drydock-cli 2.8.14__tar.gz → 2.8.15__tar.gz

drydock_cli-2.8.15/.auto_release.lock

@@ -0,0 +1 @@
+479979

{drydock_cli-2.8.14 → drydock_cli-2.8.15}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: drydock-cli
-Version: 2.8.14
+Version: 2.8.15
 Summary: Local-first CLI coding agent — tested with Gemma 4 26B via vLLM
 Project-URL: Homepage, https://github.com/fbobe321/drydock
 Project-URL: Repository, https://github.com/fbobe321/drydock

{drydock_cli-2.8.14 → drydock_cli-2.8.15}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "drydock-cli"
-version = "2.8.14"
+version = "2.8.15"
 description = "Local-first CLI coding agent — tested with Gemma 4 26B via vLLM"
 readme = "README.md"
 requires-python = ">=3.12"

drydock_cli-2.8.15/scripts/classify_pulse.sh

@@ -0,0 +1,78 @@
+#!/bin/bash
+# classify_pulse.sh
+#
+# Periodic classifier runner. Reads recent log activity, classifies into
+# buckets, dispatches to ~/.drydock/dispatch/<bucket>.jsonl queues for
+# downstream automation (autonomous_review, GraphRAG curator, Deep Noir
+# vector intake, LoRA candidate pile).
+#
+# Incremental: tracks the last-read line number for each source file in
+# ~/.drydock/dispatch/.last_pos_<slug> so each run only classifies NEW
+# lines. This prevents the queue from accumulating duplicate entries when
+# the same 500-line window is re-classified every 10 minutes.
+#
+# Sources scanned:
+#   /data3/drydock/logs/autonomous_review.log         (new lines since last run)
+#   ~/.drydock/logs/admiral_history.log if present    (new lines since last run)
+#
+# Cron entry (when wired up):
+#   */10 * * * * /data3/drydock/scripts/classify_pulse.sh
+
+set -uo pipefail
+
+DRYDOCK=/data3/drydock
+PYTHON=/home/bobef/miniconda3/bin/python3
+LOG="$DRYDOCK/logs/classify_pulse.log"
+POS_DIR="$HOME/.drydock/dispatch"
+
+mkdir -p "$DRYDOCK/logs"
+exec >>"$LOG" 2>&1
+
+echo "================ $(date -u +%FT%TZ) tick ================"
+
+REVIEW_LOG="$DRYDOCK/logs/autonomous_review.log"
+ADMIRAL_LOG="$HOME/.drydock/logs/admiral_history.log"
+
+cd "$DRYDOCK" || exit 1
+
+# scan_new_lines <log_file> <slug> <source_label>
+# Reads only lines added since the last call. Updates the position file.
+scan_new_lines() {
+    local log_file="$1" slug="$2" source_label="$3"
+    local pos_file="$POS_DIR/.last_pos_$slug"
+    local last_line=0
+    [ -f "$pos_file" ] && last_line=$(cat "$pos_file")
+    local total_lines
+    total_lines=$(wc -l < "$log_file")
+    if [ "$total_lines" -le "$last_line" ]; then
+        echo "  (no new lines since last run, skipping)"
+        echo "$total_lines" > "$pos_file"
+        return
+    fi
+    local new_count=$(( total_lines - last_line ))
+    echo "  scanning $new_count new lines (lines $((last_line+1))–$total_lines)"
+    tail -n "+$((last_line + 1))" "$log_file" | $PYTHON -m drydock.core.classifier - \
+        --format summary \
+        --source "$source_label" \
+        --dispatch 2>&1 | tail -20
+    echo "$total_lines" > "$pos_file"
+}
+
+if [ -f "$REVIEW_LOG" ]; then
+    echo "--- autonomous_review.log ---"
+    scan_new_lines "$REVIEW_LOG" "review" "autonomous_review.log"
+fi
+
+if [ -f "$ADMIRAL_LOG" ]; then
+    echo ""
+    echo "--- admiral_history.log ---"
+    scan_new_lines "$ADMIRAL_LOG" "admiral" "admiral_history.log"
+fi
+
+echo ""
+echo "--- queue depths ---"
+for q in "$POS_DIR/"*.jsonl; do
+    [ -f "$q" ] && printf "  %-40s %d\n" "$(basename "$q")" "$(wc -l < "$q")"
+done
+
+echo "done"

