@kontourai/flow-agents 0.1.1

package/evals/integration/test_console_learning_projection.sh

@@ -0,0 +1,192 @@
+#!/usr/bin/env bash
+# test_console_learning_projection.sh - Console learning projection contract and non-mutation guard
+set -uo pipefail
+
+ROOT="$(cd "$(dirname "${BASH_SOURCE[0]}")/../.." && pwd)"
+source "$ROOT/evals/lib/node.sh"
+
+FIXTURE_DIR="$ROOT/evals/fixtures/console-learning-projection"
+TMPDIR_EVAL="$(mktemp -d /private/tmp/eval-console-learning-projection.XXXXXX)"
+ARTIFACT_ROOT="$TMPDIR_EVAL/artifacts"
+KONTOUR_ROOT="$TMPDIR_EVAL/.kontour"
+GENERATED_AT="2026-06-06T20:00:00Z"
+PROJECTION="$KONTOUR_ROOT/projections/flow-agents-learning/repo-fixture-repo.json"
+pass=0
+fail=0
+
+cleanup() {
+  rm -rf "$TMPDIR_EVAL"
+}
+trap cleanup EXIT
+
+_pass() { echo "  ✓ $1"; pass=$((pass + 1)); }
+_fail() { echo "  ✗ $1"; fail=$((fail + 1)); }
+
+echo "=== Layer 2: Console Learning Projection ==="
+echo ""
+
+mkdir -p "$ARTIFACT_ROOT"
+cp -R "$FIXTURE_DIR/artifacts/." "$ARTIFACT_ROOT/"
+
+if flow_agents_build_ts 2>"$TMPDIR_EVAL/build.err"; then
+  _pass "TypeScript CLI build is available"
+else
+  _fail "TypeScript CLI build failed: $(cat "$TMPDIR_EVAL/build.err" 2>/dev/null)"
+fi
+
+find "$ARTIFACT_ROOT" -name learning.json -type f -print0 | sort -z | xargs -0 shasum -a 256 >"$TMPDIR_EVAL/before.sha"
+
+cp "$PROJECTION" "$TMPDIR_EVAL/projection-first.json" 2>/dev/null || true
+if node "$ROOT/build/src/cli.js" console-learning-projection \
+  --artifact-root "$ARTIFACT_ROOT" \
+  --kontour-root "$KONTOUR_ROOT" \
+  --scope fixture-repo \
+  --scope-kind repo \
+  --producer flow-agents-learning \
+  --generated-at "$GENERATED_AT" \
+  --json >"$TMPDIR_EVAL/run.json" 2>"$TMPDIR_EVAL/run.err"; then
+  _pass "CLI writes projection from fixture artifact root"
+else
+  _fail "CLI failed: $(cat "$TMPDIR_EVAL/run.err" 2>/dev/null)"
+fi
+
+if [[ -f "$PROJECTION" ]]; then
+  _pass "projection file exists at expected producer and scope path"
+else
+  _fail "projection file missing at $PROJECTION"
+fi
+
+if jq -e --arg path "$PROJECTION" \
+  '.scanned_learning_file_count == 2 and .emitted_learning_count == 2 and .destination == $path and .producer == "flow-agents-learning" and .scope == {"kind":"repo","id":"fixture-repo"} and .dry_run == false' \
+  "$TMPDIR_EVAL/run.json" >/dev/null 2>&1; then
+  _pass "JSON summary reports scanned files, emitted learnings, scope, producer, and destination"
+else
+  _fail "JSON summary missing expected command result"
+fi
+
+cp "$PROJECTION" "$TMPDIR_EVAL/projection-first.json" 2>/dev/null || true
+if node "$ROOT/build/src/cli.js" console-learning-projection \
+  --artifact-root "$ARTIFACT_ROOT" \
+  --kontour-root "$KONTOUR_ROOT" \
+  --scope fixture-repo \
+  --scope-kind repo \
+  --producer flow-agents-learning \
+  --generated-at "$GENERATED_AT" \
+  --json >"$TMPDIR_EVAL/run-second.json" 2>"$TMPDIR_EVAL/run-second.err"; then
+  if cmp -s "$TMPDIR_EVAL/projection-first.json" "$PROJECTION"; then
+    _pass "projection output is byte-stable with fixed generated-at"
+  else
+    _fail "projection output changed across fixed-timestamp runs"
+  fi
+else
+  _fail "second deterministic run failed: $(cat "$TMPDIR_EVAL/run-second.err" 2>/dev/null)"
+fi
+
+find "$ARTIFACT_ROOT" -name learning.json -type f -print0 | sort -z | xargs -0 shasum -a 256 >"$TMPDIR_EVAL/after.sha"
+if cmp -s "$TMPDIR_EVAL/before.sha" "$TMPDIR_EVAL/after.sha"; then
+  _pass "source learning.json files are byte-for-byte unchanged"
+else
+  _fail "source learning.json checksum changed after projection command"
+fi
+
+echo ""
+echo "--- Projection Contract ---"
+
+if jq -e --arg generated "$GENERATED_AT" '
