@kontourai/flow-agents 2.3.0 → 3.0.0

package/evals/integration/test_command_log_integrity.sh

@@ -32,11 +32,11 @@ trap cleanup EXIT
 # ── helper: seed a minimal delivered workflow artifact ────────────────────────
 seed_repo() { # $1=dir $2=slug
   local p="$1" slug="$2"
-  mkdir -p "$p/.flow-agents/$slug"
+  mkdir -p "$p/.kontourai/flow-agents/$slug"
   printf '# Repo\n' > "$p/AGENTS.md"
   printf '%s' "{\"schema_version\":\"1.0\",\"task_slug\":\"$slug\",\"status\":\"delivered\",\"phase\":\"done\",\"updated_at\":\"2026-06-23T00:00:00Z\",\"next_action\":{\"status\":\"done\",\"summary\":\"done\"}}" \
-    > "$p/.flow-agents/$slug/state.json"
-  cat > "$p/.flow-agents/$slug/$slug--deliver.md" << MD
+    > "$p/.kontourai/flow-agents/$slug/state.json"
+  cat > "$p/.kontourai/flow-agents/$slug/$slug--deliver.md" << MD
 # $slug
 
 branch: main
@@ -71,12 +71,12 @@ echo "Test 1: altered entry (flip fail→pass without fixing hash) → broken
 T1="$TMP/t1"; seed_repo "$T1" t1
 write_chained_log "$T1" t1
 
-LOG="$T1/.flow-agents/t1/command-log.jsonl"
+LOG="$T1/.kontourai/flow-agents/t1/command-log.jsonl"
 
 if [[ -f "$LOG" ]]; then _pass "T1: command-log.jsonl written"; else _fail "T1: command-log.jsonl missing"; fi
 
 # Verify clean chain (before tamper)
-chain_status=$(node -e "const g = require('$GATE'); const r = g.verifyCommandLogChain('$T1/.flow-agents/t1'); console.log(r.status);")
+chain_status=$(node -e "const g = require('$GATE'); const r = g.verifyCommandLogChain('$T1/.kontourai/flow-agents/t1'); console.log(r.status);")
 if [[ "$chain_status" == "ok" ]]; then
   _pass "T1: untampered chain verifies as ok"
 else
@@ -97,7 +97,7 @@ open(sys.argv[1], 'w').write('\n'.join(lines) + '\n')
 PY
 
 # Verify broken chain
-chain_after=$(node -e "const g = require('$GATE'); const r = g.verifyCommandLogChain('$T1/.flow-agents/t1'); console.log(r.status + ':' + r.brokenAt);")
+chain_after=$(node -e "const g = require('$GATE'); const r = g.verifyCommandLogChain('$T1/.kontourai/flow-agents/t1'); console.log(r.status + ':' + r.brokenAt);")
 if [[ "$chain_after" == "broken:1" ]]; then
   _pass "T1: tampered entry detected → broken at entry 1"
 else
@@ -108,7 +108,7 @@ fi
 # The tampered entry (lint) was a FAIL flipped to PASS — so the log now shows a false pass.
 # Since chain is broken, gate should block with integrity warning and NOT trust log passes.
 printf '%s' '{"schema_version":"1.0","task_slug":"t1","verdict":"pass","checks":[{"id":"npm-test","kind":"command","status":"pass","command":"npm test","summary":"passed"}]}' \
-  > "$T1/.flow-agents/t1/evidence.json"
+  > "$T1/.kontourai/flow-agents/t1/evidence.json"
 
 set +e
 gate_out=$(FLOW_AGENTS_GOAL_FIT_MODE=block FLOW_AGENTS_GOAL_FIT_BACKSTOP=skip \
@@ -141,7 +141,7 @@ echo "Test 2: removed/reordered entry → linkage breaks → broken → gate fla
 T2="$TMP/t2"; seed_repo "$T2" t2
 write_chained_log "$T2" t2
 
-LOG2="$T2/.flow-agents/t2/command-log.jsonl"
+LOG2="$T2/.kontourai/flow-agents/t2/command-log.jsonl"
 lines_before=$(wc -l < "$LOG2" | tr -d ' ')
 
 # Reorder: swap entry 0 and entry 1
@@ -153,7 +153,7 @@ lines[0], lines[1] = lines[1], lines[0]
 open(sys.argv[1], 'w').write('\n'.join(lines) + '\n')
 PY
 
-chain_reorder=$(node -e "const g = require('$GATE'); const r = g.verifyCommandLogChain('$T2/.flow-agents/t2'); console.log(r.status);")
+chain_reorder=$(node -e "const g = require('$GATE'); const r = g.verifyCommandLogChain('$T2/.kontourai/flow-agents/t2'); console.log(r.status);")
 if [[ "$chain_reorder" == "broken" ]]; then
   _pass "T2: reordered entries detected → broken"
 else
@@ -163,7 +163,7 @@ fi
 # Test: delete middle entry (restore then delete entry 0 so entry 1's prevHash is wrong)
 write_chained_log "$T2" t2  # re-append fresh entries (now 4 total — but that's fine for test)
 # Write a fresh log with just 2 entries and then delete the first
-LOG2_FRESH="$T2/.flow-agents/t2/command-log.jsonl"
+LOG2_FRESH="$T2/.kontourai/flow-agents/t2/command-log.jsonl"
 python3 - "$LOG2_FRESH" << 'PY'
 import sys
 lines = [l for l in open(sys.argv[1]).read().strip().split('\n') if l.strip()]
@@ -173,7 +173,7 @@ last2 = lines[-2:]
 open(sys.argv[1], 'w').write(last2[1] + '\n')
 PY
 