{drydock_cli-2.8.14 → drydock_cli-2.8.15}/trip_log.md

@@ -1,5 +1,77 @@
 # Drydock Trip Log
 
+## 2026-05-11 09:30 UTC tick
+- Stress: 1658/1658 COMPLETE (run finished normally; PID 340932 gone; no new stress batch started)
+- Write rate: N/A (cycle complete)
+- Admiral last 30 min: 0 fires
+- vLLM 400s: 0 (llamacpp-gemma4 up, balancer PID 380535 on :8001 forwarding correctly)
+- GH issues: 0 open
+- Dispatch queue: harness=103694, retrieval=175 (drain: 0 actionable — all already ingested)
+- Action this tick: no action — system idle post-cycle completion, all dominant dispatch patterns (thinking_stall) already addressed in prior commits
+
+## 2026-05-11 09:00 UTC tick
+- Stress: 1658/1658 COMPLETE (PID 340932 dead — run finished normally at ~09:00 UTC; done=88 skip=53 timeout=0 recycle=42)
+- Write rate: 5.3% (88 accepted / 1658 prompts total cycle)
+- Admiral last 30 min: 0 fires (run complete, no active session)
+- vLLM 400s: 0; balancer on :8001 healthy (PID 380535 confirmed llm_balancer.py); llamacpp-gemma4 up (docker unhealthy status is health-check noise, API responds correctly)
+- GH issues: 0 open
+- Dispatch queue: harness=103391, retrieval=175 (drain: 0 ingested — all already consumed); steering=0
+- Action this tick: no fix committed — stress run completed full 1658/1658 cycle; retrieval drain returned 0 actionable; harness:thinking_stall dominant in queue but already handled in agent_loop.py ~line 1072; all services healthy; no new drydock bugs found
+
+## 2026-05-11 08:30 UTC tick
+- Stress: 1658/1658 COMPLETE (PID 340932 dead — run finished normally; accepted=88, skipped=53, timeouts=0, elapsed=555m)
+- Write rate: 5.3% (88 accepted / 1658 prompts; note: "accepted" here counts net new file writes, not TUI interactions)
+- Admiral last 30 min: 0 fires (stress-alert:complete logged at 08:15, retry-spike alert at 08:08 was final burst before completion)
+- vLLM 400s: 0; balancer on :8001 healthy, gemma4 up
+- GH issues: 0 open
+- Dispatch queue: harness=103088, retrieval=175 (drain: 0 ingested — all 175 entries already consumed); steering=N/A
+- Action this tick: no fix committed — stress run completed 1658/1658; retrieval queue fully drained; no new drydock bugs found; harness:thinking_stall pattern remains dominant in queue but already addressed in source; system idle pending next stress batch
+
+## 2026-05-11 08:00 UTC tick
+- Stress: 1652/1658 (PID 340932 alive; done=82 skip=52 this restart; 99.6% complete — final stretch)
+- Write rate: ~61% (82 done / 134 resolved attempts this restart)
+- Admiral last 30 min: 0 new fires (admiral_history.log quiet; harness:thinking_stall still dominant in queue but all recent outcomes via source=opus/canned)
+- vLLM 400s: 0; balancer on :8001 confirmed healthy; gemma4 up (unhealthy status from docker healthcheck, not actual model failure)
+- GH issues: 0 open
+- Dispatch queue: harness=102784, retrieval=175, steering=N/A; retrieval drain: 0 ingested (all already done)
+- Action this tick: no fix committed — stress run at 99.6% completion, all services healthy, thinking_stall pattern already addressed in source; no new actionable bugs found
+
+## 2026-05-11 07:00 UTC tick
+- Stress: 1635/1658 at last babysitter tick (PID 340932 alive; done=70 skip=47 recycle=35; 98.6% complete)
+- Write rate: ~60% (70 done / 117 resolved attempts)
+- Admiral last 30 min: thinking_stall pattern dominant; all resolved ("unstuck" per queue); no new patterns
+- vLLM 400s: 0; balancer OK on :8001 (gemma4 confirmed)
+- GH issues: 0 open
+- Dispatch queue: harness=102169, retrieval=175, steering=N/A; retrieval drain: 0 ingested (all already done)
+- Action this tick: no fix committed — system healthy, stress nearly done, all patterns already addressed in source
+
+## 2026-05-11 06:02 UTC tick
+- Stress: 1620/1658 (PID 340932 alive, stress_shakedown running; near completion)
+- Write rate: ~60% based on recent done/skip ratio (last batch: +26 msgs, +4 writes)
+- Admiral last 30 min: ~20 fires (thinking_stall pattern dominant in harness queue)
+- vLLM 400s: 0
+- GH issues: 0 open
+- Dispatch queue: harness=101570, retrieval=175 (retrieval drain: 0 ingested, all already done)
+- Action this tick: no fix committed — existing thinking_stall handling in agent_loop.py is comprehensive (grep/read/write/bash all have specific nudges); stress run progressing normally; system healthy
+
+## 2026-05-11 05:31 UTC tick
+- Stress: 1604/1658 at 05:00 UTC babysitter tick (PID 340932 alive; done=49 skip=37 recycle=25; run near end)
+- Write rate: ~56% (49 done / 87 resolved attempts this restart)
