
+cmp -s "$BODY" "$TMPDIR_EVAL/rendered-body.md" && pass "body file rendering preserves multiline markdown exactly" || fail "body file rendering preserves multiline markdown exactly"
+node -e 'const fs=require("fs"); const expected=fs.readFileSync(process.argv[1],"utf8"); const actual=JSON.parse(fs.readFileSync(process.argv[2],"utf8")).body; process.exit(actual === expected ? 0 : 1);' "$BODY" "$TMPDIR_EVAL/rendered.json"
+[[ "$?" -eq 0 ]] && pass "rendered JSON body preserves real multiline text" || fail "rendered JSON body preserves real multiline text"
+[[ "$(json_query "$TMPDIR_EVAL/rendered.json" "change_provider.role")" == "ChangeProvider" ]] && pass "rendered request keeps provider-neutral ChangeProvider role" || fail "rendered request keeps provider-neutral ChangeProvider role"
+
+cat > "$TMPDIR_EVAL/closing-pass.json" <<'JSON'
+{
+  "provider": "github",
+  "default_owner": "kontourai",
+  "default_repo": "flow-agents",
+  "expected_closing_refs": ["#60"],
+  "provider_output": {
+    "recognized_closing_refs": ["kontourai/flow-agents#60"]
+  }
+}
+JSON
+
+node "$SCRIPT" validate-closing-refs \
+  --input-json "$TMPDIR_EVAL/closing-pass.json" \
+  > "$TMPDIR_EVAL/closing-pass.out"
+status=$?
+[[ "$status" -eq 0 ]] && pass "recognized closing refs pass validation" || fail "recognized closing refs pass validation"
+[[ "$(json_query "$TMPDIR_EVAL/closing-pass.out" "status")" == "pass" ]] && pass "closing ref pass result is explicit" || fail "closing ref pass result is explicit"
+
+cat > "$TMPDIR_EVAL/closing-missing.json" <<'JSON'
+{
+  "provider": "github",
+  "default_owner": "kontourai",
+  "default_repo": "flow-agents",
+  "expected_closing_refs": ["#60"],
+  "provider_output": {
+    "recognized_closing_refs": []
+  }
+}
+JSON
+
+if node "$SCRIPT" validate-closing-refs \
+  --input-json "$TMPDIR_EVAL/closing-missing.json" \
+  > "$TMPDIR_EVAL/closing-missing.out" \
+  2> "$TMPDIR_EVAL/closing-missing.err"; then
+  fail "missing recognized closing refs should fail"
+elif [[ "$(json_query "$TMPDIR_EVAL/closing-missing.out" "status")" == "fail" ]] && rg -q "missing recognized closing refs" "$TMPDIR_EVAL/closing-missing.err"; then
+  pass "missing closing refs fail with actionable output"
+else
+  fail "missing closing refs failure was not actionable"
+fi
+
+cat > "$TMPDIR_EVAL/docs-files.json" <<'JSON'
+{"files": ["docs/workflow-usage-guide.md", "docs/work-item-adapters.md"]}
+JSON
+cat > "$TMPDIR_EVAL/empty-checks.json" <<'JSON'
+[]
+JSON
+
+node "$SCRIPT" evaluate-provider-checks \
+  --change-files-json "$TMPDIR_EVAL/docs-files.json" \
+  --provider-checks-json "$TMPDIR_EVAL/empty-checks.json" \
+  > "$TMPDIR_EVAL/docs-checks.out"
+status=$?
