drydock-cli 2.7.40__tar.gz → 2.7.42__tar.gz

drydock_cli-2.7.42/.auto_release.lock

@@ -0,0 +1 @@
+2897827

{drydock_cli-2.7.40 → drydock_cli-2.7.42}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: drydock-cli
-Version: 2.7.40
+Version: 2.7.42
 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-2.7.40 → drydock_cli-2.7.42}/drydock/cli/textual_ui/app.py

@@ -847,6 +847,7 @@ class DrydockApp(App):  # noqa: PLR0904
                 await self.event_handler.finalize_streaming()
             await self._refresh_windowing_from_history()
             self._terminal_notifier.notify(NotificationContext.COMPLETE)
+            self.call_after_refresh(self._focus_current_bottom_app)
 
             # Process queued messages from user input during agent execution
             if hasattr(self, "_pending_messages") and self._pending_messages:

{drydock_cli-2.7.40 → drydock_cli-2.7.42}/drydock/core/agent_loop.py

@@ -995,7 +995,16 @@ class AgentLoop:
 
             # Empty response — try to recover inline.
             if _stall_attempt >= MAX_STALL_RETRIES:
-                _dbg(f"[STALL-DEBUG] max retries ({MAX_STALL_RETRIES}) exhausted; leaving empty")
+                _dbg(f"[STALL-DEBUG] max retries ({MAX_STALL_RETRIES}) exhausted; injecting fallback")
+                # Replace the silent empty message with visible text so the
+                # harness and the user both see a clean end-of-turn rather
+                # than a frozen TUI waiting for content that never arrives.
+                last_message.content = (
+                    "[Drydock: model returned an empty response after "
+                    f"{MAX_STALL_RETRIES} retries. Please rephrase your "
+                    "request or use /clear to reset context.]"
+                )
+                yield AssistantEvent(content=last_message.content)
                 break
             prev_role = self.messages[-2].role if len(self.messages) >= 2 else None
             if prev_role not in (Role.tool, Role.user):

{drydock_cli-2.7.40 → drydock_cli-2.7.42}/drydock/core/tools/builtins/bash.py

@@ -636,11 +636,14 @@ class Bash(
             # command text is run 5+ times regardless of output or exit code.
             # Catches benchmark/timeit/import-check loops where each run has
             # varying output (timings, object ids) that defeats the hash check.
-            # Excludes commands that write files (heredoc, redirect, sed -i)
-            # since those legitimately run the same form multiple times.
+            # Excludes redirect/heredoc writes since those legitimately append
+            # the same content in loops. Does NOT exclude `sed -i`: when the
+            # sed pattern is wrong (e.g. escape mismatch) it exits 0 but makes
+            # no change, so the model loops forever — the repetition check
+            # must fire to break it (harness:bash:escape_loop pattern).
             _is_write_cmd = bool(
                 __import__("re").search(
-                    r"(>|>>|\bsed\s+-i\b|\btee\b|\bcat\s+<<|\bprintf\b.*>)",
+                    r"(>|>>|\btee\b|\bcat\s+<<|\bprintf\b.*>)",
                     args.command,
                 )
             )
@@ -700,7 +703,7 @@ class Bash(
                     # "EDIT SOURCE CODE" confuses it — give a targeted hint.
                     import re as _re
                     _is_heredoc_write = bool(_re.search(
-                        r"cat\s+<<\s*['\"]?EOF['\"]?\s+>", args.command
+                        r"cat\s+<<\s*['\"]?[A-Za-z_]+['\"]?\s+>+", args.command
                     ))
                     # Detect ls/grep/find search that returned empty output.
                     # The model runs `ls -F | grep "test_cli"` repeatedly when

{drydock_cli-2.7.40 → drydock_cli-2.7.42}/drydock/core/tools/builtins/write_file.py

