loki-mode 9.72.0 → 9.74.0

package/SKILL.md

@@ -3,7 +3,7 @@ name: loki-mode
 description: Autonomous spec-driven build system with a built-in trust layer. It does not call work done until it is verified (RARV-C closure loop, 8 quality gates, completion council, verified-completion evidence gate). Triggers on "Loki Mode". Takes a spec (PRD, GitHub issue, OpenAPI doc, etc.) to deployed product with minimal human intervention. Provider-agnostic. Requires --dangerously-skip-permissions flag.
 ---
 
-# Loki Mode v9.72.0
+# Loki Mode v9.74.0
 
 **You are an autonomous agent. You make decisions. You do not ask questions. You do not stop.**
 
@@ -470,4 +470,4 @@ See `CHANGELOG.md` entries [7.5.7], [7.5.8], [7.5.13] for the per-fix list and r
 
 ---
 
-**v9.72.0 | [Autonomi](https://www.autonomi.dev/) flagship product | ~410 lines core**
+**v9.74.0 | [Autonomi](https://www.autonomi.dev/) flagship product | ~410 lines core**

package/VERSION

@@ -1 +1 @@
-9.72.0
+9.74.0

package/autonomy/completion-council.sh

@@ -3742,11 +3742,19 @@ except (json.JSONDecodeError, IOError, KeyError, ValueError):
 runner = d.get('runner', 'none')
 passed = d.get('pass')
 status = d.get('status', '')
+# S-157 (BACKLOG 89): same failed-count rule as council_evidence_gate. A
+# numeric failed_count above zero is a failure whatever pass says; legacy
+# numeric failed is the fallback; null, missing or a bool is unmeasured.
+def _count(v):
+    return v if isinstance(v, (int, float)) and not isinstance(v, bool) else None
+failed_n = _count(d.get('failed_count'))
+if failed_n is None:
+    failed_n = _count(d.get('failed'))
 if status == 'no_tests_run':
     print('inconclusive:%s' % runner)
 elif runner == 'none':
     print('pass:none')
-elif passed is False:
+elif passed is False or (failed_n or 0) > 0:
     print('fail:%s' % runner)
 elif passed is not True:
     print('inconclusive:%s' % runner)
@@ -4602,9 +4610,18 @@ runner = d.get('runner', 'none')
 passed = d.get('pass')
 # Affirmative green requires a REAL suite (runner != none) that recorded a
 # boolean True pass. A missing/null/non-boolean pass key or a zero-test run
-# (status no_tests_run) is inconclusive, which is not green.
+# (status no_tests_run) is inconclusive, which is not green. S-157 (BACKLOG
+# 98): a recorded failure count above zero is never green, by the same rule
+# as council_evidence_gate (numeric failed_count wins, legacy numeric failed
+# is the fallback, null/missing/bool is unmeasured).
+def _count(v):
+    return v if isinstance(v, (int, float)) and not isinstance(v, bool) else None
+failed_n = _count(d.get('failed_count'))
+if failed_n is None:
+    failed_n = _count(d.get('failed'))
 print('yes' if (runner != 'none' and passed is True
-                and d.get('status') != 'no_tests_run') else 'no')
+                and d.get('status') != 'no_tests_run'
+                and (failed_n or 0) <= 0) else 'no')
 " 2>/dev/null || echo "no")
     [ "$tr_state" = "yes" ] || return 1
 

package/autonomy/lib/proof-generator.py

@@ -911,6 +911,7 @@ def _collect_tests(loki_dir):
         "status": "not_run",
         "duration_sec": None,
     }
+    empty = dict(out)
     if not isinstance(raw, dict):
         return out
     out["runner"] = str(raw.get("runner") or "")
@@ -944,6 +945,12 @@ def _collect_tests(loki_dir):
             out["status"] = "inconclusive"
     else:
         out["status"] = "not_run"
+    # BACKLOG 98: a test-results.json left by an earlier iteration is not this
+    # run's evidence, so it cannot lift the headline (the gates' rule). An
+    # earlier red is never discarded: a stale failure still reads failed.
+    if out["status"] != "failed" and not _test_results_fresh(
+            os.path.join(loki_dir, "quality")):
+        return empty
     return out
 
 

package/autonomy/loki

@@ -28444,16 +28444,23 @@ if budget_limit is not None:
 
