loki-mode 9.17.0 → 9.18.2

package/autonomy/verify.sh

@@ -421,6 +421,30 @@ _verify_zero_tests_executed() {
 }
 
 # ---------------------------------------------------------------------------
+# Reads scripts.test out of a package.json, or prints nothing.
+#
+# Parsed as JSON, never grepped: a substring search over the whole file is the
+# exact bug this helper exists to remove, and re-introducing it one level down
+# would be invisible. A malformed package.json prints nothing, which routes to
+# runner=none -> INCONCLUSIVE, never to a guessed runner.
+_verify_pkg_test_script() {
+    local tree="$1"
+    [ -f "$tree/package.json" ] || return 0
+    python3 -c '
+import json,sys
+try:
+    with open(sys.argv[1]) as fh:
+        d = json.load(fh)
+except Exception:
+    sys.exit(0)
+if not isinstance(d, dict):
+    sys.exit(0)
+s = d.get("scripts")
+if isinstance(s, dict) and isinstance(s.get("test"), str):
+    sys.stdout.write(s["test"])
+' "$tree/package.json" 2>/dev/null || true
+}
+
 # Gate: tests (faithful port of enforce_test_coverage detection, run.sh:6624).
 #
 # Detection order mirrors the source: vitest -> jest -> mocha (package.json),
@@ -451,16 +475,52 @@ verify_gate_tests() {
     local out=""
 
     if [ -f "$tree/package.json" ]; then
-        if grep -q '"vitest"' "$tree/package.json" 2>/dev/null; then
-            runner="vitest"
-            out="$(cd "$tree" && $_vt npx vitest run 2>&1)" || rc=$?
-        elif grep -q '"jest"' "$tree/package.json" 2>/dev/null; then
-            runner="jest"
-            out="$(cd "$tree" && $_vt npx jest --passWithNoTests --forceExit 2>&1)" || rc=$?
-        elif grep -q '"mocha"' "$tree/package.json" 2>/dev/null; then
-            runner="mocha"
-            out="$(cd "$tree" && $_vt npx mocha 2>&1)" || rc=$?
-        fi
+        # WHICH runner, decided by what `npm test` ACTUALLY INVOKES -- not by
+        # what happens to be installed.
+        #
+        # The old check was `grep '"jest"' package.json`, which matches a
+        # DEVDEPENDENCY entry. On this very repo that is exactly what happened:
+        # jest is a devDependency with NO jest config, while scripts.test runs
+        # bash -n plus `node --test`. So verify hijacked the runner, jest globbed
+        # 895 files that are not jest tests, every one reported "Your test suite
+        # must contain at least one test", and `loki verify` returned BLOCKED on
+        # a clean tree -- permanently, for a defect that does not exist.
+        #
+        # A false BLOCK on the flagship verification command is worse than a
+        # missed one: it trains users to ignore the verdict.
+        #
+        # scripts.test is the project's own declaration of its runner, so it is
+        # the signal with authority here. A declared script that names none of
+        # the three is still RUN, via `npm test` -- mirroring run.sh's npm-test
+        # fallback. Falling through instead would be its own defect: this repo
+        # declares `bash -n ... && node --test ...`, and without the fallback
+        # verify skipped it and ran PYTEST over a bash/node project.
+        local _test_script=""
+        _test_script="$(_verify_pkg_test_script "$tree")"
+        case "$_test_script" in
+            *vitest*)
+                runner="vitest"
+                out="$(cd "$tree" && $_vt npx vitest run 2>&1)" || rc=$? ;;
+            *jest*)
+                runner="jest"
+                out="$(cd "$tree" && $_vt npx jest --passWithNoTests --forceExit 2>&1)" || rc=$? ;;
+            *mocha*)
+                runner="mocha"
+                out="$(cd "$tree" && $_vt npx mocha 2>&1)" || rc=$? ;;
+            "")
+                : ;;   # nothing declared: leave runner=none for the paths below
+            *"no test specified"*)
+                : ;;   # npm's placeholder is not a test script
+            *)
+                # Labelled by what it invokes, so the evidence names the real
+                # runner rather than a generic "npm".
+                case "$_test_script" in
+                    *"node --test"*|*"node:test"*) runner="node-test" ;;
+                    *pytest*)                      runner="pytest" ;;
+                    *)                             runner="npm-test" ;;
+                esac
+                out="$(cd "$tree" && $_vt npm test 2>&1)" || rc=$? ;;
+        esac
     fi
 
