loki-mode 7.81.0 → 7.82.0

package/autonomy/run.sh

@@ -3979,9 +3979,18 @@ EOF
     if [ -n "$BUDGET_LIMIT" ]; then
         # Validate budget limit is numeric before writing JSON
         if ! echo "$BUDGET_LIMIT" | grep -qE '^[0-9]+(\.[0-9]+)?$'; then
-            log_warn "Invalid BUDGET_LIMIT '$BUDGET_LIMIT', defaulting to 0"
-            BUDGET_LIMIT=0
+            log_warn "Invalid BUDGET_LIMIT '$BUDGET_LIMIT', ignoring (no cap set)"
+            BUDGET_LIMIT=""
         fi
+        # Mirror the CLI guard: a non-positive cap (0/0.00) would make
+        # check_budget_limit pause before any work runs. Treat it as "no cap"
+        # rather than a silent pre-work pause.
+        if [ -n "$BUDGET_LIMIT" ] && ! awk -v b="$BUDGET_LIMIT" 'BEGIN{exit !(b+0 > 0)}'; then
+            log_warn "BUDGET_LIMIT '$BUDGET_LIMIT' is not greater than 0, ignoring (no cap set)"
+            BUDGET_LIMIT=""
+        fi
+    fi
+    if [ -n "$BUDGET_LIMIT" ]; then
         cat > ".loki/metrics/budget.json" << BUDGET_EOF
 {
   "limit": $BUDGET_LIMIT,
@@ -10386,20 +10395,32 @@ check_primary_recovery() {
 # Returns: 0 if rate limit detected, 1 otherwise
 is_rate_limited() {
     local log_file="$1"
-
-    # Generic patterns that work across all providers
-    # - HTTP 429 status code
-    # - "rate limit" / "rate-limit" / "ratelimit" text
-    # - "too many requests" text
-    # - "quota exceeded" text
-    # - "request limit" text
-    # - "retry after" / "retry-after" headers
-    if grep -qiE '(429|rate.?limit|too many requests|quota exceeded|request limit|retry.?after)' "$log_file" 2>/dev/null; then
+    [ -f "$log_file" ] || return 1
+
+    # Only consider the TAIL of the log: a real provider rate-limit appears at the
+    # END of the iteration (the call that failed), not buried in mid-run prose.
+    # Scanning the whole file false-positived on the agent's OWN output (a build
+    # that printed or generated rate-limiting code -- "rate limit", "429",
+    # "retry-after" as source/text -- wrongly triggered a multi-minute wait).
+    local tail_txt
+    tail_txt=$(tail -n 40 "$log_file" 2>/dev/null) || return 1
+
+    # Require an ERROR CONTEXT, not a bare keyword: a rate-limit token must
+    # co-occur (same line) with a genuine provider-error frame -- an explicit
+    # error word ("Error"/"failed"/"exceeded"), an HTTP/status frame, or the
+    # canonical "429 Too Many Requests" phrasing. This distinguishes a real
+    # failing API line from the words "rate limit"/"retry-after"/"429" appearing
+    # in the model's own generated source or prose. Note: a bare "429" or a bare
+    # "retry-after" is NOT sufficient on its own (both occur in generated code).
+    local _err='(error|errored|failed|exceeded|http[ /]?[0-9]|status[: ]+[0-9]|too many requests)'
+    local _rl='(rate.?limit|too many requests|quota exceeded|request limit|429[ )"]*too many|retry.?after)'
+    if printf '%s\n' "$tail_txt" | grep -qiE "(${_rl}).*(${_err})|(${_err}).*(${_rl})" 2>/dev/null; then
         return 0
     fi
 
-    # Claude-specific: "resets Xam/pm" format
-    if grep -qE 'resets [0-9]+[ap]m' "$log_file" 2>/dev/null; then
+    # Claude-specific: the explicit "resets Xam/pm" reset-time line is itself an
+    # unambiguous provider rate-limit signal (the CLI only prints it on a limit).
+    if printf '%s\n' "$tail_txt" | grep -qE 'resets [0-9]+[ap]m' 2>/dev/null; then
         return 0
     fi
 