+  .schema == "kontour.console.projection" and
+  .version == "0.1" and
+  .generatedAt == $generated and
+  .scope == {"kind":"repo","id":"fixture-repo"} and
+  .producer == {"id":"flow-agents-learning","product":"flow-agents"} and
+  .derivedFrom.mode == "direct_snapshot" and
+  .derivedFrom.eventHistory == "unavailable" and
+  .derivedFrom.directSnapshot.emittedAt == $generated and
+  .derivedFrom.directSnapshot.producer == {"id":"flow-agents-learning","product":"flow-agents"} and
+  .derivedFrom.directSnapshot.sourceRef == {"product":"flow-agents","kind":"workflow-learning","id":".flow-agents/*/learning.json","label":"Local workflow learning sidecars"} and
+  (.learnings | length) == 2
+' "$PROJECTION" >/dev/null 2>&1; then
+  _pass "projection envelope includes Console schema, scope, producer, and direct snapshot provenance"
+else
+  _fail "projection envelope is missing required Console contract fields"
+fi
+
+if jq -e '
+  all(.learnings[]; .family == "workflow" and .nonAuthority == true) and
+  ([paths(scalars) | map(tostring) | join(".") | select(test("(^|\\.)(claims|gates|decisions|actions)(\\.|$)"))] | length) == 0 and
+  (has("refs") | not) and
+  (has("links") | not)
+' "$PROJECTION" >/dev/null 2>&1; then
+  _pass "learnings are inert and output has no authoritative fields or invalid top-level refs/links"
+else
+  _fail "projection contains authority fields, invalid refs/links, or non-inert learnings"
+fi
+
+if jq -e '
+  .learnings as $learnings |
+  ($learnings | map(.id) | unique | length) == 2 and
+  all($learnings[]; (.id | test("^learning\\.workflow\\.")) and (.subjectRef.product == "flow-agents") and (.subjectRef.kind == "workflow") and (.sourceRef.product == "flow-agents") and (.sourceRef.kind == "workflow-learning") and (.summary | length > 0)) and
+  all($learnings[]; .sourceRef.id != .extensions["flow-agents"].record_id) and
+  any($learnings[]; .subjectRef.id == "console-learning-correction" and .sourceRef.id == "console-learning-correction/record-correction-needed" and .sourceRef.label == "console-learning-correction/record-correction-needed" and .extensions["flow-agents"].record_id == "record-correction-needed") and
+  any($learnings[]; .subjectRef.id == "console-learning-open-route" and .sourceRef.id == "console-learning-open-route/record-open-route" and .sourceRef.label == "console-learning-open-route/record-open-route" and .extensions["flow-agents"].record_id == "record-open-route")
+' "$PROJECTION" >/dev/null 2>&1; then
+  _pass "learning ids and subject/source refs include task context while preserving raw record ids"
+else
+  _fail "learning ids or subject/source refs do not correlate to source records"
+fi
+
+if jq -e '
+  .learnings[] | select(.sourceRef.id == "console-learning-correction/record-correction-needed") |
+  .extensions["flow-agents"] as $ext |
+  $ext.task_slug == "console-learning-correction" and
+  $ext.record_id == "record-correction-needed" and
+  $ext.source_path == "console-learning-correction/learning.json" and
+  $ext.source_refs == ["critique.json#/critiques/0/findings/0", "evals/integration/test_console_learning_projection.sh"] and
+  $ext.routing.count == 2 and
+  $ext.routing.open == 1 and
+  $ext.routing.deferred == 1 and
+  ($ext.routing.targets | sort) == ["eval", "skill"] and
+  ($ext.routing.statuses | sort) == ["deferred", "open"] and
+  ($ext.routing.refs | sort) == ["github:kontourai/flow-agents#96", "skills/learning-review/SKILL.md"] and
+  $ext.correction.needed == true and
+  $ext.correction.type == "workflow" and
+  $ext.correction.recurrence_key == "console-learning-projection.recurrence-metadata" and