-chain_delete=$(node -e "const g = require('$GATE'); const r = g.verifyCommandLogChain('$T2/.flow-agents/t2'); console.log(r.status);")
+chain_delete=$(node -e "const g = require('$GATE'); const r = g.verifyCommandLogChain('$T2/.kontourai/flow-agents/t2'); console.log(r.status);")
 if [[ "$chain_delete" == "broken" ]]; then
   _pass "T2: removed predecessor entry detected → broken (prevHash mismatch)"
 else
@@ -191,7 +191,7 @@ printf '{"hook_event_name":"PostToolUse","tool_name":"Bash","cwd":"%s","tool_inp
 printf '{"hook_event_name":"PostToolUse","tool_name":"Bash","cwd":"%s","tool_input":{"command":"npm run build"},"tool_response":{"exitCode":1}}' "$T3" \
   | node "$CAPTURE" >/dev/null 2>&1
 
-chain_legit=$(node -e "const g = require('$GATE'); const r = g.verifyCommandLogChain('$T3/.flow-agents/t3'); console.log(r.status);")
+chain_legit=$(node -e "const g = require('$GATE'); const r = g.verifyCommandLogChain('$T3/.kontourai/flow-agents/t3'); console.log(r.status);")
 if [[ "$chain_legit" == "ok" ]]; then
   _pass "T3: untampered chained log verifies ok"
 else
@@ -200,7 +200,7 @@ fi
 
 # Evidence claims npm run build passed (it actually failed → capture log shows fail → block)
 printf '%s' '{"schema_version":"1.0","task_slug":"t3","verdict":"pass","checks":[{"id":"build","kind":"command","status":"pass","command":"npm run build","summary":"build passed"}]}' \
-  > "$T3/.flow-agents/t3/evidence.json"
+  > "$T3/.kontourai/flow-agents/t3/evidence.json"
 
 set +e
 gate3_out=$(FLOW_AGENTS_GOAL_FIT_MODE=block FLOW_AGENTS_GOAL_FIT_BACKSTOP=skip \
@@ -234,9 +234,9 @@ T4="$TMP/t4"; seed_repo "$T4" t4
 
 # Write a legacy-style log (no _chain field) — exactly like pre-B2 fixtures
 printf '%s\n' '{"command":"npm test","observedResult":"fail","exitCode":1,"capturedAt":"2026-06-23T00:00:00Z","source":"postToolUse-capture"}' \
-  > "$T4/.flow-agents/t4/command-log.jsonl"
+  > "$T4/.kontourai/flow-agents/t4/command-log.jsonl"
 
-chain_legacy=$(node -e "const g = require('$GATE'); const r = g.verifyCommandLogChain('$T4/.flow-agents/t4'); console.log(r.status);")
+chain_legacy=$(node -e "const g = require('$GATE'); const r = g.verifyCommandLogChain('$T4/.kontourai/flow-agents/t4'); console.log(r.status);")
 if [[ "$chain_legacy" == "legacy" ]]; then
   _pass "T4: unchained (legacy) log returns legacy status"
 else
@@ -245,7 +245,7 @@ fi
 
 # Evidence claims npm test passed, but legacy log shows it failed → still blocks
 printf '%s' '{"schema_version":"1.0","task_slug":"t4","verdict":"pass","checks":[{"id":"unit-tests","kind":"command","status":"pass","command":"npm test","summary":"passed"}]}' \
-  > "$T4/.flow-agents/t4/evidence.json"
+  > "$T4/.kontourai/flow-agents/t4/evidence.json"
 
 set +e
 gate4_out=$(FLOW_AGENTS_GOAL_FIT_MODE=block FLOW_AGENTS_GOAL_FIT_BACKSTOP=skip \

package/evals/integration/test_console_learning_projection.sh

@@ -102,7 +102,7 @@ if jq -e --arg generated "$GENERATED_AT" '
   .derivedFrom.eventHistory == "unavailable" and
   .derivedFrom.directSnapshot.emittedAt == $generated and
   .derivedFrom.directSnapshot.producer == {"id":"flow-agents-learning","product":"flow-agents"} and
-  .derivedFrom.directSnapshot.sourceRef == {"product":"flow-agents","kind":"workflow-learning","id":".flow-agents/*/learning.json","label":"Local workflow learning sidecars"} and
+  .derivedFrom.directSnapshot.sourceRef == {"product":"flow-agents","kind":"workflow-learning","id":".kontourai/flow-agents/*/learning.json","label":"Local workflow learning sidecars"} and
   (.learnings | length) == 2
 ' "$PROJECTION" >/dev/null 2>&1; then
   _pass "projection envelope includes Console schema, scope, producer, and direct snapshot provenance"