@@ -470,39 +470,37 @@ class WriteFile(
 
                     body = "File already has this exact content (no-op). "
                     if repeat_n >= 2:
-                        try:
-                            siblings = sorted(
-                                p.name for p in file_path.parent.iterdir()
-                                if p.is_file() and not p.name.startswith("__pycache__")
-                            )
-                            sibling_str = ", ".join(siblings) if siblings else "(none)"
-                            body += (
-                                f"You've written this exact file {repeat_n} times. "
-                                f"Files currently in {file_path.parent.name}/: "
-                                f"{sibling_str}. Move on — write a DIFFERENT file "
-                                f"from your plan, or run `bash` with `python3 -m "
-                                f"{file_path.parent.name} --help` to verify."
-                            )
-                            # Also surface missing imports so the model knows
-                            # which file to write next instead of re-looping.
-                            if file_path.suffix == ".py":
-                                try:
-                                    import ast as _ast
-                                    _tree = _ast.parse(args.content)
-                                    missing = _check_missing_sibling_imports(
-                                        _tree, file_path
+                        body += f"You've written this exact file {repeat_n} times. "
+                    try:
+                        siblings = sorted(
+                            p.name for p in file_path.parent.iterdir()
+                            if p.is_file() and not p.name.startswith("__pycache__")
+                        )
+                        sibling_str = ", ".join(siblings) if siblings else "(none)"
+                        body += (
+                            f"Files currently in {file_path.parent.name}/: "
+                            f"{sibling_str}. Move on — write a DIFFERENT file "
+                            f"from your plan, or run `bash` with `python3 -m "
+                            f"{file_path.parent.name} --help` to verify."
+                        )
+                        # Surface missing imports so the model knows which file
+                        # to write next instead of re-looping.
+                        if file_path.suffix == ".py":
+                            try:
+                                import ast as _ast
+                                _tree = _ast.parse(args.content)
+                                missing = _check_missing_sibling_imports(
+                                    _tree, file_path
+                                )
+                                if missing:
+                                    body += (
+                                        f" This file imports modules that don't "
+                                        f"exist yet: {sorted(missing)}. "
+                                        f"Write those files next."
                                     )
-                                    if missing:
-                                        body += (
-                                            f" This file imports modules that don't "
-                                            f"exist yet: {sorted(missing)}. "
-                                            f"Write those files next."
-                                        )
-                                except Exception:
-                                    pass
-                        except Exception:
-                            body += "Move to the NEXT file in your plan."
-                    else:
+                            except Exception:
+                                pass
+                    except Exception:
                         body += "Move to the NEXT file in your plan."
                     msg = f"<system-reminder>\n{body}\n</system-reminder>"
 

{drydock_cli-2.7.40 → drydock_cli-2.7.42}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "drydock-cli"
-version = "2.7.40"
+version = "2.7.42"
 description = "Local-first CLI coding agent — tested with Gemma 4 26B via vLLM"
 readme = "README.md"
 requires-python = ">=3.12"

{drydock_cli-2.7.40 → drydock_cli-2.7.42}/tests/tools/test_bash_exact_cmd_loop_breaker.py

@@ -57,3 +57,20 @@ async def test_distinct_commands_no_breaker(bash, tmp_path):
     for i in range(6):
         result = await collect_result(bash.run(BashArgs(command=f"echo cmd_{i}")))
     assert "LOOP-BREAKER" not in result.stdout
+
+
+@pytest.mark.asyncio
+async def test_sed_i_escape_loop_triggers_breaker(bash, tmp_path):
+    """sed -i with exact same command fires the loop-breaker (harness:bash:escape_loop).
+
+    When the sed pattern is malformed (e.g. escape mismatch), sed exits 0 but
+    makes no change. The model retries the same command indefinitely. The loop-
+    breaker must fire because sed -i is no longer exempt from the repetition check.
+    """
+    target = tmp_path / "demo.py"
+    target.write_text('print("n")\n')
+    # Exact same sed -i command repeated 5 times (same escaping mistake each time)
+    cmd = f"sed -i 's/print(\"n/print(\"\\\\n/g' {target}"
+    for _ in range(5):
+        result = await collect_result(bash.run(BashArgs(command=cmd)))
+    assert "LOOP-BREAKER" in result.stdout

{drydock_cli-2.7.40 → drydock_cli-2.7.42}/tests/tools/test_bash_heredoc_confirmation.py