+  $ext.correction.prevention == {"target":"eval","status":"open","ref":"evals/integration/test_console_learning_projection.sh"} and
+  $ext.outcome == "mixed" and
+  $ext.learning_status == "followup_required"
+' "$PROJECTION" >/dev/null 2>&1; then
+  _pass "correction-needed extension carries routing, correction, recurrence, and source path details"
+else
+  _fail "correction-needed extension missing required Flow Agents details"
+fi
+
+if jq -e '
+  .learnings[] | select(.sourceRef.id == "console-learning-open-route/record-open-route") |
+  .extensions["flow-agents"] as $ext |
+  $ext.task_slug == "console-learning-open-route" and
+  $ext.record_id == "record-open-route" and
+  $ext.source_path == "console-learning-open-route/learning.json" and
+  $ext.routing.count == 2 and
+  $ext.routing.open == 1 and
+  $ext.routing.accepted == 1 and
+  ($ext.routing.targets | sort) == ["backlog", "doc"] and
+  ($ext.routing.statuses | sort) == ["accepted", "open"] and
+  $ext.correction == {"needed":false} and
+  $ext.outcome == "success" and
+  $ext.learning_status == "learned"
+' "$PROJECTION" >/dev/null 2>&1; then
+  _pass "non-correction open-route extension carries routing and correction state"
+else
+  _fail "non-correction open-route extension missing required Flow Agents details"
+fi
+
+if git diff --quiet -- schemas/workflow-learning.schema.json; then
+  _pass "workflow-learning source schema is unchanged"
+else
+  _fail "schemas/workflow-learning.schema.json has unexpected diff"
+fi
+
+echo ""
+echo "Result: $pass passed, $fail failed"
+if [[ "$fail" -gt 0 ]]; then
+  exit 1
+fi

package/evals/integration/test_context_map.sh

@@ -0,0 +1,65 @@
+#!/usr/bin/env bash
+# test_context_map.sh — Generated context map drift and content checks
+set -uo pipefail
+
+ROOT="$(cd "$(dirname "${BASH_SOURCE[0]}")/../.." && pwd)"
+MAP="$ROOT/docs/context-map.md"
+TMPDIR_EVAL="$(mktemp -d)"
+trap 'rm -rf "$TMPDIR_EVAL"' EXIT
+
+errors=0
+_pass() { echo "  ✓ $1"; }
+_fail() { echo "  ✗ $1"; errors=$((errors + 1)); }
+
+if (cd "$ROOT" && npm run context-map:check) >"$TMPDIR_EVAL/check.out" 2>"$TMPDIR_EVAL/check.err"; then
+  _pass "context map is current"
+else
+  _fail "context map is stale: $(cat "$TMPDIR_EVAL/check.out" "$TMPDIR_EVAL/check.err")"
+fi
+
+if (cd "$ROOT" && npm run context-map -- --output "$TMPDIR_EVAL/context-map.md") >"$TMPDIR_EVAL/generate.out" 2>"$TMPDIR_EVAL/generate.err" && cmp -s "$MAP" "$TMPDIR_EVAL/context-map.md"; then
+  _pass "context map generation is deterministic"
+else
+  _fail "context map generation is not deterministic"
+fi
+
+if (cd "$ROOT" && npm run context-map -- --include-runtime --output "$TMPDIR_EVAL/runtime-context-map.md") >"$TMPDIR_EVAL/runtime.out" 2>"$TMPDIR_EVAL/runtime.err" &&
+  rg -q 'Current Workflow State' "$TMPDIR_EVAL/runtime-context-map.md"; then
+  _pass "context map supports optional runtime workflow state"
+else
+  _fail "runtime context map generation failed"
+fi
+
+for expected in \
+  'Repository Shape' \
+  'Core Commands' \
+  'Workflow Sidecars' \
+  'Workflow Skills' \
+  'Support Skills' \
+  'Agents' \
+  'Optional Powers' \
+  'Packs' \
+  'Context Loading Rules' \
+  'npm run context-map:check' \
+  'packaging/packs.json' \
+  'workflow-release.schema.json' \
+  'workflow-learning.schema.json' \
+  'plan-work' \
+  'tool-planner' \
+  'Eval-first execution' \
+  'Research-before-coding workflow' \
+  'Save durable knowledge'; do
+  if rg -q "$expected" "$MAP"; then
+    _pass "context map includes $expected"
+  else
+    _fail "context map missing $expected"
+  fi
+done
+
+if [[ "$errors" -eq 0 ]]; then
+  echo "Context map integration passed."