package/evals/integration/test_critique_supersession_roundtrip.sh

@@ -0,0 +1,182 @@
+#!/usr/bin/env bash
+# test_critique_supersession_roundtrip.sh — Regression evals for the trust-ledger defects
+# fixed under #267 (critique supersession), #268 (lossless, type-stable check/critique
+# round-trip), and #282 (first-class superseded_by affordance).
+#
+# Proves:
+#   (a) record-critique twice for the same id (fail → pass, SAME reviewer) supersedes the fail:
+#       the effective/live state is the pass, the historical fail is retained (status=superseded,
+#       metadata.superseded_by) but excluded from evaluation, no duplicate claim ids, and CI
+#       trust-reconcile exits 0.
+#   (b) HEADLINE: a --flow-id (builder.build/verify) session — ensure-session → record-evidence
+#       (mixed session-local kinds) → record-critique → trust-reconcile — exits 0. This exact flow
+#       could never converge before (critique claims were re-absorbed as command-less test_output
+#       checks → permanent [not-run] divergence; check kinds collapsed to the declared claimType).
+#   (c) record-evidence AFTER a critique preserves the critique history (previously hardcoded []).
+#   (d) ANTI-GAMING: a DIFFERENT reviewer cannot supersede a reviewer's fail — the disputed
+#       critique stays live and trust-reconcile still exits 1 (worker cannot bury a reviewer finding).
+#   (e) VALIDATOR (#282): a top-level pass tolerates a superseded historical fail member, but a
+#       LIVE (non-superseded) fail still triggers "required critique must pass".
+#
+# Deterministic, no model spend, self-cleaning.
+# Usage: bash evals/integration/test_critique_supersession_roundtrip.sh
+
+set -uo pipefail
+
+ROOT="$(cd "$(dirname "${BASH_SOURCE[0]}")/../.." && pwd)"
+source "$ROOT/evals/lib/node.sh"
+RECON="$ROOT/scripts/ci/trust-reconcile.js"
+WRITER="workflow-sidecar"
+
+TMP="$(mktemp -d)"
+errors=0
+_pass() { echo "  PASS: $1"; }
+_fail() { echo "  FAIL: $1"; errors=$((errors + 1)); }
+cleanup() { rm -rf "$TMP"; }
+trap cleanup EXIT
+
+# seed <aroot> <slug> [flow-args...] — ensure-session + init-plan
+seed() {
+  local aroot="$1"; local slug="$2"; shift 2
+  mkdir -p "$aroot"
+  flow_agents_node "$WRITER" ensure-session --artifact-root "$aroot" --task-slug "$slug" \
+    --title "T" --summary "S" --timestamp "2026-07-01T00:00:00Z" "$@" >/dev/null 2>&1
+  flow_agents_node "$WRITER" init-plan "$aroot/$slug/$slug--deliver.md" \
+    --source-request "R" --summary "S" --timestamp "2026-07-01T00:00:00Z" >/dev/null 2>&1
+}
+
+# ─── (a) same-reviewer supersession: fail → pass ──────────────────────────────
+echo ""
+echo "=== (a) record-critique fail→pass (same reviewer) supersedes, history intact, reconcile 0 ==="
+A_AROOT="$TMP/a/aroot"; A_SLUG="supersede-same"; A_DIR="$A_AROOT/$A_SLUG"
+seed "$A_AROOT" "$A_SLUG"
+flow_agents_node "$WRITER" record-evidence "$A_DIR" --verdict pass \
+  --check-json '{"id":"c1","kind":"diff","status":"pass","summary":"diff check"}' \
+  --timestamp "2026-07-01T00:01:00Z" >/dev/null 2>&1
+flow_agents_node "$WRITER" record-critique "$A_DIR" --id rv --reviewer alice --verdict fail \
+  --summary "found a bug" --timestamp "2026-07-01T00:02:00Z" >/dev/null 2>&1
+flow_agents_node "$WRITER" record-critique "$A_DIR" --id rv --reviewer alice --verdict pass \
+  --summary "bug fixed" --timestamp "2026-07-01T00:03:00Z" >/dev/null 2>&1
+
+node - "$A_DIR/trust.bundle" << 'NODE'
+const fs = require('fs');
+const b = JSON.parse(fs.readFileSync(process.argv[2], 'utf8'));
+const crit = (b.claims||[]).filter(c => c.metadata && c.metadata.origin === 'critique');
+const live = crit.filter(c => !c.metadata.superseded_by);
+const hist = crit.filter(c => c.metadata.superseded_by);
+if (crit.length !== 2) throw new Error('expected 2 critique claims (1 live + 1 history), got ' + crit.length);
+if (live.length !== 1 || live[0].value !== 'pass') throw new Error('expected exactly 1 LIVE critique with value=pass, got ' + JSON.stringify(live.map(c=>c.value)));
+if (hist.length !== 1 || hist[0].value !== 'fail' || hist[0].status !== 'superseded') throw new Error('expected historical fail with status=superseded');
+if (!hist[0].metadata.superseded_by) throw new Error('history missing first-class metadata.superseded_by');
+const ids = (b.claims||[]).map(c=>c.id); const dup = ids.filter((x,i)=>ids.indexOf(x)!==i);
+if (dup.length) throw new Error('duplicate claim ids in bundle: ' + dup.join(','));
+console.log('live=pass, history=fail(superseded), no duplicate ids');
+NODE
+if [[ $? -eq 0 ]]; then _pass "(a) effective state=pass, historical fail retained+superseded, no dup ids"; else _fail "(a) supersession/history assertion failed"; fi
+
+TRUST_RECONCILE_COMMANDS="true" node "$RECON" --bundle "$A_DIR/trust.bundle" --repo-root "$TMP/a" >"$TMP/a-recon.log" 2>&1
+if [[ $? -eq 0 ]]; then _pass "(a) trust-reconcile exits 0 (resolved session converges)"; else _fail "(a) trust-reconcile did NOT converge: $(cat "$TMP/a-recon.log")"; fi
+
+# ─── (b) HEADLINE: --flow-id session converges ────────────────────────────────
+echo ""
+echo "=== (b) HEADLINE --flow-id builder.build/verify: mixed evidence + critique → reconcile 0 ==="
+B_AROOT="$TMP/b/aroot"; B_SLUG="flowid-converge"; B_DIR="$B_AROOT/$B_SLUG"
+seed "$B_AROOT" "$B_SLUG" --flow-id builder.build --step-id verify
+flow_agents_node "$WRITER" record-evidence "$B_DIR" --verdict pass \
+  --check-json '{"id":"k-diff","kind":"diff","status":"pass","summary":"diff excerpt"}' \
+  --check-json '{"id":"k-policy","kind":"policy","status":"pass","summary":"policy rule"}' \
+  --check-json '{"id":"k-ext","kind":"external","status":"pass","summary":"attested"}' \
+  --timestamp "2026-07-01T00:01:00Z" >/dev/null 2>&1
+flow_agents_node "$WRITER" record-critique "$B_DIR" --id code-review --reviewer alice --verdict pass \
+  --summary "looks good" --timestamp "2026-07-01T00:02:00Z" >/dev/null 2>&1
+
+node - "$B_DIR/trust.bundle" << 'NODE'
+const fs = require('fs');
+const b = JSON.parse(fs.readFileSync(process.argv[2], 'utf8'));