+[[ "$status" -eq 0 ]] && pass "docs-only missing provider checks are accepted as skip" || fail "docs-only missing provider checks are accepted as skip"
+[[ "$(json_query "$TMPDIR_EVAL/docs-checks.out" "evidence_status")" == "skip" ]] && pass "docs-only missing checks map to evidence skip" || fail "docs-only missing checks map to evidence skip"
+[[ "$(json_query "$TMPDIR_EVAL/docs-checks.out" "release_gate_status")" == "not_required" ]] && pass "docs-only missing checks map to release not_required" || fail "docs-only missing checks map to release not_required"
+
+for risk in runtime schema package hook security; do
+  case "$risk" in
+    runtime) path="scripts/flow-kit.js" ;;
+    schema) path="schemas/workflow-evidence.schema.json" ;;
+    package) path="package.json" ;;
+    hook) path="scripts/hooks/quality-gate.js" ;;
+    security) path="security/policy.md" ;;
+  esac
+  printf '{"files":["%s"]}\n' "$path" > "$TMPDIR_EVAL/$risk-files.json"
+  if node "$SCRIPT" evaluate-provider-checks \
+    --change-files-json "$TMPDIR_EVAL/$risk-files.json" \
+    --provider-checks-json "$TMPDIR_EVAL/empty-checks.json" \
+    > "$TMPDIR_EVAL/$risk-checks.out" \
+    2> "$TMPDIR_EVAL/$risk-checks.err"; then
+    fail "$risk missing provider checks should not pass"
+  elif [[ "$(json_query "$TMPDIR_EVAL/$risk-checks.out" "evidence_status")" == "not_verified" ]] \
+    && [[ "$(json_query "$TMPDIR_EVAL/$risk-checks.out" "release_gate_status")" == "hold" ]]; then
+    pass "$risk missing provider checks map to not_verified and hold"
+  else
+    fail "$risk missing provider checks did not map to not_verified and hold"
+  fi
+done
+
+mkdir -p "$TMPDIR_EVAL/final-state"
+cat > "$TMPDIR_EVAL/final-state/evidence.json" <<'JSON'
+{
+  "schema_version": "1.0",
+  "task_slug": "final-state",
+  "updated_at": "2026-05-29T00:00:00Z",
+  "verdict": "pass",
+  "summary": "Final orchestrated evidence passed.",
+  "checks": [
+    {"id": "focused-tests", "kind": "test", "status": "pass", "summary": "Focused tests passed."}
+  ]
+}
+JSON
+cat > "$TMPDIR_EVAL/final-state/release.json" <<'JSON'
+{
+  "schema_version": "1.0",
+  "task_slug": "final-state",
+  "updated_at": "2026-05-29T00:00:00Z",
+  "decision": "merge",
+  "scope": "Final state reconciliation fixture.",
+  "summary": "Final release state passed.",
+  "evidence_ref": "evidence.json",
+  "gates": [
+    {"id": "merge", "kind": "merge", "status": "pass", "required": true, "summary": "Merge gate passed.", "evidence_refs": ["evidence.json"]}
+  ],
+  "rollback_plan": {"status": "not_required", "summary": "Merge-only fixture."}
+}
+JSON
+
+node "$SCRIPT" reconcile-final-state "$TMPDIR_EVAL/final-state" > "$TMPDIR_EVAL/reconcile.out"
+status=$?
