prizmkit 1.1.90 → 1.1.91
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/bundled/VERSION.json +3 -3
- package/bundled/dev-pipeline/lib/heartbeat.sh +56 -22
- package/bundled/dev-pipeline-windows/lib/common.ps1 +61 -13
- package/bundled/dev-pipeline-windows/lib/pipeline.ps1 +17 -5
- package/bundled/dev-pipeline-windows/run-recovery.ps1 +7 -3
- package/bundled/skills/_metadata.json +1 -1
- package/package.json +1 -1
package/bundled/VERSION.json
CHANGED
|
@@ -43,6 +43,8 @@ start_heartbeat() {
|
|
|
43
43
|
local prev_size=0
|
|
44
44
|
local prev_max_size=0
|
|
45
45
|
local prev_child_activity_signature=""
|
|
46
|
+
local prev_progress_signature=""
|
|
47
|
+
local prev_error_loop_signature=""
|
|
46
48
|
local stale_seconds=0
|
|
47
49
|
while kill -0 "$cli_pid" 2>/dev/null; do
|
|
48
50
|
sleep "$heartbeat_interval"
|
|
@@ -77,9 +79,10 @@ start_heartbeat() {
|
|
|
77
79
|
local child_activity_signature=""
|
|
78
80
|
local child_total_bytes=0
|
|
79
81
|
local child_session_count=0
|
|
82
|
+
local progress_signature=""
|
|
80
83
|
if [[ -f "$progress_json" ]]; then
|
|
81
|
-
local
|
|
82
|
-
|
|
84
|
+
local progress_activity_data
|
|
85
|
+
progress_activity_data=$(python3 - "$progress_json" <<'PY' 2>/dev/null || true
|
|
83
86
|
import json
|
|
84
87
|
import sys
|
|
85
88
|
|
|
@@ -89,14 +92,31 @@ try:
|
|
|
89
92
|
except Exception:
|
|
90
93
|
sys.exit(0)
|
|
91
94
|
|
|
92
|
-
|
|
95
|
+
child_signature = str(progress.get("child_activity_signature") or "")
|
|
93
96
|
total_bytes = int(progress.get("child_total_bytes") or 0)
|
|
94
97
|
session_count = len(progress.get("child_session_files") or [])
|
|
95
|
-
|
|
98
|
+
progress_signature = json.dumps(
|
|
99
|
+
[
|
|
100
|
+
progress.get("current_phase") or "",
|
|
101
|
+
progress.get("current_tool") or "",
|
|
102
|
+
int(progress.get("message_count") or 0),
|
|
103
|
+
int(progress.get("total_tool_calls") or 0),
|
|
104
|
+
int(progress.get("active_subagent_count") or 0),
|
|
105
|
+
int(progress.get("subagent_spawn_count") or 0),
|
|
106
|
+
child_signature,
|
|
107
|
+
total_bytes,
|
|
108
|
+
session_count,
|
|
109
|
+
progress.get("fatal_error_code") or "",
|
|
110
|
+
progress.get("terminal_result_text") or "",
|
|
111
|
+
],
|
|
112
|
+
separators=(",", ":"),
|
|
113
|
+
)
|
|
114
|
+
sep = "\x1f"
|
|
115
|
+
print(sep.join([child_signature, str(total_bytes), str(session_count), progress_signature]))
|
|
96
116
|
PY
|
|
97
117
|
)
|
|
98
|
-
if [[ -n "$
|
|
99
|
-
IFS=$'\
|
|
118
|
+
if [[ -n "$progress_activity_data" ]]; then
|
|
119
|
+
IFS=$'\037' read -r child_activity_signature child_total_bytes child_session_count progress_signature <<< "$progress_activity_data"
|
|
100
120
|
fi
|
|
101
121
|
fi
|
|
102
122
|
|
|
@@ -106,6 +126,12 @@ PY
|
|
|
106
126
|
fi
|
|
107
127
|
prev_child_activity_signature="$child_activity_signature"
|
|
108
128
|
|
|
129
|
+
local progress_advanced=0
|
|
130
|
+
if [[ -n "$progress_signature" && "$progress_signature" != "$prev_progress_signature" ]]; then
|
|
131
|
+
progress_advanced=1
|
|
132
|
+
fi
|
|
133
|
+
prev_progress_signature="$progress_signature"
|
|
134
|
+
|
|
109
135
|
local effective_stale_kill_threshold="$stale_kill_threshold"