+// Type stability: no critique claim may be re-absorbed as a check, and check kinds/evidenceTypes
+// must survive the round-trip (a session-local kind must NOT flip to test_output).
+const checks = (b.claims||[]).filter(c => c.metadata && c.metadata.origin === 'check');
+const crit = (b.claims||[]).filter(c => c.metadata && c.metadata.origin === 'critique');
+if (crit.length !== 1) throw new Error('expected exactly 1 critique claim (no re-absorption/duplication), got ' + crit.length);
+const evByClaim = {}; for (const e of (b.evidence||[])) (evByClaim[e.claimId] ||= []).push(e);
+for (const c of checks) {
+  const ets = (evByClaim[c.id]||[]).map(e=>e.evidenceType);
+  if (ets.includes('test_output')) throw new Error('session-local check ' + c.id + ' flipped to test_output evidence (round-trip kind instability)');
+}
+console.log(checks.length + ' check claims, ' + crit.length + ' critique claim; no test_output flip');
+NODE
+if [[ $? -eq 0 ]]; then _pass "(b) round-trip type-stable under --flow-id (no critique→check absorption, no evidenceType flip)"; else _fail "(b) round-trip type instability under --flow-id"; fi
+
+TRUST_RECONCILE_COMMANDS="true" node "$RECON" --bundle "$B_DIR/trust.bundle" --repo-root "$TMP/b" >"$TMP/b-recon.log" 2>&1
+if [[ $? -eq 0 ]]; then _pass "(b) HEADLINE: --flow-id session trust-reconcile exits 0 (converges)"; else _fail "(b) HEADLINE --flow-id session did NOT converge: $(cat "$TMP/b-recon.log")"; fi
+
+# ─── (c) record-evidence after critiques preserves history ────────────────────
+echo ""
+echo "=== (c) record-evidence AFTER a critique preserves critique history (#268) ==="
+flow_agents_node "$WRITER" record-evidence "$B_DIR" --verdict pass \
+  --check-json '{"id":"k-diff2","kind":"diff","status":"pass","summary":"another diff"}' \
+  --timestamp "2026-07-01T00:04:00Z" >/dev/null 2>&1
+node - "$B_DIR/trust.bundle" << 'NODE'
+const fs = require('fs');
+const b = JSON.parse(fs.readFileSync(process.argv[2], 'utf8'));
+const crit = (b.claims||[]).filter(c => c.metadata && c.metadata.origin === 'critique');
+if (crit.length !== 1) throw new Error('critique history dropped by record-evidence — expected 1, got ' + crit.length);
+if (crit[0].value !== 'pass') throw new Error('critique value changed');
+console.log('critique survived record-evidence');
+NODE
+if [[ $? -eq 0 ]]; then _pass "(c) critique history survives a later record-evidence call"; else _fail "(c) record-evidence dropped critique history"; fi
+
+# ─── (d) ANTI-GAMING: cross-reviewer cannot supersede a reviewer fail ─────────
+echo ""
+echo "=== (d) ANTI-GAMING: a different reviewer cannot supersede a reviewer's fail ==="
+D_AROOT="$TMP/d/aroot"; D_SLUG="supersede-cross"; D_DIR="$D_AROOT/$D_SLUG"
+seed "$D_AROOT" "$D_SLUG"
+flow_agents_node "$WRITER" record-evidence "$D_DIR" --verdict pass \
+  --check-json '{"id":"c1","kind":"diff","status":"pass","summary":"diff check"}' \
+  --timestamp "2026-07-01T00:01:00Z" >/dev/null 2>&1
+flow_agents_node "$WRITER" record-critique "$D_DIR" --id rv --reviewer reviewer-bob --verdict fail \
+  --summary "reviewer found a real bug" --timestamp "2026-07-01T00:02:00Z" >/dev/null 2>&1
+flow_agents_node "$WRITER" record-critique "$D_DIR" --id rv --reviewer worker-mallory --verdict pass \
+  --summary "worker claims fixed" --timestamp "2026-07-01T00:03:00Z" >/dev/null 2>&1
+node - "$D_DIR/trust.bundle" << 'NODE'
+const fs = require('fs');
+const b = JSON.parse(fs.readFileSync(process.argv[2], 'utf8'));
+const crit = (b.claims||[]).filter(c => c.metadata && c.metadata.origin === 'critique');
+const live = crit.filter(c => !c.metadata.superseded_by);
+const liveFail = live.filter(c => c.value === 'fail' || c.status === 'disputed');
+if (!liveFail.length) throw new Error('ANTI-GAMING VIOLATION: reviewer fail was superseded by a different reviewer');
+if (crit.some(c => c.metadata.superseded_by)) throw new Error('unexpected supersession across reviewers');
+console.log('reviewer fail stays LIVE (' + liveFail.length + '); cross-reviewer supersession refused');
+NODE
+if [[ $? -eq 0 ]]; then _pass "(d) reviewer's fail is NOT superseded by a different reviewer (stays live)"; else _fail "(d) ANTI-GAMING regression: cross-reviewer supersession occurred"; fi
+TRUST_RECONCILE_COMMANDS="true" node "$RECON" --bundle "$D_DIR/trust.bundle" --repo-root "$TMP/d" >"$TMP/d-recon.log" 2>&1
+if [[ $? -ne 0 ]] && grep -q "session-local-failed" "$TMP/d-recon.log"; then _pass "(d) trust-reconcile still exits 1 (live disputed critique blocks)"; else _fail "(d) reconcile should have blocked on the live reviewer fail"; fi
+
+# ─── (e) VALIDATOR (#282): superseded fail tolerated, live fail blocks ────────
+echo ""
+echo "=== (e) VALIDATOR: superseded historical fail tolerated; live fail blocks (#282) ==="
+# Reuse (a)'s superseded bundle: pass top-level with a superseded fail member.
+va_out="$(flow_agents_node validate-workflow-artifacts "$A_AROOT" --require-sidecars --require-critique 2>&1)"
+if echo "$va_out" | grep -q "required critique must pass"; then
+  _fail "(e) validator wrongly rejected a superseded historical fail"
+else
+  _pass "(e) validator tolerates a superseded historical fail member (no 'required critique must pass')"
+fi
+# Live fail (d's cross-reviewer bundle has a live disputed critique) must still be rejected.
+vd_out="$(flow_agents_node validate-workflow-artifacts "$D_AROOT" --require-sidecars --require-critique 2>&1)"
+if echo "$vd_out" | grep -q "required critique must pass"; then
+  _pass "(e) validator still rejects a LIVE (non-superseded) fail critique"
+else
+  _fail "(e) validator failed to reject a live fail critique"
+fi
+
+# ─── Summary ──────────────────────────────────────────────────────────────────
+echo ""
+echo "────────────────────────────────────────────"
+if [[ $errors -eq 0 ]]; then
+  echo "test_critique_supersession_roundtrip: all checks passed."
+  exit 0
+else
+  echo "test_critique_supersession_roundtrip: $errors check(s) failed."
+  exit 1
+fi