+- Admiral last 30 min: dispatch queue harness=101255 entries, all thinking_stall — addressed by prior commits; no new patterns
+- vLLM 400s: 0; balancer OK on :8001 (gemma4 forwarding confirmed)
+- GH issues: 0 open
+- Dispatch queue: harness=101255, retrieval=175 (0 actionable, all already ingested)
+- Action this tick: no action — healthy; retrieval drain: 0 new ingests; stress at 97% with no intervention needed
+
+## 2026-05-11 05:01 UTC tick
+- Stress: 1604/1658 (PID 340932 alive, etime ~6h; done=49 skip=37 timeout=0 recycle=25 — 43% skip rate due to TUI-not-accepting, known approval-modal issue)
+- Write rate: 49 done / 87 attempts = ~56% success rate this restart (babysitter ticked 05:00 UTC, 1 min ago)
+- Admiral last 30 min: dispatch queue dominated by thinking_stall (100930 harness entries, unchanged pattern) — already addressed by prior commits
+- vLLM 400s: 0 (container "unhealthy" health check but forwarding fine on :8000; balancer OK on :8001)
+- GH issues: 0 open
+- Dispatch queue: harness=100930, retrieval=175 (0 actionable — all already ingested)
+- Action this tick: no action — system healthy; retrieval drain: 0 new ingests; stress near end of run (54 steps remaining); no new actionable patterns beyond already-fixed admiral commits
+
 ## 2026-05-11 01:04 UTC tick
 - Stress: 1551/1658 (PID 340932 alive; brief SKIP spiral of 12 consecutive SKIPs during TUI recycles, recovered — new session session_20260511_010327 created and active with tool calls at idx ~1552)
 - Write rate: 22 done / 11 skip in first 33 steps of current restart (~67%)
@@ -4493,3 +4565,39 @@ restarted, cron self-match bug fixed in this same session).
 - GH issues: 0 open
 - Dispatch queue: harness=100594, retrieval=175 (drain: 0 new actionable — all already ingested)
 - Action this tick: no action — stress run progressing (~1580/1658), system healthy, all dispatch patterns (thinking_stall dominates at 100K+ entries) already addressed by prior commits
+
+## 2026-05-11 06:30 UTC tick
+- Stress: 1517/1658 (resuming at step 1517, PID 340932 alive, elapsed 7h30m)
+- Write rate: ~56% (from prior tick estimate)
+- Admiral last 30 min: 0 fires (no new active sessions)
+- vLLM 400s: 0 (llamacpp-gemma4 "unhealthy" label cosmetic; balancer PID 380535 forwarding on :8001 correctly)
+- GH issues: 0 open
+- Dispatch queue: harness=101873, retrieval=175 (drain: 0 new — all already ingested)
+- Action this tick: no action — system healthy, stress near completion, thinking_stall dispatch pattern (101K entries) already addressed by prior commits
+
+## 2026-05-11 07:30 UTC tick
+- Stress: 1645/1658 (nearly complete, 43 skips — TUI prompt-acceptance retries)
+- Write rate: ~15 writes per last 20 prompts
+- Admiral last 30 min: 0 fires
+- vLLM 400s: 0 (llamacpp-gemma4 "unhealthy" health-check label cosmetic; balancer serving correctly)
+- GH issues: 0 open
+- Dispatch queue: harness=102472, retrieval=175 (drain: 0 actionable — all already ingested)
+- Action this tick: no action — system healthy. Dominant dispatch patterns (thinking_stall 71K, bash_generic 13K, hallucinated_name 6K, search_replace:not_found 5K) all addressed by prior commits. Stress run 13 steps from completion.
+
+## 2026-05-11 10:00 UTC tick
+- Stress: 1658/1658 COMPLETE (run finished; PID 340932 gone; no new batch started)
+- Write rate: N/A (cycle complete)
+- Admiral last 30 min: 0 fires
+- vLLM 400s: 0 (balancer PID 380535 on :8001 healthy)
+- GH issues: 0 open
+- Dispatch queue: harness=103997, retrieval=175 (drain: 0 actionable — all already ingested)
+- Action this tick: no action — system idle post-cycle, dominant thinking_stall pattern (103K entries) already addressed by prior commits, no new actionable patterns
+
+## 2026-05-11 10:30 UTC tick
+- Stress: 1658/1658 COMPLETE (run finished at ~09:00 UTC; babysitter confirmed done=88 skip=53 recycle=42; no restart needed)
+- Write rate: N/A (cycle complete)
+- Admiral last 30 min: 0 fires
+- vLLM 400s: 0 (balancer PID 380535 on :8001 healthy; docker logs error is cosmetic — llamacpp container absent, gemma4 serving fine)
+- GH issues: 0 open
+- Dispatch queue: harness=103997 (104K, bloated by duplicates — fixed), retrieval=175 (drain: 0 actionable — all already ingested)
+- Action this tick: committed fix bd8201a — classify_pulse.sh was re-scanning last 500 lines of admiral_history.log every 10 min without cross-run dedup, causing 76x duplicate entries (13,464 bash_generic records from 177 unique evidence strings). Switched to incremental line-count tracking per source file (~/.drydock/dispatch/.last_pos_<slug>). Initialized position files to current EOF so next tick starts clean. Auto-release will ship at next 0/6/12/18 CDT tick.