|
|
110
136
|
if [[ $stale_kill_threshold -gt 0 && -f "$progress_json" ]]; then
|
|
111
137
|
local extended_threshold
|
|
@@ -172,8 +198,11 @@ PY
|
|
|
172
198
|
|
|
173
199
|
# Check for error-loop: agent is actively producing output but results are
|
|
174
200
|
# all read-offset errors or wasted calls. This is a stuck agent that appears
|
|
175
|
-
# "active" by log growth but is accomplishing nothing.
|
|
201
|
+
# "active" by log growth but is accomplishing nothing. Only a newly advancing
|
|
202
|
+
# error signature counts; stale historical errors must not turn later normal
|
|
203
|
+
# output into a stale kill.
|
|
176
204
|
local error_loop_detected=false
|
|
205
|
+
local error_loop_signature=""
|
|
177
206
|
if [[ $effective_stale_kill_threshold -gt 0 && $growth -gt 0 && -f "$progress_json" ]]; then
|
|
178
207
|
local error_loop_flag
|
|
179
208
|
error_loop_flag=$(python3 - "$progress_json" <<'PY' 2>/dev/null || true
|
|
@@ -187,33 +216,38 @@ errors = progress.get("errors", [])
|
|
|
187
216
|
if isinstance(errors, list) and len(errors) >= 5:
|
|
188
217
|
recent = errors[-5:]
|
|
189
218
|
if all(isinstance(e, dict) and e.get("type") in ("read_offset_overflow", "wasted_call") for e in recent):
|
|
190
|
-
print("
|
|
219
|
+
print(json.dumps(recent, sort_keys=True, separators=(",", ":")))
|
|
191
220
|
PY
|
|
192
221
|
)
|
|
193
|
-
if [[ "$error_loop_flag"
|
|
194
|
-
|
|
222
|
+
if [[ -n "$error_loop_flag" ]]; then
|
|
223
|
+
error_loop_signature="$error_loop_flag"
|
|
224
|
+
if [[ -n "$prev_error_loop_signature" && "$error_loop_signature" != "$prev_error_loop_signature" ]]; then
|
|
225
|
+
error_loop_detected=true
|
|
226
|
+
fi
|
|
195
227
|
fi
|
|
196
228
|
fi
|
|
229
|
+
prev_error_loop_signature="$error_loop_signature"