package/evals/integration/test_dual_emit_flow_step.sh

@@ -103,6 +103,151 @@ else
   _fail "resolver: CJS require failed"
 fi
 
+# Test 6: composed Builder closeout step resolves through builder.publish-learn
+if node --input-type=module << NODEEOF
+import { resolveFlowStep } from '${_RESOLVER_MOD}';
+const r = resolveFlowStep('builder.build', 'pr-open', '${ROOT}');
+if (!r) throw new Error('expected non-null result for builder.build/pr-open');
+if (r.flowId !== 'builder.build') throw new Error('expected active flowId builder.build, got ' + r.flowId);
+if (r.sourceFlowId !== 'builder.publish-learn') throw new Error('expected sourceFlowId builder.publish-learn, got ' + r.sourceFlowId);
+if (r.gateId !== 'builder.publish-learn:pr-open-gate') throw new Error('expected composed gate id, got ' + r.gateId);
+const claim = r.gateExpects.find(e => e.bundle_claim.claimType === 'builder.pr-open.pull-request');
+if (!claim) throw new Error('expected builder.pr-open.pull-request in composed expects');
+NODEEOF
+then
+  _pass "resolver: builder.build/pr-open resolves composed builder.publish-learn gate"
+else
+  _fail "resolver: builder.build/pr-open composed gate failed"
+fi
+
+# Test 7: composed child gates must export every imported expectation
+COMPOSE_DEFS="$TMP/compose-defs"
+mkdir -p "$COMPOSE_DEFS"
+cat > "$COMPOSE_DEFS/test.parent.flow.json" <<'JSON'
+{
+  "id": "test.parent",
+  "version": "1.0.0",
+  "steps": [{ "id": "compose", "next": null, "uses_flow": "test.child" }],
+  "gates": {}
+}
+JSON
+cat > "$COMPOSE_DEFS/test.child.flow.json" <<'JSON'
+{
+  "id": "test.child",
+  "version": "1.0.0",
+  "steps": [{ "id": "compose", "next": null }],
+  "exports": ["test.child.allowed"],
+  "gates": {
+    "compose-gate": {
+      "step": "compose",
+      "expects": [
+        {
+          "id": "allowed",
+          "kind": "trust.bundle",
+          "required": true,
+          "bundle_claim": {
+            "claimType": "test.child.allowed",
+            "subjectType": "flow-step",
+            "accepted_statuses": ["trusted", "accepted"]
+          }
+        }
+      ]
+    }
+  }
+}
+JSON
+
+if FLOW_AGENTS_FLOW_DEFS_DIR="$COMPOSE_DEFS" node --input-type=module << NODEEOF
+import { resolveFlowStep } from '${_RESOLVER_MOD}';
+const r = resolveFlowStep('test.parent', 'compose', '${ROOT}');
+if (!r) throw new Error('expected exported composition to resolve');
+if (r.sourceFlowId !== 'test.child') throw new Error('expected sourceFlowId test.child, got ' + r.sourceFlowId);
+if (!r.gateExpects.some(e => e.bundle_claim.claimType === 'test.child.allowed')) throw new Error('missing exported claim');
+NODEEOF
+then
+  _pass "resolver: composed child exports allow imported gate expectations"
+else
+  _fail "resolver: exported composed child claim failed"
+fi
+
+cat > "$COMPOSE_DEFS/test.child.flow.json" <<'JSON'
+{
+  "id": "test.child",
+  "version": "1.0.0",
+  "steps": [{ "id": "compose", "next": null }],
+  "exports": ["test.child.other"],
+  "gates": {
+    "compose-gate": {
+      "step": "compose",
+      "expects": [
+        {
+          "id": "allowed",
+          "kind": "trust.bundle",
+          "required": true,
+          "bundle_claim": {
+            "claimType": "test.child.allowed",
+            "subjectType": "flow-step",
+            "accepted_statuses": ["trusted", "accepted"]
+          }
+        }
+      ]
+    }
+  }
+}
+JSON
+
+if FLOW_AGENTS_FLOW_DEFS_DIR="$COMPOSE_DEFS" node --input-type=module << NODEEOF
+import { resolveFlowStep } from '${_RESOLVER_MOD}';
+const r = resolveFlowStep('test.parent', 'compose', '${ROOT}');
+if (r !== null) throw new Error('expected non-exported composition to fail closed');
+NODEEOF
+then
+  _pass "resolver: non-exported composed child claims fail closed"
+else
+  _fail "resolver: non-exported composed child claim was imported"
+fi
+
+cat > "$COMPOSE_DEFS/test.child.flow.json" <<'JSON'
+{
+  "id": "test.child",
+  "version": "1.0.0",
+  "steps": [{ "id": "compose", "next": null, "uses_flow": "test.child" }],
+  "exports": ["test.child.allowed"],
+  "gates": {}
+}
+JSON
+
+if FLOW_AGENTS_FLOW_DEFS_DIR="$COMPOSE_DEFS" node --input-type=module << NODEEOF
+import { resolveFlowStep } from '${_RESOLVER_MOD}';
+const r = resolveFlowStep('test.parent', 'compose', '${ROOT}');
+if (r !== null) throw new Error('expected self-cycle composition to return null');
+NODEEOF
+then
+  _pass "resolver: composed flow self-cycle returns null"
+else
+  _fail "resolver: composed flow self-cycle did not fail closed"
+fi
+
+cat > "$COMPOSE_DEFS/test.parent.flow.json" <<'JSON'
+{
+  "id": "test.parent",
+  "version": "1.0.0",
+  "steps": [{ "id": "compose", "next": null, "uses_flow": "../outside" }],
+  "gates": {}
+}
+JSON
+
+if FLOW_AGENTS_FLOW_DEFS_DIR="$COMPOSE_DEFS" node --input-type=module << NODEEOF
+import { resolveFlowStep } from '${_RESOLVER_MOD}';
+const r = resolveFlowStep('test.parent', 'compose', '${ROOT}');
+if (r !== null) throw new Error('expected invalid uses_flow id to return null');
+NODEEOF
+then
+  _pass "resolver: invalid composed uses_flow id returns null"
+else
+  _fail "resolver: invalid composed uses_flow id did not fail closed"
+fi
+
 echo ""
 echo "── P-d declared-only: session WITH active_flow_id=builder.build / active_step_id=verify ──"
 