package/completions/_loki

@@ -67,8 +67,12 @@ function _loki {
                     _arguments \
                         '--follow[Follow logs in real-time]' \
                         '-f[Follow logs in real-time]' \
-                        '--lines[Number of lines]:number:' \
-                        '-n[Number of lines]:number:'
+                        '--tail[Number of lines to show]:number:' \
+                        '-n[Number of lines to show]:number:' \
+                        '--all[Show all lines]' \
+                        '-a[Show all lines]' \
+                        '--help[Show help]' \
+                        '-h[Show help]'
                     ;;
                 issue)
                     _loki_issue
@@ -150,7 +154,63 @@ function _loki_commands {
         'share:Share session report as GitHub Gist'
         'proof:Inspect/share proof-of-run artifacts'
         'context:Context window management'
+        'ctx:Context window management (alias)'
         'code:Codebase intelligence'
+        'web:Start the web dashboard'
+        'plan:Preview the build plan for a spec'
+        'report:Reporting commands (cost, kpis, share)'
+        'cost:Show cost of recent runs'
+        'kpis:Key performance indicators'
+        'stats:Session statistics'
+        'preview:Open the running app preview'
+        'deploy:Advisory deploy command (print-only)'
+        'docker:Run Loki in a Docker container'
+        'mcp:MCP server commands'
+        'magic:Spec-driven component generation'
+        'assets:Export/import shareable team assets'
+        'spec:Living spec drift detection'
+        'verify:Deterministic PR verification'
+        'grill:Interrogate a spec before building'
+        'trust:Visible trust trajectory'
+        'trust-metrics:Trust metrics report'
+        'why:Explain the last build outcome'
+        'heal:Legacy system healing'
+        'modernize:Modernize a legacy system'
+        'migrate:Migration commands'
+        'analyze:Analyze a codebase'
+        'compliance:Compliance reporting'
+        'crash:Crash report management'
+        'otel:OpenTelemetry management'
+        'wiki:Wiki commands'
+        'explain:Explain Loki concepts'
+        'optimize:Optimization commands'
+        'review:Code review commands'
+        'export:Export session data'
+        'test:Run tests'
+        'ci:CI helpers'
+        'watch:Watch mode'
+        'audit:Audit commands'
+        'syslog:System log commands'
+        'cluster:Cluster commands'
+        'failover:Failover commands'
+        'remote:Remote execution commands'
+        'worktree:Git worktree commands'
+        'wt:Git worktree commands (alias)'
+        'trigger:Trigger commands'
+        'cleanup:Clean up Loki state'
+        'rollback:Roll back a change'
+        'update:Update Loki Mode'
+        'self-update:Update Loki Mode'
+        'setup-skill:Set up the Claude Code skill'
+        'quickstart:Quickstart guide'
+        'sentrux:Sentrux integration'
+        'state:Inspect run state'
+        'template:PRD template commands'
+        'ultracode:Ultra code review'
+        'voice:Voice commands'
+        'bench:Run benchmarks'
+        'open:Open a project resource'
+        'rc:Release candidate commands'
         'version:Show version'
         'completions:Output shell completions'
         'help:Show help'
@@ -160,7 +220,7 @@ function _loki_commands {
 
 function _loki_start {
     _arguments \
-        '--provider[AI provider]:provider name:(claude codex gemini)' \
+        '--provider[AI provider]:provider name:(claude codex cline aider)' \
         '--parallel[Parallel mode]' \
         '--background[Background mode]' \
         '--bg[Background mode]' \

package/completions/loki.bash

@@ -5,7 +5,7 @@ _loki_completion() {
     _init_completion || return
 
     # Main subcommands (must match autonomy/loki main case statement)
-    local main_commands="start quick monitor demo init stop pause resume status dashboard logs serve api sandbox notify import github issue config provider reset memory compound checkpoint council dogfood projects enterprise secrets doctor watchdog audit metrics syslog onboard share proof explain plan report test ci watch telemetry agent context code run export review optimize heal migrate cluster worktree trigger failover remote version completions help"
+    local main_commands="start quick monitor demo init stop pause resume status dashboard web serve api sandbox notify import github issue config provider reset memory compound checkpoint council dogfood projects enterprise secrets doctor watchdog audit metrics syslog onboard share proof explain plan report cost 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 [[ $cword -eq 1 ]]; then
@@ -18,7 +18,10 @@ _loki_completion() {
         start)
             # If the previous word was --provider, show provider names
             if [[ "$prev" == "--provider" ]]; then
-                COMPREPLY=( $(compgen -W "claude codex gemini" -- "$cur") )
+                # Active providers (loki rejects gemini since v7.5.18; cline/aider
+                # are supported). Keep in sync with providers/loader.sh
+                # SUPPORTED_PROVIDERS.
+                COMPREPLY=( $(compgen -W "claude codex cline aider" -- "$cur") )
                 return 0
             fi
 