+[[ "$status" -eq 0 ]] && pass "final reconciled evidence and release sidecars pass" || fail "final reconciled evidence and release sidecars pass"
+[[ "$(json_query "$TMPDIR_EVAL/reconcile.out" "status")" == "pass" ]] && pass "final reconciliation result is explicit" || fail "final reconciliation result is explicit"
+
+if [[ "$errors" -eq 0 ]]; then
+  echo "Publish change helper checks passed"
+else
+  echo "Publish change helper checks failed: $errors"
+fi
+
+exit "$errors"

package/evals/integration/test_pull_work_provider.sh

@@ -0,0 +1,140 @@
+#!/usr/bin/env bash
+set -uo pipefail
+
+ROOT="$(cd "$(dirname "${BASH_SOURCE[0]}")/../.." && pwd)"
+SETTINGS="$ROOT/context/settings/backlog-provider-settings.json"
+FIXTURE="$ROOT/evals/fixtures/pull-work-provider/github-issues.json"
+SCRIPT="$ROOT/scripts/pull-work-provider.js"
+TMPDIR_EVAL="$(mktemp -d)"
+trap 'rm -rf "$TMPDIR_EVAL"' EXIT
+
+errors=0
+pass() { echo "  ✓ $1"; }
+fail() { echo "  ✗ $1"; errors=$((errors + 1)); }
+
+json_query() {
+  node -e 'const fs=require("fs"); let cur=JSON.parse(fs.readFileSync(process.argv[1],"utf8")); for (const part of process.argv[2].split(".")) cur=part==="length" ? cur.length : (Array.isArray(cur) ? cur[Number(part)] : cur[part]); console.log(cur);' "$1" "$2"
+}
+
+echo "=== Pull Work Provider Normalization ==="
+
+node "$SCRIPT" \
+  --settings-json "$SETTINGS" \
+  --issues-json "$FIXTURE" \
+  --resolved-ref 'kontourai/flow#2=closed' \
+  --current-ref main \
+  --current-sha cccccccccccccccccccccccccccccccccccccccc \
+  --changed-file docs/readme.md \
+  --commits-since dddddddddddddddddddddddddddddddddddddddd=2 \
+  --commits-since eeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeee=12 \
+  --now 2026-06-03T00:00:00Z \
+  > "$TMPDIR_EVAL/normalized.json"
+status=$?
+[[ "$status" -eq 0 ]] && pass "normalizer exits successfully" || fail "normalizer exits successfully"
+
+[[ "$(json_query "$TMPDIR_EVAL/normalized.json" "items.0.id")" == "github:kontourai/flow-agents#22" ]] && pass "normalizes provider-qualified id" || fail "normalizes provider-qualified id"
+[[ "$(json_query "$TMPDIR_EVAL/normalized.json" "items.0.source_provider.role")" == "WorkItemProvider" ]] && pass "preserves work item provider ref" || fail "preserves work item provider ref"
+[[ "$(json_query "$TMPDIR_EVAL/normalized.json" "items.0.board_membership.role")" == "BoardProvider" ]] && pass "preserves board provider ref" || fail "preserves board provider ref"
+[[ "$(json_query "$TMPDIR_EVAL/normalized.json" "items.0.artifact_refs.0")" == ".flow-agents/flow-agents-kit-platform-backlog/flow-agents-kit-platform-backlog--idea-to-backlog.md" ]] && pass "preserves source artifact ref" || fail "preserves source artifact ref"
+[[ "$(json_query "$TMPDIR_EVAL/normalized.json" "items.0.blockers.0.ref.owner")" == "kontourai" ]] && pass "preserves blocker owner" || fail "preserves blocker owner"
+[[ "$(json_query "$TMPDIR_EVAL/normalized.json" "items.0.blockers.0.ref.repo")" == "flow" ]] && pass "preserves cross-repo blocker repo" || fail "preserves cross-repo blocker repo"
+[[ "$(json_query "$TMPDIR_EVAL/normalized.json" "items.0.dependency_impacts.0.ref.owner")" == "kontourai" ]] && pass "projects prose blocker impact owner" || fail "projects prose blocker impact owner"
+[[ "$(json_query "$TMPDIR_EVAL/normalized.json" "items.0.dependency_impacts.0.ref.repo")" == "flow" ]] && pass "projects prose blocker impact repo" || fail "projects prose blocker impact repo"
+[[ "$(json_query "$TMPDIR_EVAL/normalized.json" "items.0.dependency_impacts.0.cross_repo")" == "true" ]] && pass "marks prose blocker impact cross-repo" || fail "marks prose blocker impact cross-repo"