package/evals/integration/test_enforcer_expects_driven.sh

@@ -34,11 +34,11 @@ trap cleanup EXIT
 # ─── helper: seed a minimal delivered workflow artifact ───────────────────────
 seed_repo() { # $1=dir $2=slug
   local p="$1" slug="$2"
-  mkdir -p "$p/.flow-agents/$slug"
+  mkdir -p "$p/.kontourai/flow-agents/$slug"
   printf '# Repo\n' > "$p/AGENTS.md"
   printf '%s' "{\"schema_version\":\"1.0\",\"task_slug\":\"$slug\",\"status\":\"delivered\",\"phase\":\"done\",\"updated_at\":\"2026-06-26T00:00:00Z\",\"next_action\":{\"status\":\"done\",\"summary\":\"done\"}}" \
-    > "$p/.flow-agents/$slug/state.json"
-  cat > "$p/.flow-agents/$slug/$slug--deliver.md" << MD
+    > "$p/.kontourai/flow-agents/$slug/state.json"
+  cat > "$p/.kontourai/flow-agents/$slug/$slug--deliver.md" << MD
 # $slug
 
 branch: main
@@ -67,12 +67,12 @@ seed_repo "$T1_DIR" "declares-tampered"
 
 # current.json: active flow
 printf '%s' '{"artifact_dir":"declares-tampered","active_flow_id":"builder.build","active_step_id":"verify"}' \
-  > "$T1_DIR/.flow-agents/current.json"
+  > "$T1_DIR/.kontourai/flow-agents/current.json"
 
-python3 - "$T1_DIR/.flow-agents/declares-tampered/trust.bundle" << 'PY'
+python3 - "$T1_DIR/.kontourai/flow-agents/declares-tampered/trust.bundle" << 'PY'
 import json, sys
 bundle = {
-    "schemaVersion": 3,
+    "schemaVersion": 5,
     "source": "flow-agents/workflow-sidecar",
     "claims": [{
         "id": "c1",
@@ -152,12 +152,12 @@ T2_DIR="$TMP/t2"
 seed_repo "$T2_DIR" "declares-clean"
 
 printf '%s' '{"artifact_dir":"declares-clean","active_flow_id":"builder.build","active_step_id":"verify"}' \
-  > "$T2_DIR/.flow-agents/current.json"
+  > "$T2_DIR/.kontourai/flow-agents/current.json"
 
-python3 - "$T2_DIR/.flow-agents/declares-clean/trust.bundle" << 'PY'
+python3 - "$T2_DIR/.kontourai/flow-agents/declares-clean/trust.bundle" << 'PY'
 import json, sys
 bundle = {
-    "schemaVersion": 3,
+    "schemaVersion": 5,
     "source": "flow-agents/workflow-sidecar",
     "claims": [{
         "id": "c2",
@@ -227,12 +227,12 @@ seed_repo "$T3_DIR" "no-flow"
 