|
|
197
230
|
|
|
198
231
|
# Track progress staleness. Parent sessions can sit in a wait/polling
|
|
199
|
-
# tool while child transcripts
|
|
200
|
-
#
|
|
201
|
-
#
|
|
232
|
+
# tool while child transcripts or structured progress counters keep
|
|
233
|
+
# advancing, so both child activity and progress.json state changes
|
|
234
|
+
# count as progress. Error loops bypass normal growth-as-progress
|
|
235
|
+
# because the log is only growing with repeated failed reads or wasted
|
|
236
|
+
# calls.
|
|
202
237
|
#
|
|
203
238
|
# When the main session log has been truncated (e.g. by subagent stdout
|
|
204
|
-
# redirection severing the file descriptor), use
|
|
205
|
-
# primary progress
|
|
206
|
-
#
|
|
239
|
+
# redirection severing the file descriptor), use structured progress and
|
|
240
|
+
# child activity as the primary progress signals — the parent process is
|
|
241
|
+
# still running and children may still be producing output.
|
|
207
242
|
if [[ "$error_loop_detected" == "true" ]]; then
|
|
208
243
|
stale_seconds=$effective_stale_kill_threshold
|
|
209
244
|
elif [[ "$log_truncated" == "true" ]]; then
|
|
210
|
-
|
|
211
|
-
# Don't increment staleness — the process is still working.
|
|
212
|
-
if [[ $child_growth -gt 0 ]]; then
|
|
245
|
+
if [[ $progress_advanced -gt 0 || $child_growth -gt 0 ]]; then
|
|
213
246
|
stale_seconds=0
|
|
247
|
+
else
|
|
248
|
+
stale_seconds=$((stale_seconds + heartbeat_interval))
|
|
214
249
|
fi
|
|
215
|
-
|
|
216
|
-
elif [[ $growth -le 0 && $child_growth -eq 0 ]]; then
|
|
250
|
+
elif [[ $growth -le 0 && $child_growth -eq 0 && $progress_advanced -eq 0 ]]; then
|
|
217
251
|
stale_seconds=$((stale_seconds + heartbeat_interval))
|
|
218
252
|
else
|
|
219
253
|
stale_seconds=0
|
|
@@ -251,7 +285,7 @@ PY
|
|
|
251
285
|
local status_icon
|
|
252
286
|
if $log_truncated; then
|
|
253
287
|
status_icon="${RED}⚠${NC}"
|
|
254
|
-
elif [[ $growth -gt 0 || $child_growth -gt 0 ]]; then
|
|
288
|
+
elif [[ $growth -gt 0 || $child_growth -gt 0 || $progress_advanced -gt 0 ]]; then
|
|
255
289
|
status_icon="${GREEN}▶${NC}"
|
|
256
290
|
else
|
|
257
291
|
status_icon="${YELLOW}⏸${NC}"
|
|
@@ -467,13 +467,14 @@ function Get-PrizmEffectiveStaleKillThreshold {
|
|
|
467
467
|
return [Math]::Max($BaseThreshold * 4, 3600)
|
|
468
468
|
}
|
|
469
469
|
|
|
470
|
-
function Get-
|
|
470
|
+
function Get-PrizmProgressActivity {
|
|
471
471
|
param([string]$ProgressFile)
|
|
472
472
|
|
|
473
473
|
$empty = [pscustomobject]@{
|
|
474
|
-
|
|
475
|
-
|
|
476
|
-
|
|
474
|
+
ChildSignature = ''
|
|
475
|
+
ChildTotalBytes = 0
|
|
476
|
+
ChildSessionCount = 0
|
|
477
|
+
ProgressSignature = ''
|
|
477
478
|
}
|
|
478
479
|
if (-not (Test-Path $ProgressFile)) { return $empty }
|
|
479
480
|
|
|
@@ -483,25 +484,72 @@ function Get-PrizmProgressChildActivity {
|
|
|
483
484
|
return $empty
|
|
484
485
|
}
|
|
485
486
|
|
|
486
|
-
$
|
|
487
|
+
$childSignature = ''
|
|
487
488
|
if ($progress.PSObject.Properties['child_activity_signature'] -and $progress.child_activity_signature) {
|
|
488
|
-
$
|
|
489
|
+
$childSignature = [string]$progress.child_activity_signature
|
|
489
490
|
}
|
|
490
491
|
|
|
491
|
-
$
|
|
492
|
+
$childTotalBytes = [int64]0
|
|
492
493
|
if ($progress.PSObject.Properties['child_total_bytes']) {
|
|
493
|
-
[int64]::TryParse([string]$progress.child_total_bytes, [ref]$
|
|
494
|
+
[int64]::TryParse([string]$progress.child_total_bytes, [ref]$childTotalBytes) | Out-Null
|
|
494
495
|
}
|
|
495
496
|
|
|
496
|
-
$
|
|
497
|
+
$childSessionCount = 0
|
|
497
498
|
if ($progress.PSObject.Properties['child_session_files'] -and $progress.child_session_files) {
|
|
498
|
-
$
|
|
499
|
+
$childSessionCount = @($progress.child_session_files).Count
|
|
500
|
+
}
|
|
501
|
+
|
|
502
|
+
$messageCount = 0
|
|
503
|
+
if ($progress.PSObject.Properties['message_count']) {
|
|
504
|
+
[int]::TryParse([string]$progress.message_count, [ref]$messageCount) | Out-Null