@@ -119,7 +122,7 @@ _loki_completion() {
 
         logs)
             if [[ "$cur" == -* ]]; then
-                COMPREPLY=( $(compgen -W "--follow -f --lines -n --help" -- "$cur") )
+                COMPREPLY=( $(compgen -W "--tail -n --all -a --follow -f --help -h" -- "$cur") )
                 return 0
             fi
             ;;

package/dashboard/__init__.py

@@ -7,7 +7,7 @@ Modules:
     control: Session control API (start/stop/pause/resume)
 """
 
-__version__ = "7.81.0"
+__version__ = "7.82.0"
 
 # Expose the control app for easy import
 try:

package/dashboard/audit.py

@@ -21,6 +21,7 @@ import logging.handlers
 import os
 import socket
 import sys
+import threading
 from datetime import datetime, timezone
 from pathlib import Path
 from typing import Any, Optional
@@ -47,6 +48,16 @@ _SYSLOG_PROTO = os.environ.get("LOKI_AUDIT_SYSLOG_PROTO", "udp").lower().strip()
 INTEGRITY_ENABLED = os.environ.get("LOKI_AUDIT_NO_INTEGRITY", "").lower() not in ("true", "1", "yes")
 _last_hash: str = "0" * 64  # Genesis hash
 
+# Serializes the chain read-modify-write + file append in log_event(). Without
+# it, concurrent callers (the dashboard fans audit writes out across async
+# handlers / asyncio.to_thread workers) interleave the unsynchronized
+# _last_hash RMW with the file append: lines get written in a different order
+# than the hashes were chained, which breaks the tamper-evident chain so
+# verify_all_logs() reports valid:False even though no entry was tampered with.
+# Holding this lock makes "compute hash, update _last_hash, append the line" a
+# single atomic step so on-disk line order always matches chain order.
+_hash_lock = threading.Lock()
+
 
 def _recover_last_hash() -> str:
     """Recover the last integrity hash from the most recent audit log file.
@@ -254,20 +265,35 @@ def log_event(
         "details": details or {},
     }
 
-    # Tamper-evident chain hash
+    # Tamper-evident chain hash + file append, serialized as one atomic step.
+    #
+    # The chain hash is a read-modify-write of the module-global _last_hash, and
+    # the line must land in the file in the same order the hashes were chained.
+    # _hash_lock guards the whole "compute hash -> update _last_hash -> append"
+    # critical section so concurrent callers cannot interleave and break the
+    # chain (see _hash_lock definition above).
+    #
+    # NOTE for async callers: log_event() does blocking file I/O while holding
+    # this lock. When called from an asyncio handler it SHOULD be offloaded with
+    # `await asyncio.to_thread(audit.log_event, ...)` so a slow disk does not
+    # stall the dashboard event loop. The thread-safety here is what makes that
+    # offload safe; rewriting every call site to async is a separate, larger
+    # change and is intentionally not done here.
     global _last_hash
-    if INTEGRITY_ENABLED:
-        entry_json = json.dumps(entry, sort_keys=True, default=str)
-        entry["_integrity_hash"] = _compute_chain_hash(entry_json, _last_hash)
-        _last_hash = entry["_integrity_hash"]
+    with _hash_lock:
+        if INTEGRITY_ENABLED:
+            entry_json = json.dumps(entry, sort_keys=True, default=str)
+            entry["_integrity_hash"] = _compute_chain_hash(entry_json, _last_hash)
+            _last_hash = entry["_integrity_hash"]
 
-    log_file = _get_current_log_file()
-    _rotate_logs_if_needed(log_file)
+        log_file = _get_current_log_file()
+        _rotate_logs_if_needed(log_file)
 
-    with open(log_file, "a") as f:
-        f.write(json.dumps(entry) + "\n")
+        with open(log_file, "a") as f:
+            f.write(json.dumps(entry) + "\n")
 