+[[ "$(json_query "$TMPDIR_EVAL/normalized.json" "items.0.dependency_impacts.0.known_status")" == "closed" ]] && pass "projects resolved prose blocker status" || fail "projects resolved prose blocker status"
+[[ "$(json_query "$TMPDIR_EVAL/normalized.json" "items.0.dependency_impacts.0.impact_state")" == "resolved" ]] && pass "projects resolved prose blocker impact state" || fail "projects resolved prose blocker impact state"
+[[ "$(json_query "$TMPDIR_EVAL/normalized.json" "items.0.readiness.classification")" == "ready" ]] && pass "#22 ready when flow#2 is closed" || fail "#22 ready when flow#2 is closed"
+[[ "$(json_query "$TMPDIR_EVAL/normalized.json" "items.1.readiness.classification")" == "blocked" ]] && pass "blocked item classified blocked" || fail "blocked item classified blocked"
+[[ "$(json_query "$TMPDIR_EVAL/normalized.json" "items.2.readiness.classification")" == "related-only" ]] && pass "research item classified related-only" || fail "research item classified related-only"
+[[ "$(json_query "$TMPDIR_EVAL/normalized.json" "items.3.readiness.classification")" == "in_progress" ]] && pass "in-progress item classified in_progress" || fail "in-progress item classified in_progress"
+[[ "$(json_query "$TMPDIR_EVAL/normalized.json" "items.4.readiness.classification")" == "stale" ]] && pass "old unshaped item classified stale" || fail "old unshaped item classified stale"
+[[ "$(json_query "$TMPDIR_EVAL/normalized.json" "items.5.planned_base_ref")" == "main" ]] && pass "marker normalizes planned_base_ref" || fail "marker normalizes planned_base_ref"
+[[ "$(json_query "$TMPDIR_EVAL/normalized.json" "items.5.planned_base_sha")" == "aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa" ]] && pass "marker normalizes planned_base_sha" || fail "marker normalizes planned_base_sha"
+[[ "$(json_query "$TMPDIR_EVAL/normalized.json" "items.5.planned_at")" == "2026-06-03T03:23:14Z" ]] && pass "marker normalizes planned_at" || fail "marker normalizes planned_at"
+[[ "$(json_query "$TMPDIR_EVAL/normalized.json" "items.5.planning_artifact_ref")" == ".flow-agents/idea-to-backlog-source-revision-structured-blockers/idea-to-backlog-source-revision-structured-blockers--plan.md" ]] && pass "marker normalizes planning_artifact_ref" || fail "marker normalizes planning_artifact_ref"
+[[ "$(json_query "$TMPDIR_EVAL/normalized.json" "items.5.planning_scope_refs.0")" == "skills/idea-to-backlog/SKILL.md" ]] && pass "marker normalizes planning_scope_refs" || fail "marker normalizes planning_scope_refs"
+[[ "$(json_query "$TMPDIR_EVAL/normalized.json" "items.5.source_revisions.1.repo")" == "kontourai/flow" ]] && pass "marker preserves repo-scoped source_revisions" || fail "marker preserves repo-scoped source_revisions"
+[[ "$(json_query "$TMPDIR_EVAL/normalized.json" "items.5.blockers.0.ref.owner")" == "kontourai" ]] && pass "structured blocker preserves provider owner" || fail "structured blocker preserves provider owner"
+[[ "$(json_query "$TMPDIR_EVAL/normalized.json" "items.5.blockers.0.ref.repo")" == "flow" ]] && pass "structured blocker preserves provider repo" || fail "structured blocker preserves provider repo"
+[[ "$(json_query "$TMPDIR_EVAL/normalized.json" "items.5.blockers.0.ref.number")" == "2" ]] && pass "structured blocker preserves provider number" || fail "structured blocker preserves provider number"
+[[ "$(json_query "$TMPDIR_EVAL/normalized.json" "items.5.blockers.1.type")" == "text" ]] && pass "structured blocker preserves text blocker" || fail "structured blocker preserves text blocker"