 # Prefer the recorded cumulative spend; fall back to the current run only when
 # the state file carries no figure. Never silently substitute 0.0 for "unknown".
-if isinstance(budget_used_recorded, (int, float)):
+# A recorded ZERO is not a reading: run.sh check_budget_limit writes
+# "budget_used": 0 for a run it never measured. Same rule as the dashboard
+# (server.py budget readers): only a positive recorded figure counts, and a
+# measured zero comes from current_cost, which collect_efficiency leaves None
+# unless a record passes record_is_measured.
+if isinstance(budget_used_recorded, (int, float)) and budget_used_recorded > 0:
     budget_used = budget_used_recorded
 elif isinstance(current_cost, (int, float)):
     budget_used = current_cost
 else:
-    budget_used = 0.0
+    budget_used = None
 status = "none"
 percent_used = None
 remaining = None
-if budget_limit is not None and budget_limit > 0:
+if budget_limit is not None and budget_limit > 0 and budget_used is None:
+    status = "unknown"
+elif budget_limit is not None and budget_limit > 0:
     percent_used = round(budget_used / budget_limit * 100, 2)
     remaining = max(0.0, budget_limit - budget_used)
     if budget_used >= budget_limit:
@@ -28487,7 +28494,7 @@ if show_json:
         "by_model": {k: round(v, 6) for k, v in by_model.items()},
         "budget": {
             "limit": budget_limit,
-            "used": round(budget_used, 6),
+            "used": round(budget_used, 6) if budget_used is not None else None,
             "remaining": round(remaining, 6) if remaining is not None else None,
             "percent_used": percent_used,
             "status": status,
@@ -28538,13 +28545,17 @@ if budget_limit is None:
     print(DIM + "  When set, Loki warns at 80% and hard-stops at 100%." + NC)
 else:
     col = GREEN
-    if status == "warn":
+    if status in ("warn", "unknown"):
         col = YELLOW
     elif status == "exceeded":
         col = RED
     print("  Cap:       " + _fmt_usd(budget_limit))
-    print("  Used:      " + _fmt_usd(budget_used) + " (" + col + str(percent_used) + "%" + NC + ")")
-    print("  Remaining: " + _fmt_usd(remaining))
+    if budget_used is None:
+        print("  Used:      not recorded (no measured iteration cost)")
+        print("  Remaining: not recorded")
+    else:
+        print("  Used:      " + _fmt_usd(budget_used) + " (" + col + str(percent_used) + "%" + NC + ")")
+        print("  Remaining: " + _fmt_usd(remaining))
     print("  Status:    " + col + BOLD + status.upper() + NC)
     if status == "warn":
         print(YELLOW + "  Warning: at or above 80% of cap. Run continues; hard-stop at 100%." + NC)

package/autonomy/run.sh

@@ -13875,7 +13875,20 @@ sys.stdout.write(t.strip())
                    echo "$LOKI_MONOREPO_TEST_CMD" | grep -qE '[;|`$]|&&|\|\||>>|<<'; then
                     log_error "LOKI_MONOREPO_TEST_CMD rejected (only [A-Za-z0-9_./= -] allowed): $LOKI_MONOREPO_TEST_CMD"
                     test_runner="monorepo-custom-rejected"
-                    details="monorepo-custom: rejected by whitelist (gate skipped, inconclusive)"
+                    # BACKLOG 62: nothing ran, so record the no-runner shape
+                    # (see the runner=="none" return below) instead of falling
+                    # through with test_passed=true, which wrote pass:true and
+                    # touched unit-tests.pass. The runner label stays distinct:
+                    # the council reads a non-boolean pass on a named runner as
+                    # NO_PASS (inconclusive pass-through). The command itself is
+                    # not echoed into the JSON (it failed the whitelist).
+                    rm -f "$quality_dir/unit-tests.pass" 2>/dev/null || true
+                    cat > "$quality_dir/test-results.json" << TREOF
+{"timestamp":"$(date -u +%Y-%m-%dT%H:%M:%SZ)","runner":"monorepo-custom-rejected","pass":"inconclusive","summary":"LOKI_MONOREPO_TEST_CMD rejected by whitelist (not run)","command":null,"exit_code":null,"status":"not_run","passed_count":null,"failed_count":null,"verification_gap":"test_command_rejected"}
+TREOF
+                    printf '%s\n' "${ITERATION_COUNT:-0}" > "$quality_dir/.test-results.iter" 2>/dev/null || true
+                    _LOKI_TEST_SUITE_STATUS=not_run
+                    return 0
                 else
                     test_runner="monorepo-custom"
                     local output

package/dashboard/__init__.py

@@ -7,7 +7,7 @@ Modules:
     control: Session control API (start/stop/pause/resume)
 """
 