+  exit 0
+fi
+
+echo "Context map integration failed: $errors issue(s)."
+exit 1

package/evals/integration/test_effective_backlog_settings.sh

@@ -0,0 +1,58 @@
+#!/usr/bin/env bash
+set -uo pipefail
+
+ROOT="$(cd "$(dirname "${BASH_SOURCE[0]}")/../.." && pwd)"
+SCRIPT="$ROOT/scripts/effective-backlog-settings.js"
+PROJECT="$ROOT/context/settings/backlog-provider-settings.json"
+GLOBAL="$ROOT/evals/fixtures/backlog-provider-settings/global-default.json"
+OVERRIDE="$ROOT/evals/fixtures/backlog-provider-settings/project-override.json"
+EMPTY="$(mktemp)"
+trap 'rm -f "$EMPTY"' EXIT
+printf '{"schema_version":"1.0","projects":[]}\n' > "$EMPTY"
+
+errors=0
+pass() { echo "  ✓ $1"; }
+fail() { echo "  ✗ $1"; errors=$((errors + 1)); }
+
+json_value() {
+  node -e 'const fs=require("fs"); let cur=JSON.parse(fs.readFileSync(0,"utf8")); for (const part of process.argv[1].split(".")) cur=Array.isArray(cur) ? cur[Number(part)] : cur[part]; console.log(cur);' "$1"
+}
+
+echo "=== Effective Backlog Settings ==="
+
+default_result="$(cd "$ROOT" && npm run --silent effective-backlog-settings -- --repo-path . --json)"
+[[ "$(printf '%s' "$default_result" | json_value status)" == "configured" ]] && pass "default project settings resolve through npm command" || fail "default project settings resolve through npm command"
+[[ "$(printf '%s' "$default_result" | json_value settings.work_item_provider.kind)" == "github" ]] && pass "default provider kind is github" || fail "default provider kind is github"
+[[ "$(printf '%s' "$default_result" | json_value settings.work_item_provider.repo.owner)" == "kontourai" ]] && pass "default repo owner is kontourai" || fail "default repo owner is kontourai"
+[[ "$(printf '%s' "$default_result" | json_value settings.work_item_provider.repo.name)" == "flow-agents" ]] && pass "default repo name is flow-agents" || fail "default repo name is flow-agents"
+[[ "$(printf '%s' "$default_result" | json_value settings.board_provider.board.number)" == "1" ]] && pass "default project number is 1" || fail "default project number is 1"
+! printf '%s' "$default_result" | rg -q '/build/context/settings/backlog-provider-settings.json' && pass "default settings path is not build-relative" || fail "default settings path is not build-relative"
+
+configured="$(node "$SCRIPT" --repo-path "$ROOT" --project-settings "$PROJECT" --global-settings "$EMPTY" --json)"
+[[ "$(printf '%s' "$configured" | json_value status)" == "configured" ]] && pass "configured repo resolves" || fail "configured repo resolves"
+[[ "$(printf '%s' "$configured" | json_value settings.work_item_provider.kind)" == "github" ]] && pass "provider kind is github" || fail "provider kind is github"
+[[ "$(printf '%s' "$configured" | json_value settings.work_item_provider.repo.owner)" == "kontourai" ]] && pass "repo owner is kontourai" || fail "repo owner is kontourai"
+[[ "$(printf '%s' "$configured" | json_value settings.work_item_provider.repo.name)" == "flow-agents" ]] && pass "repo name is flow-agents" || fail "repo name is flow-agents"
+[[ "$(printf '%s' "$configured" | json_value settings.board_provider.board.number)" == "1" ]] && pass "project number is 1" || fail "project number is 1"
+printf '%s' "$configured" | rg -F -q '"ready_statuses": [' && pass "filters include ready statuses" || fail "filters include ready statuses"
+printf '%s' "$configured" | rg -F -q '"include_labels": []' && pass "filters do not require labels by default" || fail "filters do not require labels by default"