+[[ "$(json_query "$TMPDIR_EVAL/normalized.json" "items.5.blockers.1.evidence")" == "Product decision on rollout scope." ]] && pass "structured blocker preserves text evidence" || fail "structured blocker preserves text evidence"
+[[ "$(json_query "$TMPDIR_EVAL/normalized.json" "items.5.blockers.2.evidence")" == "Blocked by product decision on rollout scope." ]] && pass "prose blocker remains alongside structured blockers" || fail "prose blocker remains alongside structured blockers"
+[[ "$(json_query "$TMPDIR_EVAL/normalized.json" "items.5.readiness.classification")" == "blocked" ]] && pass "text blocker keeps marker item blocked" || fail "text blocker keeps marker item blocked"
+[[ "$(json_query "$TMPDIR_EVAL/normalized.json" "items.6.blockers.0.ref.repo")" == "flow-agents" ]] && pass "invalid marker falls back to prose blocker repo" || fail "invalid marker falls back to prose blocker repo"
+[[ "$(json_query "$TMPDIR_EVAL/normalized.json" "items.6.readiness.classification")" == "blocked" ]] && pass "invalid marker fallback preserves blocked readiness" || fail "invalid marker fallback preserves blocked readiness"
+[[ "$(json_query "$TMPDIR_EVAL/normalized.json" "items.7.revision_freshness.classification")" == "fresh" ]] && pass "fresh item freshness classified fresh" || fail "fresh item freshness classified fresh"
+[[ "$(json_query "$TMPDIR_EVAL/normalized.json" "items.7.revision_freshness.planned_base_sha")" == "cccccccccccccccccccccccccccccccccccccccc" ]] && pass "freshness preserves planned base sha" || fail "freshness preserves planned base sha"
+[[ "$(json_query "$TMPDIR_EVAL/normalized.json" "items.7.revision_freshness.current_sha")" == "cccccccccccccccccccccccccccccccccccccccc" ]] && pass "freshness reports current sha" || fail "freshness reports current sha"
+[[ "$(json_query "$TMPDIR_EVAL/normalized.json" "items.7.revision_freshness.planned_age_days")" == "1" ]] && pass "freshness reports planned age" || fail "freshness reports planned age"
+[[ "$(json_query "$TMPDIR_EVAL/normalized.json" "items.8.revision_freshness.classification")" == "drifted" ]] && pass "drifted item freshness classified drifted" || fail "drifted item freshness classified drifted"
+[[ "$(json_query "$TMPDIR_EVAL/normalized.json" "items.8.revision_freshness.commits_since_planned_base")" == "2" ]] && pass "drifted item reports commits since planned base" || fail "drifted item reports commits since planned base"
+[[ "$(json_query "$TMPDIR_EVAL/normalized.json" "items.8.revision_freshness.planning_scope_intersections.length")" == "0" ]] && pass "drifted item reports no scope intersections" || fail "drifted item reports no scope intersections"
+[[ "$(json_query "$TMPDIR_EVAL/normalized.json" "items.9.revision_freshness.classification")" == "stale" ]] && pass "stale item freshness classified stale" || fail "stale item freshness classified stale"
+[[ "$(json_query "$TMPDIR_EVAL/normalized.json" "items.9.revision_freshness.route_recommendation.target")" == "idea-to-backlog" ]] && pass "stale item routes back to idea-to-backlog" || fail "stale item routes back to idea-to-backlog"
+[[ "$(json_query "$TMPDIR_EVAL/normalized.json" "items.9.readiness.classification")" == "stale" ]] && pass "stale freshness constrains readiness" || fail "stale freshness constrains readiness"
+[[ "$(json_query "$TMPDIR_EVAL/normalized.json" "items.10.revision_freshness.classification")" == "not_verified" ]] && pass "legacy item without planned_base_sha is not_verified" || fail "legacy item without planned_base_sha is not_verified"