-    # Forward to syslog if configured
+    # Forward to syslog if configured (outside the lock: fire-and-forget and
+    # must never extend the critical section / block other writers).
     _forward_to_syslog(entry)
 
     return entry

package/dashboard/registry.py

@@ -140,6 +140,9 @@ def register_project(
     path: str,
     name: Optional[str] = None,
     alias: Optional[str] = None,
+    pid: Optional[int] = None,
+    port: Optional[int] = None,
+    status: Optional[str] = None,
 ) -> dict:
     """
     Register a project in the registry.
@@ -148,6 +151,18 @@ def register_project(
         path: Absolute path to the project directory
         name: Display name (defaults to directory name)
         alias: Short alias for CLI usage
+        pid: Live runtime pid for the project's current build (optional)
+        port: Live runtime port for the project's current build (optional)
+        status: Runtime status override, e.g. "active" / "running" (optional)
+
+    The pid / port / status kwargs let a caller register a project AND set its
+    runtime fields in one atomic, locked load->mutate->save. Before this, the
+    runner had to call register_project() and then do a SECOND, unlocked
+    load-mutate-save to stamp pid/port/status, which lost-updates a concurrent
+    writer (two writers each load the registry, set their own runtime fields,
+    and save -- the last save wins and silently drops the other's entry). Set
+    these here so the whole change happens inside _registry_lock(). When all
+    three are omitted the behavior is unchanged (backward compatible).
 
     Returns:
         The registered project entry
@@ -188,6 +203,16 @@ def register_project(
             }
             registry["projects"][project_id] = project
 
+        # Atomically stamp runtime fields inside the same lock (both the create
+        # and update branches). None means "leave as-is" so omitting these
+        # keeps the legacy behavior.
+        if pid is not None:
+            project["pid"] = pid
+        if port is not None:
+            project["port"] = port
+        if status is not None:
+            project["status"] = status
+
         _save_registry(registry)
     return project
 
@@ -438,7 +463,6 @@ def sync_registry_with_discovery() -> dict:
     Returns:
         Summary of sync results
     """
-    registry = _load_registry()
     discovered = discover_projects()
 
     # Track results
@@ -446,31 +470,50 @@ def sync_registry_with_discovery() -> dict:
     updated = []
     missing = []
 
-    # Add/update discovered projects
-    discovered_paths = set()
-    for project_info in discovered:
-        path = project_info["path"]
-        discovered_paths.add(path)
-        project_id = _generate_project_id(path)
-
-        if project_id not in registry["projects"]:
-            # New project
-            project = register_project(path)
-            added.append(project)
-        else:
-            # Update existing
-            project = registry["projects"][project_id]
-            project["has_loki_dir"] = True
-            project["updated_at"] = datetime.now(timezone.utc).isoformat()
-            updated.append(project)
+    # Mutate a single registry copy under the lock and save it ONCE at the end.
+    # Previously this loaded the registry, then called register_project() in the
+    # loop (which loads+mutates+saves its OWN copy), and finally saved the stale
+    # original over the top -- silently dropping every newly added project (the
+    # summary said "added: N" while the on-disk registry kept zero of them). All
+    # mutations now happen on the same in-memory dict that gets persisted.
+    with _registry_lock():
+        registry = _load_registry()
 
-    # Check for missing projects
-    for project_id, project in registry["projects"].items():
-        if not os.path.isdir(project["path"]):
-            project["status"] = "missing"
-            missing.append(project)
+        # Add/update discovered projects
+        for project_info in discovered:
+            path = project_info["path"]
+            project_id = _generate_project_id(path)
+
+            if project_id not in registry["projects"]:
+                # New project
+                now = datetime.now(timezone.utc).isoformat()
+                project = {
+                    "id": project_id,
+                    "path": path,
+                    "name": os.path.basename(path),
+                    "alias": None,
+                    "registered_at": now,
+                    "updated_at": now,
+                    "last_accessed": None,
+                    "has_loki_dir": os.path.isdir(os.path.join(path, ".loki")),
+                    "status": "active",
+                }
+                registry["projects"][project_id] = project
+                added.append(project)
+            else:
+                # Update existing
+                project = registry["projects"][project_id]
+                project["has_loki_dir"] = True
+                project["updated_at"] = datetime.now(timezone.utc).isoformat()
+                updated.append(project)
 
-    _save_registry(registry)
+        # Check for missing projects
+        for project_id, project in registry["projects"].items():
+            if not os.path.isdir(project["path"]):
+                project["status"] = "missing"
+                missing.append(project)
+
+        _save_registry(registry)
 
     return {
         "added": len(added),