 # No current.json flow keys (empty current.json that is still valid)
 printf '%s' '{"artifact_dir":"no-flow"}' \
-  > "$T3_DIR/.flow-agents/current.json"
+  > "$T3_DIR/.kontourai/flow-agents/current.json"
 
-python3 - "$T3_DIR/.flow-agents/no-flow/trust.bundle" << 'PY'
+python3 - "$T3_DIR/.kontourai/flow-agents/no-flow/trust.bundle" << 'PY'
 import json, sys
 bundle = {
-    "schemaVersion": 3,
+    "schemaVersion": 5,
     "source": "flow-agents/workflow-sidecar",
     "claims": [{
         "id": "c3",

package/evals/integration/test_evidence_capture_hook.sh

@@ -2,7 +2,7 @@
 # test_evidence_capture_hook.sh — Capture-first evidence determinism contracts.
 #
 # Part A: evidence-capture.js deterministically records command executions to
-#         .flow-agents/<slug>/command-log.jsonl (machine-recorded, not model-claimed).
+#         .kontourai/flow-agents/<slug>/command-log.jsonl (machine-recorded, not model-claimed).
 # Part B: stop-goal-fit.js cross-references evidence.json claimed-pass command
 #         checks against the capture log, and re-runs a TRUSTED backstop command
 #         only when the log has no execution for a claimed-pass command.
@@ -23,10 +23,10 @@ _fail() { echo "  ✗ $1"; errors=$((errors + 1)); }
 # ---- helpers -------------------------------------------------------------
 seed_repo() { # $1 dir, $2 slug
   local p="$1" slug="$2"
-  mkdir -p "$p/.flow-agents/$slug"
+  mkdir -p "$p/.kontourai/flow-agents/$slug"
   printf '# Repo\n' > "$p/AGENTS.md"
-  printf '%s' "{\"schema_version\":\"1.0\",\"task_slug\":\"$slug\",\"status\":\"delivered\",\"phase\":\"done\",\"updated_at\":\"2026-06-23T00:00:00Z\",\"next_action\":{\"status\":\"done\",\"summary\":\"done\"}}" > "$p/.flow-agents/$slug/state.json"
-  cat > "$p/.flow-agents/$slug/$slug--deliver.md" <<MD
+  printf '%s' "{\"schema_version\":\"1.0\",\"task_slug\":\"$slug\",\"status\":\"delivered\",\"phase\":\"done\",\"updated_at\":\"2026-06-23T00:00:00Z\",\"next_action\":{\"status\":\"done\",\"summary\":\"done\"}}" > "$p/.kontourai/flow-agents/$slug/state.json"
+  cat > "$p/.kontourai/flow-agents/$slug/$slug--deliver.md" <<MD
 # $slug
 
 branch: main
@@ -59,7 +59,7 @@ printf '{"hook_event_name":"PostToolUse","tool_name":"Bash","cwd":"%s","tool_inp
 # A non-command tool (Write) must NOT be captured.
 printf '{"hook_event_name":"PostToolUse","tool_name":"Write","cwd":"%s","tool_input":{"file_path":"/tmp/x"}}' "$A" | capture
 
-LOG="$A/.flow-agents/t1/command-log.jsonl"
+LOG="$A/.kontourai/flow-agents/t1/command-log.jsonl"
 if [[ -f "$LOG" ]]; then _pass "capture writes command-log.jsonl"; else _fail "capture did not write command-log.jsonl"; fi
 
 lines=$(wc -l < "$LOG" | tr -d ' ')
@@ -90,8 +90,8 @@ else _fail "capture should be non-blocking and echo stdin"; fi
 # ============================================================================
 echo "Part B1: log contradicts claimed pass → block"
 B="$TMP/contradict"; seed_repo "$B" t1
-printf '%s' '{"schema_version":"1.0","task_slug":"t1","verdict":"pass","checks":[{"id":"unit-tests","kind":"command","status":"pass","command":"npm test","summary":"tests passed"}]}' > "$B/.flow-agents/t1/evidence.json"
-printf '%s\n' '{"command":"npm test","observedResult":"fail","exitCode":1,"capturedAt":"2026-06-23T00:00:00Z","source":"postToolUse-capture"}' > "$B/.flow-agents/t1/command-log.jsonl"
+printf '%s' '{"schema_version":"1.0","task_slug":"t1","verdict":"pass","checks":[{"id":"unit-tests","kind":"command","status":"pass","command":"npm test","summary":"tests passed"}]}' > "$B/.kontourai/flow-agents/t1/evidence.json"
+printf '%s\n' '{"command":"npm test","observedResult":"fail","exitCode":1,"capturedAt":"2026-06-23T00:00:00Z","source":"postToolUse-capture"}' > "$B/.kontourai/flow-agents/t1/command-log.jsonl"
 