+[[ "$(json_query "$TMPDIR_EVAL/normalized.json" "items.10.revision_freshness.reasons.0.code")" == "missing_planned_base_sha" ]] && pass "legacy item records missing planned_base_sha reason" || fail "legacy item records missing planned_base_sha reason"
+[[ "$(json_query "$TMPDIR_EVAL/normalized.json" "items.10.readiness.classification")" == "ready" ]] && pass "legacy freshness gap preserves readiness while reporting not_verified freshness" || fail "legacy freshness gap preserves readiness while reporting not_verified freshness"
+[[ "$(json_query "$TMPDIR_EVAL/normalized.json" "items.11.blockers.0.ref.owner")" == "kontourai" ]] && pass "structured-only blocker preserves owner" || fail "structured-only blocker preserves owner"
+[[ "$(json_query "$TMPDIR_EVAL/normalized.json" "items.11.blockers.0.ref.repo")" == "flow" ]] && pass "structured-only blocker preserves cross-repo repo" || fail "structured-only blocker preserves cross-repo repo"
+[[ "$(json_query "$TMPDIR_EVAL/normalized.json" "items.11.blockers.0.ref.number")" == "2" ]] && pass "structured-only blocker preserves number" || fail "structured-only blocker preserves number"
+[[ "$(json_query "$TMPDIR_EVAL/normalized.json" "items.11.dependency_impacts.0.ref.owner")" == "kontourai" ]] && pass "structured-only dependency impact preserves owner" || fail "structured-only dependency impact preserves owner"
+[[ "$(json_query "$TMPDIR_EVAL/normalized.json" "items.11.dependency_impacts.0.ref.repo")" == "flow" ]] && pass "structured-only dependency impact preserves repo" || fail "structured-only dependency impact preserves repo"
+[[ "$(json_query "$TMPDIR_EVAL/normalized.json" "items.11.dependency_impacts.0.ref.number")" == "2" ]] && pass "structured-only dependency impact preserves number" || fail "structured-only dependency impact preserves number"
+[[ "$(json_query "$TMPDIR_EVAL/normalized.json" "items.11.dependency_impacts.0.source.type")" == "provider_ref" ]] && pass "structured-only dependency impact records source blocker type" || fail "structured-only dependency impact records source blocker type"
+[[ "$(json_query "$TMPDIR_EVAL/normalized.json" "items.11.dependency_impacts.0.source.evidence")" == "Requires Flow contract issue first." ]] && pass "structured-only dependency impact records source evidence" || fail "structured-only dependency impact records source evidence"
+[[ "$(json_query "$TMPDIR_EVAL/normalized.json" "items.11.dependency_impacts.0.cross_repo")" == "true" ]] && pass "structured-only dependency impact is cross-repo" || fail "structured-only dependency impact is cross-repo"
+[[ "$(json_query "$TMPDIR_EVAL/normalized.json" "items.11.dependency_impacts.0.known_status")" == "closed" ]] && pass "structured-only dependency impact records resolved status" || fail "structured-only dependency impact records resolved status"
+[[ "$(json_query "$TMPDIR_EVAL/normalized.json" "items.11.dependency_impacts.0.impact_state")" == "resolved" ]] && pass "structured-only dependency impact state is resolved" || fail "structured-only dependency impact state is resolved"
+[[ "$(json_query "$TMPDIR_EVAL/normalized.json" "items.11.readiness.classification")" == "ready" ]] && pass "structured-only item ready when cross-repo blocker is closed" || fail "structured-only item ready when cross-repo blocker is closed"
+
+node "$SCRIPT" \
+  --settings-json "$SETTINGS" \
+  --issues-json "$FIXTURE" \
+  --current-ref main \
+  --current-sha ffffffffffffffffffffffffffffffffffffffff \
+  --now 2026-06-03T00:00:00Z \
+  > "$TMPDIR_EVAL/missing-drift-evidence.json"
+[[ "$(json_query "$TMPDIR_EVAL/missing-drift-evidence.json" "items.8.revision_freshness.classification")" == "not_verified" ]] && pass "moved target without drift evidence is not_verified" || fail "moved target without drift evidence is not_verified"