+printf '%s' "$configured" | rg -q '"prefer_finishing_active_work": true' && pass "WIP policy prefers finishing active work" || fail "WIP policy prefers finishing active work"
+
+override_result="$(node "$SCRIPT" --repo-path "$ROOT" --project-settings "$OVERRIDE" --global-settings "$GLOBAL" --json)"
+[[ "$(printf '%s' "$override_result" | json_value source)" == "project" ]] && pass "project settings override global defaults" || fail "project settings override global defaults"
+[[ "$(printf '%s' "$override_result" | json_value settings.board_provider.board.number)" == "1" ]] && pass "project override keeps Project 1" || fail "project override keeps Project 1"
+! printf '%s' "$override_result" | rg -q 'global-ready' && pass "project override replaces global filters" || fail "project override replaces global filters"
+
+missing_output="$(node "$SCRIPT" --repo-path "$ROOT" --project-settings "$EMPTY" --global-settings "$EMPTY" --json 2>/dev/null)"
+missing_status=$?
+[[ "$missing_status" -eq 2 ]] && pass "missing settings exits with ask-user status" || fail "missing settings exits with ask-user status"
+[[ "$(printf '%s' "$missing_output" | json_value status)" == "ask_user" ]] && pass "missing settings reports ask_user" || fail "missing settings reports ask_user"
+printf '%s' "$missing_output" | rg -q 'WorkItemProvider and BoardProvider' && pass "missing settings names provider roles" || fail "missing settings names provider roles"
+
+if [[ "$errors" -eq 0 ]]; then
+  echo "Effective backlog settings checks passed"
+else
+  echo "Effective backlog settings checks failed: $errors"
+fi
+
+exit "$errors"

package/evals/integration/test_fixture_retirement_audit.sh

@@ -0,0 +1,58 @@
+#!/usr/bin/env bash
+set -uo pipefail
+
+ROOT="$(cd "$(dirname "${BASH_SOURCE[0]}")/../.." && pwd)"
+source "$ROOT/evals/lib/node.sh"
+
+TMPDIR_EVAL="$(mktemp -d)"
+trap 'rm -rf "$TMPDIR_EVAL"' EXIT
+
+errors=0
+pass() { echo "  ✓ $1"; }
+fail() { echo "  ✗ $1"; errors=$((errors + 1)); }
+
+echo "=== Fixture Retirement Audit ==="
+
+flow_agents_node fixture-retirement-audit --json > "$TMPDIR_EVAL/audit.json"
+status=$?
+[[ "$status" -eq 0 ]] && pass "fixture audit exits successfully" || fail "fixture audit exits successfully"
+
+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" "$2"
+}
+
+[[ "$(json_query "$TMPDIR_EVAL/audit.json" "totals.scanned")" == "10" ]] && pass "audit scans all fixture groups" || fail "audit scans all fixture groups"
+[[ "$(json_query "$TMPDIR_EVAL/audit.json" "totals.retire_candidates")" == "0" ]] && pass "audit finds no unowned retire candidates" || fail "audit finds no unowned retire candidates"
+[[ "$(json_query "$TMPDIR_EVAL/audit.json" "totals.kept")" == "10" ]] && pass "audit keeps all owned fixture groups" || fail "audit keeps all owned fixture groups"
+
+node - "$TMPDIR_EVAL/audit.json" <<'NODE'
+const fs = require("node:fs");
+const audit = JSON.parse(fs.readFileSync(process.argv[2], "utf8"));
+for (const item of audit.fixtures) {
+  if (item.classification !== "kept") throw new Error(`${item.fixture} should be kept`);
+  if (!item.owners.length) throw new Error(`${item.fixture} has no owners`);
+  if (!item.direct_refs.length) throw new Error(`${item.fixture} has no direct refs`);
+  if (!item.reasons.includes("owned fixture with live eval/script references")) throw new Error(`${item.fixture} missing keep reason`);
+}
+NODE
+status=$?