-__version__ = "9.72.0"
+__version__ = "9.74.0"
 
 # Expose the control app for easy import
 try:

package/dashboard/server.py

@@ -7852,6 +7852,8 @@ def _compute_cost_snapshot() -> dict:
     budget_remaining = None
     # Did ANY record carry an observed value? Not "was a file present".
     cost_recorded = False
+    # Tokens were measured (tracker fallback) but no USD figure was recorded.
+    cost_unknown = False
 
     # Read efficiency files (one JSON file per iteration/task).
     # Use the iteration-*.json pattern so this reader sees the same
@@ -7929,25 +7931,32 @@ def _compute_cost_snapshot() -> dict:
                 total_input = totals.get("total_input", 0)
                 total_output = totals.get("total_output", 0)
                 if total_input > 0 or total_output > 0:
-                    # Real observed tokens from the context tracker: this IS a
-                    # measurement, even if the recorded USD total happens to
-                    # be 0.
+                    # Real observed tokens from the context tracker: the TOKENS
+                    # are a measurement. The USD figure is only a measurement
+                    # when the tracker recorded a number; a missing one is
+                    # unknown (null), while a recorded 0.0 stays 0.0.
                     cost_recorded = True
-                    estimated_cost = totals.get("total_cost_usd", 0.0)
+                    usd = totals.get("total_cost_usd")
+                    if isinstance(usd, (int, float)) and not isinstance(usd, bool):
+                        estimated_cost = usd
+                    else:
+                        cost_unknown = True
+                    # No provider recorded -> the model is unknown, not sonnet.
+                    model = str(ctx.get("provider") or "unknown").lower()
                     # Rebuild by_model and by_phase from per_iteration data
                     for it in ctx.get("per_iteration", []):
                         inp = it.get("input_tokens", 0)
                         out = it.get("output_tokens", 0)
-                        cost = it.get("cost_usd", 0)
-                        model = ctx.get("provider", "sonnet").lower()
+                        cost = it.get("cost_usd")
+                        cost_is_num = isinstance(cost, (int, float)) and not isinstance(cost, bool)
                         if model not in by_model:
-                            by_model[model] = {"input_tokens": 0, "output_tokens": 0, "cost_usd": 0.0}
+                            by_model[model] = {"input_tokens": 0, "output_tokens": 0, "cost_usd": 0.0,
+                                               "measured": False, "tokens_measured": True}
                         by_model[model]["input_tokens"] += inp
                         by_model[model]["output_tokens"] += out
-                        by_model[model]["cost_usd"] += cost
-                        # Observed tokens from the tracker (cost_recorded above).
-                        by_model[model]["measured"] = True
-                        by_model[model]["tokens_measured"] = True
+                        if cost_is_num:
+                            by_model[model]["cost_usd"] += cost
+                            by_model[model]["measured"] = True
             except (json.JSONDecodeError, KeyError, TypeError, AttributeError):
                 pass
 
@@ -7964,7 +7973,7 @@ def _compute_cost_snapshot() -> dict:
             # Spend against the cap only when something was measured: with
             # nothing recorded, "used" is unknown, not $0.00, and "remaining"
             # is unknown, not the whole cap.
-            if budget_limit is not None and cost_recorded:
+            if budget_limit is not None and cost_recorded and not cost_unknown:
                 budget_used = estimated_cost
                 budget_remaining = max(0.0, budget_limit - budget_used)
         except (json.JSONDecodeError, KeyError):
@@ -7988,7 +7997,7 @@ def _compute_cost_snapshot() -> dict:
             total_input + total_output + total_cache_read + total_cache_creation
         ) if cost_recorded else None,
         "cache_hit_ratio": round(total_cache_read / _read_in, 4) if _read_in > 0 else None,
-        "estimated_cost_usd": round(estimated_cost, 6) if cost_recorded else None,
+        "estimated_cost_usd": round(estimated_cost, 6) if cost_recorded and not cost_unknown else None,
         "cost_recorded": cost_recorded,
         "by_phase": {k: {
             "input_tokens": v["input_tokens"] if v.get("tokens_measured") else None,