     if [ "$runner" = "none" ]; then
@@ -1060,10 +1120,38 @@ print("%d %d %d %d" % (crit, high, mod, low))
 ' 2>/dev/null || echo "")"
                 if [ -n "$sev" ]; then
                     read -r _c _h _m _l <<<"$sev"
+                    # SHIPPED vs dev. The audit above covers ALL dependencies,
+                    # which is right for a gate -- a compromised build tool is a
+                    # real risk -- but the FINDING must say which of the two it
+                    # is. This repo reports 4 high CVEs while `npm audit
+                    # --omit=dev` reports ZERO: nothing a user installs is
+                    # vulnerable. A receipt that says "4 high severity
+                    # vulnerabilities" with no such qualifier reads as "the
+                    # shipped product is vulnerable", which is a materially
+                    # different and false claim.
+                    #
+                    # Measured separately rather than subtracted: the two runs
+                    # count different dependency graphs, so arithmetic on the
+                    # totals would not be sound.
+                    local _prod_hc=""
+                    _prod_hc="$(cd "$tree" && npm audit --omit=dev --json 2>/dev/null | python3 -c '
+import sys, json
+try:
+    v = json.load(sys.stdin).get("metadata", {}).get("vulnerabilities", {})
+except Exception:
+    sys.exit(0)
+print(v.get("critical", 0) + v.get("high", 0))
+' 2>/dev/null || echo "")"
+                    local _scope_note=""
+                    if [ "$_prod_hc" = "0" ]; then
+                        _scope_note=" All are in devDependencies; \`npm audit --omit=dev\` reports 0 high/critical, so nothing in the shipped dependency tree is affected."
+                    elif [ -n "$_prod_hc" ]; then
+                        _scope_note=" $_prod_hc of these are in the SHIPPED (non-dev) dependency tree."
+                    fi
                     if [ "$_c" -gt 0 ] || [ "$_h" -gt 0 ]; then
-                        _verify_add_gate "dependency_audit" "fail" "npm-audit" "$_c critical, $_h high CVEs" "true"
+                        _verify_add_gate "dependency_audit" "fail" "npm-audit" "$_c critical, $_h high CVEs (shipped high/critical: ${_prod_hc:-unmeasured})" "true"
                         _verify_add_finding "High" "dependencies" "deterministic:npm-audit" "package-lock.json" "null" \
-                            "npm audit found $_c critical and $_h high severity vulnerabilities."
+                            "npm audit found $_c critical and $_h high severity vulnerabilities.${_scope_note}"
                     elif [ "$_m" -gt 0 ]; then
                         _verify_add_gate "dependency_audit" "fail" "npm-audit" "$_m moderate CVEs" "true"
                         _verify_add_finding "Medium" "dependencies" "deterministic:npm-audit" "package-lock.json" "null" \

package/bin/loki

@@ -216,8 +216,14 @@ fi
 # we never disclose for an egress that will not happen. This mirrors the
 # cli_command sites below. No arguments are read, so nothing user-authored can
 # reach the payload.
+#
+# `quickstart` and `demo` are here because they are the two commands `loki
+# doctor` actually RECOMMENDS as a first build. Without them the funnel missed
+# every user who followed the product's own advice, which is precisely the
+# population whose drop-off this measures. Both are already in the
+# _loki_known_command allowlist, so `entry=` stays a bounded enum.
 case "${1:-}" in
-    start|run|quick)
+    start|run|quick|quickstart|demo)
         if command -v curl &>/dev/null && [ -f "$REPO_ROOT/autonomy/telemetry.sh" ]; then
             if ( SCRIPT_DIR="$REPO_ROOT/autonomy"; source "$SCRIPT_DIR/telemetry.sh" 2>/dev/null \
                  && declare -f _loki_analytics_enabled >/dev/null 2>&1 && _loki_analytics_enabled \