|
|
505
|
+
}
|
|
506
|
+
$totalToolCalls = 0
|
|
507
|
+
if ($progress.PSObject.Properties['total_tool_calls']) {
|
|
508
|
+
[int]::TryParse([string]$progress.total_tool_calls, [ref]$totalToolCalls) | Out-Null
|
|
509
|
+
}
|
|
510
|
+
$activeSubagentCount = 0
|
|
511
|
+
if ($progress.PSObject.Properties['active_subagent_count']) {
|
|
512
|
+
[int]::TryParse([string]$progress.active_subagent_count, [ref]$activeSubagentCount) | Out-Null
|
|
513
|
+
}
|
|
514
|
+
$subagentSpawnCount = 0
|
|
515
|
+
if ($progress.PSObject.Properties['subagent_spawn_count']) {
|
|
516
|
+
[int]::TryParse([string]$progress.subagent_spawn_count, [ref]$subagentSpawnCount) | Out-Null
|
|
517
|
+
}
|
|
518
|
+
$currentPhase = if ($progress.PSObject.Properties['current_phase'] -and $progress.current_phase) { [string]$progress.current_phase } else { '' }
|
|
519
|
+
$currentTool = if ($progress.PSObject.Properties['current_tool'] -and $progress.current_tool) { [string]$progress.current_tool } else { '' }
|
|
520
|
+
$fatalErrorCode = if ($progress.PSObject.Properties['fatal_error_code'] -and $progress.fatal_error_code) { [string]$progress.fatal_error_code } else { '' }
|
|
521
|
+
$terminalResultText = if ($progress.PSObject.Properties['terminal_result_text'] -and $progress.terminal_result_text) { [string]$progress.terminal_result_text } else { '' }
|
|
522
|
+
|
|
523
|
+
$progressSignature = @(
|
|
524
|
+
$currentPhase,
|
|
525
|
+
$currentTool,
|
|
526
|
+
$messageCount,
|
|
527
|
+
$totalToolCalls,
|
|
528
|
+
$activeSubagentCount,
|
|
529
|
+
$subagentSpawnCount,
|
|
530
|
+
$childSignature,
|
|
531
|
+
$childTotalBytes,
|
|
532
|
+
$childSessionCount,
|
|
533
|
+
$fatalErrorCode,
|
|
534
|
+
$terminalResultText
|
|
535
|
+
) | ConvertTo-Json -Compress
|
|
536
|
+
|
|
537
|
+
return [pscustomobject]@{
|
|
538
|
+
ChildSignature = $childSignature
|
|
539
|
+
ChildTotalBytes = $childTotalBytes
|
|
540
|
+
ChildSessionCount = $childSessionCount
|
|
541
|
+
ProgressSignature = $progressSignature
|
|
499
542
|
}
|
|
543
|
+
}
|
|
544
|
+
|
|
545
|
+
function Get-PrizmProgressChildActivity {
|
|
546
|
+
param([string]$ProgressFile)
|
|
500
547
|
|
|
548
|
+
$activity = Get-PrizmProgressActivity -ProgressFile $ProgressFile
|
|
501
549
|
return [pscustomobject]@{
|
|
502
|
-
Signature = $
|
|
503
|
-
TotalBytes = $
|
|
504
|
-
SessionCount = $
|
|
550
|
+
Signature = $activity.ChildSignature
|
|
551
|
+
TotalBytes = $activity.ChildTotalBytes
|
|
552
|
+
SessionCount = $activity.ChildSessionCount
|
|
505
553
|
}
|
|
506
554
|
}
|
|
507
555
|
|
|
@@ -740,7 +740,9 @@ function Invoke-PrizmPipeline {
|
|
|
740
740
|
$elapsedSeconds = 0
|
|
741
741
|
$staleSeconds = 0
|
|
742
742
|
$previousLogSize = 0
|
|
743
|
+
$previousProgressSignature = ''
|
|
743
744
|
$previousChildActivitySignature = ''
|
|
745
|
+
$previousErrorLoopSignature = ''
|
|
744
746
|
$wasTimedOut = $false
|
|
745
747
|
$staleKillMarker = Join-Path $logsDir 'stale-kill.json'
|
|
746
748
|
$wasStaleKilled = $false
|
|
@@ -758,16 +760,22 @@ function Invoke-PrizmPipeline {
|
|
|
758
760
|
$growth = $currentLogSize - $previousLogSize
|
|
759
761
|
$previousLogSize = $currentLogSize
|
|
760
762
|
|
|
761
|
-
$
|
|
762
|
-
$
|
|
763
|
+
$progressActivity = Get-PrizmProgressActivity -ProgressFile $progressJson
|
|
764
|
+
$progressSignature = [string]$progressActivity.ProgressSignature
|
|
765
|
+
$progressAdvanced = ($progressSignature -and $progressSignature -ne $previousProgressSignature)
|
|
766
|
+
$previousProgressSignature = $progressSignature
|
|
767
|
+
$childSignature = [string]$progressActivity.ChildSignature
|
|
763
768
|
$childAdvanced = ($childSignature -and $childSignature -ne $previousChildActivitySignature)
|
|
764
769
|
$previousChildActivitySignature = $childSignature
|
|
765
770
|
|
|
766
771
|
$effectiveStaleKillThreshold = Get-PrizmEffectiveStaleKillThreshold -ProgressFile $progressJson -BaseThreshold $staleKillThreshold