+[[ "$status" -eq 0 ]] && pass "all kept fixtures have owners and live refs" || fail "all kept fixtures have owners and live refs"
+
+flow_agents_node fixture-retirement-audit > "$TMPDIR_EVAL/audit.txt"
+grep -q "Retire candidates: 0" "$TMPDIR_EVAL/audit.txt" && pass "text audit reports zero retire candidates" || fail "text audit reports zero retire candidates"
+grep -q "kept: evals/fixtures/hook-influence" "$TMPDIR_EVAL/audit.txt" && pass "text audit lists kept hook influence fixture" || fail "text audit lists kept hook influence fixture"
+
+flow_agents_node fixture-retirement-audit --help > "$TMPDIR_EVAL/help.txt"
+if grep -Eq -- "--(apply|delete|archive)" "$TMPDIR_EVAL/help.txt"; then
+  fail "help does not advertise destructive fixture actions"
+else
+  pass "help does not advertise destructive fixture actions"
+fi
+
+if [[ "$errors" -eq 0 ]]; then
+  echo "Fixture retirement audit checks passed"
+else
+  echo "Fixture retirement audit checks failed: $errors"
+fi
+
+exit "$errors"

package/evals/integration/test_flow_agents_statusline.sh

@@ -0,0 +1,93 @@
+#!/usr/bin/env bash
+set -euo pipefail
+
+ROOT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")/../.." && pwd)"
+TMPDIR_EVAL="$(mktemp -d /tmp/flow-agents-statusline.XXXXXX)"
+pass=0
+fail=0
+
+cleanup() {
+  rm -rf "$TMPDIR_EVAL"
+}
+trap cleanup EXIT
+
+_pass() { echo "  ✓ $1"; pass=$((pass + 1)); }
+_fail() { echo "  ✗ $1"; fail=$((fail + 1)); }
+
+echo "=== Layer 2: Flow Agents Statusline ==="
+echo ""
+
+WORKSPACE="$TMPDIR_EVAL/workspace"
+TASK_DIR="$WORKSPACE/.flow-agents/status-demo"
+mkdir -p "$TASK_DIR"
+
+cat >"$WORKSPACE/.flow-agents/current.json" <<'JSON'
+{
+  "schema_version": "1.0",
+  "active_slug": "status-demo",
+  "artifact_dir": "status-demo",
+  "owner": "codex",
+  "updated_at": "2026-05-25T00:00:00Z",
+  "source": "test"
+}
+JSON
+
+cat >"$TASK_DIR/state.json" <<'JSON'
+{
+  "schema_version": "1.0",
+  "task_slug": "status-demo",
+  "status": "needs_decision",
+  "phase": "evidence",
+  "updated_at": "2026-05-25T00:01:00Z",
+  "next_action": {
+    "status": "needs_user",
+    "summary": "Review the release hold and decide whether the missing approval is accepted.",
+    "target_phase": "release"
+  }
+}
+JSON
+
+cat >"$TASK_DIR/acceptance.json" <<'JSON'
+{
+  "schema_version": "1.0",
+  "task_slug": "status-demo",
+  "criteria": [
+    {"id": "one", "description": "First criterion", "status": "pass"},
+    {"id": "two", "description": "Second criterion", "status": "pending"},
+    {"id": "three", "description": "Third criterion", "status": "accepted_gap"}
+  ],
+  "goal_fit": {"status": "pending", "summary": "Pending"}
+}
+JSON
+
+if output="$(cd "$WORKSPACE" && node "$ROOT_DIR/scripts/statusline/flow-agents-statusline.js" <<JSON
+{"cwd":"$WORKSPACE"}
+JSON
+)"; then
+  if [[ "$output" == *"Flow Agents: status-demo"* && "$output" == *"evidence/needs_decision"* && "$output" == *"2/3 AC"* && "$output" == *"next:"* ]]; then
+    _pass "statusline renders active workflow, phase/status, progress, and next action"
+  else
+    _fail "statusline output missing expected fields: $output"
+  fi
+else
+  _fail "statusline command failed"
+fi
+
+EMPTY="$TMPDIR_EVAL/empty"
+mkdir -p "$EMPTY"
+if output="$(cd "$EMPTY" && node "$ROOT_DIR/scripts/statusline/flow-agents-statusline.js" <<JSON
+{"cwd":"$EMPTY"}
+JSON
+)"; then
+  if [[ "$output" == "Flow Agents: no active workflow" ]]; then
+    _pass "statusline handles workspaces without workflow state"
+  else
+    _fail "statusline empty-workspace output was unexpected: $output"
+  fi
+else
+  _fail "statusline empty-workspace command failed"
+fi
+
+echo ""
+echo "Statusline results: $pass passed, $fail failed"
+[[ $fail -eq 0 ]]

package/evals/integration/test_flow_kit_repository.sh

@@ -0,0 +1,90 @@
+#!/usr/bin/env bash
+# test_flow_kit_repository.sh — Validate local Flow Kit repository fixtures and diagnostics.