 if FLOW_AGENTS_GOAL_FIT_MODE=block FLOW_AGENTS_GOAL_FIT_BACKSTOP=skip node "$GATE" >/dev/null 2>"$TMP/b1.err" <<JSON
 {"hook_event_name":"Stop","cwd":"$B"}
@@ -109,8 +109,8 @@ fi
 # ============================================================================
 echo "Part B2: log confirms claimed pass → satisfied, no re-run"
 C="$TMP/confirm"; seed_repo "$C" t1
-printf '%s' '{"schema_version":"1.0","task_slug":"t1","verdict":"pass","checks":[{"id":"unit-tests","kind":"command","status":"pass","command":"npm test","summary":"tests passed"}]}' > "$C/.flow-agents/t1/evidence.json"
-printf '%s\n' '{"command":"npm test","observedResult":"pass","exitCode":0,"capturedAt":"2026-06-23T00:00:00Z","source":"postToolUse-capture"}' > "$C/.flow-agents/t1/command-log.jsonl"
+printf '%s' '{"schema_version":"1.0","task_slug":"t1","verdict":"pass","checks":[{"id":"unit-tests","kind":"command","status":"pass","command":"npm test","summary":"tests passed"}]}' > "$C/.kontourai/flow-agents/t1/evidence.json"
+printf '%s\n' '{"command":"npm test","observedResult":"pass","exitCode":0,"capturedAt":"2026-06-23T00:00:00Z","source":"postToolUse-capture"}' > "$C/.kontourai/flow-agents/t1/command-log.jsonl"
 # A poisoned npm on PATH proves the gate does NOT re-run when the log confirms.
 POISON="$TMP/poison"; mkdir -p "$POISON"
 printf '#!/usr/bin/env bash\necho "npm should not run" >&2\nexit 99\n' > "$POISON/npm"; chmod +x "$POISON/npm"
@@ -127,7 +127,7 @@ else _pass "gate trusts the log on a confirmed pass and does not re-run the back
 echo "Part B3: never-captured claim → trusted manifest backstop catches a fail"
 D="$TMP/backstop"; seed_repo "$D" t1
 printf '%s' '{"name":"x","scripts":{"test":"exit 7"}}' > "$D/package.json"
-printf '%s' '{"schema_version":"1.0","task_slug":"t1","verdict":"pass","checks":[{"id":"unit-tests","kind":"command","status":"pass","command":"npm test","summary":"tests passed"}]}' > "$D/.flow-agents/t1/evidence.json"
+printf '%s' '{"schema_version":"1.0","task_slug":"t1","verdict":"pass","checks":[{"id":"unit-tests","kind":"command","status":"pass","command":"npm test","summary":"tests passed"}]}' > "$D/.kontourai/flow-agents/t1/evidence.json"
 # command-log.jsonl intentionally absent — the command was never actually run.
 
 if FLOW_AGENTS_GOAL_FIT_MODE=block node "$GATE" >/dev/null 2>"$TMP/b3.err" <<JSON
@@ -146,9 +146,10 @@ fi
 # ============================================================================
 echo "Part B4: never-captured claim, nothing trusted resolves → NOT_VERIFIED"
 E="$TMP/notverified"; seed_repo "$E" t1
-printf '%s' '{"schema_version":"1.0","task_slug":"t1","verdict":"pass","checks":[{"id":"custom","kind":"command","status":"pass","command":"./my-thing.sh","summary":"ran custom"}]}' > "$E/.flow-agents/t1/evidence.json"
+printf '%s' '{"schema_version":"1.0","task_slug":"t1","status":"in_progress","phase":"verification","updated_at":"2026-06-23T00:00:00Z","next_action":{"status":"continue","summary":"verify command evidence"}}' > "$E/.kontourai/flow-agents/t1/state.json"
+printf '%s' '{"schema_version":"1.0","task_slug":"t1","verdict":"pass","checks":[{"id":"custom","kind":"command","status":"pass","command":"./my-thing.sh","summary":"ran custom"}]}' > "$E/.kontourai/flow-agents/t1/evidence.json"
 
-if FLOW_AGENTS_GOAL_FIT_MODE=block node "$GATE" >/dev/null 2>"$TMP/b4.err" <<JSON
+if FLOW_AGENTS_GOAL_FIT_MODE=block FLOW_AGENTS_GOAL_FIT_RECHECK=false node "$GATE" >/dev/null 2>"$TMP/b4.err" <<JSON
 {"hook_event_name":"Stop","cwd":"$E"}
 JSON
 then _fail "gate should not silently pass an un-captured, un-verifiable claimed-pass command"
@@ -164,7 +165,7 @@ fi
 # ============================================================================
 echo "Part B5: free-form model command re-run is opt-in only"
 F="$TMP/recheck"; seed_repo "$F" t1
-printf '%s' '{"schema_version":"1.0","task_slug":"t1","verdict":"pass","checks":[{"id":"custom","kind":"command","status":"pass","command":"exit 5","summary":"ran custom"}]}' > "$F/.flow-agents/t1/evidence.json"
+printf '%s' '{"schema_version":"1.0","task_slug":"t1","verdict":"pass","checks":[{"id":"custom","kind":"command","status":"pass","command":"exit 5","summary":"ran custom"}]}' > "$F/.kontourai/flow-agents/t1/evidence.json"
 # Opt-in ON: the model's free-form "exit 5" is re-run and fails → block.
 if FLOW_AGENTS_GOAL_FIT_MODE=block FLOW_AGENTS_GOAL_FIT_RECHECK=true node "$GATE" >/dev/null 2>"$TMP/b5.err" <<JSON
 {"hook_event_name":"Stop","cwd":"$F"}

package/evals/integration/test_fixture_retirement_audit.sh

@@ -21,9 +21,9 @@ 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")" == "11" ]] && pass "audit scans all fixture groups" || fail "audit scans all fixture groups"
+[[ "$(json_query "$TMPDIR_EVAL/audit.json" "totals.scanned")" == "15" ]] && 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")" == "11" ]] && pass "audit keeps all owned fixture groups" || fail "audit keeps all owned fixture groups"
+[[ "$(json_query "$TMPDIR_EVAL/audit.json" "totals.kept")" == "15" ]] && 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");