|
|
767
772
|
|
|
768
773
|
# Check for error-loop: agent is actively producing output but results are
|
|
769
|
-
# all read-offset errors or wasted calls.
|
|
774
|
+
# all read-offset errors or wasted calls. Only a newly advancing error
|
|
775
|
+
# signature counts; stale historical errors must not turn later normal output
|
|
776
|
+
# into a stale kill.
|
|
770
777
|
$errorLoopDetected = $false
|
|
778
|
+
$errorLoopSignature = ''
|
|
771
779
|
if ($effectiveStaleKillThreshold -gt 0 -and $growth -gt 0 -and (Test-Path $progressJson)) {
|
|
772
780
|
try {
|
|
773
781
|
$progress = Get-Content $progressJson -Raw | ConvertFrom-Json
|
|
@@ -778,17 +786,21 @@ function Invoke-PrizmPipeline {
|
|
|
778
786
|
$_.type -in @("read_offset_overflow", "wasted_call")
|
|
779
787
|
}).Count -eq 5
|
|
780
788
|
if ($allBad) {
|
|
781
|
-
$
|
|
789
|
+
$errorLoopSignature = ($recent | ConvertTo-Json -Compress)
|
|
790
|
+
if ($previousErrorLoopSignature -and $errorLoopSignature -ne $previousErrorLoopSignature) {
|
|
791
|
+
$errorLoopDetected = $true
|
|
792
|
+
}
|
|
782
793
|
}
|
|
783
794
|
}
|
|
784
795
|
} catch {
|
|
785
796
|
# Ignore JSON parse errors — progress file may be incomplete or malformed
|
|
786
797
|
}
|
|
787
798
|
}
|
|
799
|
+
$previousErrorLoopSignature = $errorLoopSignature
|
|
788
800
|
|
|
789
801
|
if ($errorLoopDetected) {
|
|
790
802
|
$staleSeconds = $effectiveStaleKillThreshold
|
|
791
|
-
} elseif ($growth -gt 0 -or $childAdvanced) {
|
|
803
|
+
} elseif ($growth -gt 0 -or $childAdvanced -or $progressAdvanced) {
|
|
792
804
|
$staleSeconds = 0
|
|
793
805
|
} else {
|
|
794
806
|
$staleSeconds += $waitSeconds
|
|
@@ -121,6 +121,7 @@ $job = Start-Job -ScriptBlock {
|
|
|
121
121
|
$elapsedSeconds = 0
|
|
122
122
|
$staleSeconds = 0
|
|
123
123
|
$previousLogSize = 0
|
|
124
|
+
$previousProgressSignature = ''
|
|
124
125
|
$previousChildActivitySignature = ''
|
|
125
126
|
$wasTimedOut = $false
|
|
126
127
|
$wasStaleKilled = $false
|
|
@@ -136,12 +137,15 @@ while ($true) {
|
|
|
136
137
|
$growth = $currentLogSize - $previousLogSize
|
|
137
138
|
$previousLogSize = $currentLogSize
|
|
138
139
|
|
|
139
|
-
$
|
|
140
|
-
$
|
|
140
|
+
$progressActivity = Get-PrizmProgressActivity -ProgressFile $progressPath
|
|
141
|
+
$progressSignature = [string]$progressActivity.ProgressSignature
|
|
142
|
+
$progressAdvanced = ($progressSignature -and $progressSignature -ne $previousProgressSignature)
|
|
143
|
+
$previousProgressSignature = $progressSignature
|
|
144
|
+
$childSignature = [string]$progressActivity.ChildSignature
|
|
141
145
|
$childAdvanced = ($childSignature -and $childSignature -ne $previousChildActivitySignature)
|
|
142
146
|
$previousChildActivitySignature = $childSignature
|
|
143
147
|
|
|
144
|
-
if ($growth -gt 0 -or $childAdvanced) { $staleSeconds = 0 } else { $staleSeconds += $waitSeconds }
|
|
148
|
+
if ($growth -gt 0 -or $childAdvanced -or $progressAdvanced) { $staleSeconds = 0 } else { $staleSeconds += $waitSeconds }
|
|
145
149
|
|
|
146
150
|
if ($timeoutSeconds -gt 0 -and $elapsedSeconds -ge $timeoutSeconds) {
|
|
147
151
|
$wasTimedOut = $true
|