+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)); }
+
+run_validator() {
+  local fixture="$1"
+  local output="$2"
+  flow_agents_node "$ROOT/scripts/validate-source-tree.js" --kit "$ROOT/evals/fixtures/flow-kit-repository/$fixture" >"$output" 2>&1
+}
+
+expect_pass() {
+  local fixture="$1"
+  local output="$TMP_DIR/$fixture.out"
+  if run_validator "$fixture" "$output"; then
+    if rg -q 'Flow Kit repository validation passed' "$output"; then
+      pass "$fixture passes local kit repository validation"
+    else
+      fail "$fixture did not print pass diagnostic"
+      sed -n '1,120p' "$output"
+    fi
+  else
+    fail "$fixture should pass local kit repository validation"
+    sed -n '1,120p' "$output"
+  fi
+}
+
+expect_fail() {
+  local fixture="$1"
+  local pattern="$2"
+  local output="$TMP_DIR/$fixture.out"
+  if run_validator "$fixture" "$output"; then
+    fail "$fixture should fail local kit repository validation"
+    sed -n '1,120p' "$output"
+    return
+  fi
+  if rg -q "$pattern" "$output"; then
+    pass "$fixture fails with actionable diagnostic"
+  else
+    fail "$fixture diagnostic missing pattern: $pattern"
+    sed -n '1,160p' "$output"
+  fi
+}
+
+echo "=== Flow Kit Repository Fixture Checks ==="
+expect_pass "valid-local-kit"
+expect_fail "invalid-schema-version" '\.schema_version must be "1\.0"'
+expect_fail "invalid-missing-schema-version" '\.schema_version must be "1\.0"'
+expect_fail "invalid-id" '\.id must be a stable kebab-case string'
+expect_fail "invalid-missing-id" '\.id must be a stable kebab-case string'
+expect_fail "invalid-name" '\.name must be a non-empty string'
+expect_fail "invalid-missing-flow" 'flows\[0\]\.path points at missing Flow Definition'
+expect_fail "invalid-absolute-path" 'flows\[0\]\.path must be relative'
+expect_fail "invalid-traversal" "flows\\[0\\]\\.path must stay inside the kit directory"
+expect_fail "invalid-malformed-json" 'invalid JSON'
+expect_fail "invalid-asset-section" '\.docs must be a list'
+expect_fail "invalid-duplicate-flow" "flows\\[1\\]\\.id duplicates"
+
+echo ""
+echo "=== Builder Kit Shared Validation Check ==="
+builder_output="$TMP_DIR/source-tree.out"
+if flow_agents_node "$ROOT/scripts/validate-source-tree.js" >"$builder_output" 2>&1; then
+  if rg -q 'Source tree validation passed' "$builder_output"; then
+    pass "source-tree validation keeps Builder Kit on shared validation path"
+  else
+    fail "source-tree validation did not print pass diagnostic"
+    sed -n '1,160p' "$builder_output"
+  fi
+else
+  fail "source-tree validation failed while checking Builder Kit"
+  sed -n '1,220p' "$builder_output"
+fi
+
+echo ""
+if [[ "$errors" -eq 0 ]]; then
+  echo "Flow Kit repository fixture checks passed."
+  exit 0
+fi
+
+echo "Flow Kit repository fixture checks failed: $errors issue(s)."
+exit 1