package/completions/_loki

@@ -120,7 +120,6 @@ function _loki {
 function _loki_commands {
     local -a commands
     commands=(
-        'gates:What blocks here vs only advises, and the promotion knob'
         'start:Start Loki Mode'
         'quick:Quick single-task mode'
         'quickstart:Guided first build from your idea'
@@ -173,8 +172,6 @@ function _loki_commands {
         'share:Share session report as GitHub Gist'
         'proof:Inspect/share proof-of-run artifacts'
         'outcomes:What happened to the work AFTER the receipt (reverted/reworked/survived)'
-        'verdict:The five measured trust signals in one readable block'
-        'readiness:Can an agent verify its own work in this repo? (measured, not LLM-scored)'
         'preview:Preview the locally-running app'
         'deploy:Deploy the built product (CI/CD-aware)'
         'context:Context window management'

package/completions/loki.bash

@@ -5,8 +5,8 @@ _loki_completion() {
     _init_completion || return
 
     # Main subcommands (must match autonomy/loki main case statement)
-    local main_commands="start quick monitor demo tour welcome init stop pause resume steer status next ship dashboard web serve api sandbox notify import github issue config provider reset memory compound checkpoint council dogfood projects enterprise secrets cockpit secure own handoff doctor watchdog audit metrics syslog onboard share proof receipt outcomes verdict readiness explain plan report cost estimate kpis stats test ci watch telemetry agent context ctx code run export review optimize heal modernize migrate cluster worktree wt trigger failover gates remote deploy docker mcp magic assets analyze compliance crash open otel preview quickstart rc rollback self-update sentrux setup-skill spec state template trust trust-metrics ultracode update verify voice why wiki bench cleanup logs grill docs cp version completions help"
-    local main_commands="start quick monitor demo tour welcome init stop pause resume steer status next ship dashboard web serve api sandbox notify import github issue intent config provider reset memory compound checkpoint council dogfood projects enterprise secrets cockpit secure own handoff doctor watchdog audit metrics syslog onboard share proof receipt explain plan report cost estimate kpis stats test ci watch telemetry agent context ctx code run export review optimize heal modernize migrate cluster worktree wt trigger failover gates remote deploy docker mcp magic assets analyze compliance crash open otel preview quickstart rc rollback self-update sentrux setup-skill spec state template trust trust-metrics ultracode update verify voice why wiki bench cleanup logs grill docs cp version completions help"
+    local main_commands="start quick monitor demo tour welcome init stop pause resume steer status next ship dashboard web serve api sandbox notify import github issue config provider reset memory compound checkpoint council dogfood projects enterprise secrets cockpit secure own handoff doctor watchdog audit metrics syslog onboard share proof receipt outcomes explain plan report cost estimate kpis stats test ci watch telemetry agent context ctx code run export review optimize heal modernize migrate cluster worktree wt trigger failover remote deploy docker mcp magic assets analyze compliance crash open otel preview quickstart rc rollback self-update sentrux setup-skill spec state template trust trust-metrics ultracode update verify voice why wiki bench cleanup logs grill docs cp version completions help"
+    local main_commands="start quick monitor demo tour welcome init stop pause resume steer status next ship dashboard web serve api sandbox notify import github issue intent config provider reset memory compound checkpoint council dogfood projects enterprise secrets cockpit secure own handoff doctor watchdog audit metrics syslog onboard share proof receipt explain plan report cost estimate kpis stats test ci watch telemetry agent context ctx code run export review optimize heal modernize migrate cluster worktree wt trigger failover remote deploy docker mcp magic assets analyze compliance crash open otel preview quickstart rc rollback self-update sentrux setup-skill spec state template trust trust-metrics ultracode update verify voice why wiki bench cleanup logs grill docs cp version completions help"
 
     # 1. If we are on the first argument (subcommand)
     if [[ $cword -eq 1 ]]; then

package/dashboard/__init__.py

@@ -7,7 +7,7 @@ Modules:
     control: Session control API (start/stop/pause/resume)
 """
 
-__version__ = "9.17.0"
+__version__ = "9.18.2"
 