+[[ "$(json_query "$TMPDIR_EVAL/missing-drift-evidence.json" "items.8.revision_freshness.reasons.1.code")" == "missing_drift_evidence" ]] && pass "moved target records missing drift evidence reason" || fail "moved target records missing drift evidence reason"
+[[ "$(json_query "$TMPDIR_EVAL/missing-drift-evidence.json" "items.8.readiness.classification")" == "blocked" ]] && pass "missing drift evidence blocks ready classification" || fail "missing drift evidence blocks ready classification"
+
+node "$SCRIPT" \
+  --settings-json "$SETTINGS" \
+  --issues-json "$FIXTURE" \
+  --current-ref main \
+  --current-sha cccccccccccccccccccccccccccccccccccccccc \
+  --changed-file context/contracts/work-item-contract.md \
+  --commits-since dddddddddddddddddddddddddddddddddddddddd=2 \
+  --now 2026-06-03T00:00:00Z \
+  > "$TMPDIR_EVAL/material-contract-drift.json"
+[[ "$(json_query "$TMPDIR_EVAL/material-contract-drift.json" "items.8.revision_freshness.classification")" == "stale" ]] && pass "contract drift marks non-scoped planned item stale" || fail "contract drift marks non-scoped planned item stale"
+[[ "$(json_query "$TMPDIR_EVAL/material-contract-drift.json" "items.8.revision_freshness.reasons.2.code")" == "material_freshness_files_changed" ]] && pass "contract drift records material freshness reason" || fail "contract drift records material freshness reason"
+[[ "$(json_query "$TMPDIR_EVAL/material-contract-drift.json" "items.9.revision_freshness.planning_scope_intersections.0")" == "context/contracts/work-item-contract.md" ]] && pass "stale item reports planning scope intersection" || fail "stale item reports planning scope intersection"
+
+node "$SCRIPT" \
+  --settings-json - \
+  --issues-json "$FIXTURE" \
+  --resolved-ref 'kontourai/flow#2=closed' \
+  < "$SETTINGS" \
+  > "$TMPDIR_EVAL/stdin-settings.json"
+[[ "$(json_query "$TMPDIR_EVAL/stdin-settings.json" "items.0.readiness.classification")" == "ready" ]] && pass "normalizer accepts stdin settings JSON" || fail "normalizer accepts stdin settings JSON"
+
+node "$SCRIPT" \
+  --settings-json "$SETTINGS" \
+  --issues-json "$FIXTURE" \
+  > "$TMPDIR_EVAL/unresolved.json"
+[[ "$(json_query "$TMPDIR_EVAL/unresolved.json" "items.0.readiness.classification")" == "blocked" ]] && pass "#22 blocked when flow#2 state is unknown" || fail "#22 blocked when flow#2 state is unknown"
+[[ "$(json_query "$TMPDIR_EVAL/unresolved.json" "items.5.readiness.classification")" == "blocked" ]] && pass "marker item blocked when provider ref and text blocker are unresolved" || fail "marker item blocked when provider ref and text blocker are unresolved"
+[[ "$(json_query "$TMPDIR_EVAL/unresolved.json" "items.11.dependency_impacts.0.known_status")" == "unknown" ]] && pass "structured-only dependency impact records unknown status without resolved ref" || fail "structured-only dependency impact records unknown status without resolved ref"
+[[ "$(json_query "$TMPDIR_EVAL/unresolved.json" "items.11.dependency_impacts.0.impact_state")" == "unknown" ]] && pass "structured-only dependency impact state is unknown without resolved ref" || fail "structured-only dependency impact state is unknown without resolved ref"
+[[ "$(json_query "$TMPDIR_EVAL/unresolved.json" "items.11.readiness.classification")" == "blocked" ]] && pass "structured-only blocker blocks without prose parsing" || fail "structured-only blocker blocks without prose parsing"
+
+if [[ "$errors" -eq 0 ]]; then
+  echo "Pull work provider checks passed"
+else
+  echo "Pull work provider checks failed: $errors"
+fi
+
+exit "$errors"