drydock_cli-2.8.14/.auto_release.lock

@@ -1 +0,0 @@
-411933

drydock_cli-2.8.14/scripts/classify_pulse.sh

@@ -1,58 +0,0 @@
-#!/bin/bash
-# classify_pulse.sh
-#
-# Periodic classifier runner. Reads recent log activity, classifies into
-# buckets, dispatches to ~/.drydock/dispatch/<bucket>.jsonl queues for
-# downstream automation (autonomous_review, GraphRAG curator, Deep Noir
-# vector intake, LoRA candidate pile).
-#
-# Idempotent: dispatcher dedups (pattern_id, evidence) per run, so
-# re-classifying the same lines is cheap and doesn't bloat queues.
-#
-# Sources scanned:
-#   /data3/drydock/logs/autonomous_review.log         (last 500 lines)
-#   ~/.drydock/logs/admiral_history.log if present    (last 500 lines)
-#
-# Cron entry (when wired up):
-#   */10 * * * * /data3/drydock/scripts/classify_pulse.sh
-
-set -uo pipefail
-
-DRYDOCK=/data3/drydock
-PYTHON=/home/bobef/miniconda3/bin/python3
-LOG="$DRYDOCK/logs/classify_pulse.log"
-
-mkdir -p "$DRYDOCK/logs"
-exec >>"$LOG" 2>&1
-
-echo "================ $(date -u +%FT%TZ) tick ================"
-
-REVIEW_LOG="$DRYDOCK/logs/autonomous_review.log"
-ADMIRAL_LOG="$HOME/.drydock/logs/admiral_history.log"
-
-cd "$DRYDOCK" || exit 1
-
-if [ -f "$REVIEW_LOG" ]; then
-    echo "--- autonomous_review.log ---"
-    tail -500 "$REVIEW_LOG" | $PYTHON -m drydock.core.classifier - \
-        --format summary \
-        --source autonomous_review.log \
-        --dispatch 2>&1 | tail -20
-fi
-
-if [ -f "$ADMIRAL_LOG" ]; then
-    echo ""
-    echo "--- admiral_history.log ---"
-    tail -500 "$ADMIRAL_LOG" | $PYTHON -m drydock.core.classifier - \
-        --format summary \
-        --source admiral_history.log \
-        --dispatch 2>&1 | tail -20
-fi
-
-echo ""
-echo "--- queue depths ---"
-for q in "$HOME/.drydock/dispatch/"*.jsonl; do
-    [ -f "$q" ] && printf "  %-40s %d\n" "$(basename "$q")" "$(wc -l < "$q")"
-done
-
-echo "done"