 # Expose the control app for easy import
 try:

package/dashboard/server.py

@@ -8354,93 +8354,6 @@ async def get_trust_trajectory():
     return traj
 
 
-# =============================================================================
-# Gate policy API: which gates block here, and which only advise?
-# =============================================================================
-
-_GATE_POLICY_MODULE = None  # cached import of autonomy/lib/gate_policy.py
-
-
-def _load_gate_policy_module():
-    """Import the shared gate-policy reporter (single source of truth).
-
-    Same shape as _load_trust_module above: the derivation lives in
-    autonomy/lib/gate_policy.py so this endpoint, the bash `loki gates`, and the
-    test suite all agree, and it loads via importlib because autonomy/lib is not
-    an importable package. Cached after first load, None when unavailable.
-    """
-    global _GATE_POLICY_MODULE
-    if _GATE_POLICY_MODULE is not None:
-        return _GATE_POLICY_MODULE
-    repo_root = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))
-    mod_path = os.path.join(repo_root, "autonomy", "lib", "gate_policy.py")
-    if not os.path.isfile(mod_path):
-        return None
-    try:
-        import importlib.util as _ilu
-        spec = _ilu.spec_from_file_location("gate_policy", mod_path)
-        if spec is None or spec.loader is None:
-            return None
-        mod = _ilu.module_from_spec(spec)
-        spec.loader.exec_module(mod)
-        _GATE_POLICY_MODULE = mod
-        return mod
-    except Exception:
-        return None
-
-
-@app.get("/api/gate-policy", dependencies=[Depends(auth.require_scope("read"))])
-async def get_gate_policy():
-    """Which quality gates block on this repo, and which only advise.
-
-    Read-only by construction. There is deliberately no POST counterpart:
-    promoting an advisory gate to blocking stays an explicit operator act via
-    the environment variable named in each record's `promote_with`, because a
-    surface that can silently start blocking is the thing operators most
-    reasonably fear. This endpoint reports; it never promotes.
-
-    Honest-data rule, and the reason `audit_hits` is nullable: an unmeasured
-    gate reports null, NEVER 0. Zero is the positive claim that the gate ran and
-    never fired; an absent ledger is not evidence of that. See the shape
-    contract in the gate_policy module docstring.
-
-    Fails open like /api/trust/trajectory: a missing module returns a renderable
-    available=False payload rather than a 500. Unlike that sibling the assess()
-    call is also wrapped, because it reads a ledger file that an unrelated
-    writer can leave malformed, and a broken ledger must not take down a
-    read-only report.
-
-    Response keys are the module shape contract plus `available` (bool). On the
-    fail-open path `available` is False, `status` is "unavailable" (a value the
-    module contract does not itself define, since assess() never returned), an
-    `error` string explains why, and `gates` is empty. A UI should branch on
-    `available`, not on a non-empty `gates`.
-    """
-    mod = _load_gate_policy_module()
-    if mod is None:
-        return {
-            "schema_version": 1,
-            "available": False,
-            "error": "gate_policy module not found",
-            "status": "unavailable",
-            "ledger": "absent",
-            "gates": [],
-        }
-    try:
-        res = mod.assess(str(_get_loki_dir()))
-    except Exception as e:
-        return {
-            "schema_version": 1,
-            "available": False,
-            "error": f"gate policy assessment failed: {e}",
-            "status": "unavailable",
-            "ledger": "absent",
-            "gates": [],
-        }
-    res["available"] = True
-    return res
-
-
 # =============================================================================
 # Pricing API
 # =============================================================================
@@ -10270,71 +10183,6 @@ async def remove_checklist_waiver(item_id: str):
 # Council Hard Gate Endpoint (Phase 4)
 # =============================================================================
 
-# Receipts name gates in snake_case; the UI rows are display names. Mapping is
-# explicit rather than derived (a lower()/replace() heuristic would silently
-# mis-map "Test Suite" <-> "test_coverage" and "Test Mutation" <->
-# "mutation_integrity", which are different gates).
-_RECEIPT_GATE_NAMES = {
-    "static_analysis": "Static Analysis",
-    "test_coverage": "Test Suite",
-    "code_review": "Blind Code Review",
-    "anti_sycophancy": "Anti-Sycophancy",
-    "mock_integrity": "Mock Integrity",
-    "mutation_integrity": "Test Mutation",
-    "doc_coverage": "Documentation Coverage",
-    "magic_debate": "Magic Modules Debate",
-}
-
-
-# Receipts say "passed"/"failed"; the UI switches on "pass"/"fail"
-# (dashboard-ui/components/loki-quality-gates.js:51). Passing the receipt's word
-# through unchanged matched nothing, so every gate fell to the "pending" default
-# and the page still read "0 Pass, 0 Fail, 8 Pending" while showing real
-# timestamps -- half-fixed, and only visible by loading the page.
-#
-# An UNRECOGNISED status maps to nothing and the gate stays pending. Guessing
-# (e.g. treating any unknown word as a pass) is how a future status string would
-# silently become a green badge.
-_RECEIPT_GATE_STATUS = {
-    "passed": "pass", "pass": "pass", "ok": "pass",
-    "failed": "fail", "fail": "fail", "blocked": "fail",
-}
-
-
-def _receipt_backed_gates():
-    """_DEFAULT_QUALITY_GATES, with results filled in from the newest receipt.
-
-    Fail-open by design: any error returns the plain defaults, so a malformed
-    receipt degrades to "pending" rather than breaking the Quality page.
-
-    Gates absent from the receipt keep status "pending" and get NO last_checked.
-    An absent measurement is never rendered as a result.
-    """
-    gates = [dict(g) for g in _DEFAULT_QUALITY_GATES]
-    try:
-        proofs = sorted((_get_loki_dir() / "proofs").glob("*/proof.json"))
-        if not proofs:
-            return gates
-        receipt = json.loads(proofs[-1].read_text())
-        checked_at = receipt.get("generated_at") or ""
-        by_display = {}
-        for entry in receipt.get("quality_gates", {}).get("gates", []) or []:
-            display = _RECEIPT_GATE_NAMES.get(entry.get("name", ""))
-            status = _RECEIPT_GATE_STATUS.get(str(entry.get("status", "")).lower())
-            if display and status:
-                by_display[display] = status
-        for gate in gates:
-            status = by_display.get(gate.get("name"))
-            if status:
-                gate["status"] = status
-                if checked_at:
-                    gate["last_checked"] = checked_at
-                gate["source"] = "receipt"
-    except (OSError, ValueError, KeyError, TypeError, AttributeError):
-        return [dict(g) for g in _DEFAULT_QUALITY_GATES]
-    return gates
-
-
 _DEFAULT_QUALITY_GATES = [
     {"name": "Static Analysis", "description": "CodeQL, ESLint/Pylint, type-checker findings on the diff", "status": "pending"},
     {"name": "Test Suite", "description": "Project test runner pass/fail (red blocks)", "status": "pending"},
@@ -10373,22 +10221,7 @@ async def get_council_gate():
         except (json.JSONDecodeError, IOError):
             data = {"blocked": False, "gates": _DEFAULT_QUALITY_GATES, "error": "Failed to read gate file"}
     else:
-        # No live gate file. Fall back to the most recent RECEIPT, which already
-        # records per-gate results -- rather than serving eight rows that all
-        # say "Last checked: Never" while .loki/proofs/ holds the real answers.
-        #
-        # That "Never" was not a display bug: /api/council/gate served
-        # _DEFAULT_QUALITY_GATES verbatim, and those entries carry no
-        # last_checked, so the UI honestly rendered "Never" for every gate on a
-        # repo with 9 receipts.
-        #
-        # ONLY gates the receipt actually names are filled in. Measured across
-        # all 9 receipts here, that is 3 of 8 (static_analysis, code_review,
-        # doc_coverage); the other five stay "pending" with no timestamp,
-        # because a gate that never ran must not inherit a neighbour's result.
-        # Filling all eight from a 3-gate receipt is the exact false-green this
-        # codebase exists to prevent.
-        data = {"blocked": False, "gates": _receipt_backed_gates()}
+        data = {"blocked": False, "gates": _DEFAULT_QUALITY_GATES}
 
     # Verified-completion evidence gate (additive).
     if evidence_file.exists():