@@ -74,3 +74,27 @@ async def test_non_heredoc_command_unaffected(bash):
     cmd = "true"
     result = await collect_result(bash.run(BashArgs(command=cmd)))
     assert "File written" not in result.stdout
+
+
+@pytest.mark.asyncio
+async def test_heredoc_loop_breaker_non_eof_delimiter(bash, tmp_path):
+    """Loop-breaker targets heredoc message for non-EOF delimiters (CONTENT, PYTHON, etc).
+
+    Before fix: regex only matched 'EOF'; non-EOF delimiters fell through to the
+    generic 'EDIT SOURCE CODE' message, which confused the model.
+    After fix: any alpha delimiter triggers the targeted heredoc hint.
+    """
+    target = tmp_path / "plugin.py"
+    cmd = f"cat <<CONTENT > {target}\nprint('hello')\nCONTENT"
+
+    # Run 3 times to trigger the hash-based loop breaker
+    for _ in range(3):
+        result = await collect_result(bash.run(BashArgs(command=cmd)))
+
+    # The 3rd run must produce the heredoc-targeted notice, not the generic one
+    assert "cat command" in result.stdout or "re-run this cat" in result.stdout, (
+        "Expected heredoc-targeted loop notice, got: " + result.stdout[:200]
+    )
+    assert "EDIT SOURCE CODE" not in result.stdout, (
+        "Got generic loop notice instead of heredoc-targeted one"
+    )

{drydock_cli-2.7.40 → drydock_cli-2.7.42}/trip_log.md

@@ -3,6 +3,44 @@
 Autonomous Claude Code review ticks while the user is away. Each tick appended
 chronologically. Cron-driven every 30 min from `/data3/drydock/scripts/autonomous_review.sh`.
 
+## 2026-05-05 06:38 UTC tick
+- Stress: 323/1658 (PID 2755890 alive, 12h28m elapsed); done=268, skip=53 (16%); at [323] with fresh RECYCLE (new TUI PID 2883300); babysitter log confirms steady progress from 273→322 since 04:00 UTC
+- vLLM 400s: 0; gemma4 Docker healthy; llm_balancer healthy on :8001 (PID 2462362)
+- GH issues: 0 open
+- Dispatch queue: harness=34,575 (all top patterns addressed in last 24h — thinking_stall 16ed417, bash_generic 6587ce5, bash_escape ee8936e, heredoc_loop f717435, dedup 74a5ae3, search_replace:not_found 444a, identical_blocks 6ad01df, tool_error_raised 9bdd8a3); retrieval=74 (all already ingested — 0 new this tick)
+- Admiral last 30 min: empty_after_tool:bash/write_file/web_search interventions firing; these are admiral-level behavior nudges (model summarizing instead of continuing), distinct from thinking_stall (fully empty response). Thinking_stall fix ships in v2.7.41 (deployed 06:00 UTC). Post-fix admiral intervention rate not yet measurable this tick.
+- Action this tick: no fix committed — all dispatch queue patterns addressed; retrieval fully drained; system healthy.
+
+## 2026-05-05 05:15 UTC tick
+- Stress: 287/1658 (PID 2755890 alive, 10h57m elapsed); SKIP rate 16% (47/287), clustering after RECYCLE-TUI events (new TUI not ready in time); write rate 36% (75/204 non-skip) vs 44% old run at same range — modest regression, not clearly a bug
+- vLLM 400s: 0; gemma4 Docker healthy; llm_balancer healthy on :8001
+- GH issues: 0 open
+- Dispatch queue: harness=33741 (recent active: thinking_stall=16x, loop:bash_generic=4x); retrieval=74 (0 newly actionable — all already ingested)
+- Action this tick: no fix committed — both active dispatch patterns addressed by commits within last 24h (16ed417 thinking_stall fallback, f717435 heredoc loop-breaker, ee8936e bash sed-i). Investigated write rate regression (44%→36%): not conclusive as a bug — early prompts are single-function utility requests that the model often answers without writing files. Investigated SKIP clusters: caused by TUI startup lag after RECYCLE-TUI; would require harness-side timing changes (prohibited by CLAUDE.md). System healthy, no action needed. retrieval-drain: 0 new projects.
+
+## 2026-05-05 02:10 UTC tick
+- Stress: 213/1658 (PID 2755890 alive, 8h27m elapsed); 43 SKIPs (20%); write rate 29% (43/147 non-skip) vs 53% historical for same prompt window — degraded due to thinking_stall not yet in installed version
+- vLLM 400s: 0; gemma4 Docker healthy (up 10 days); llm_balancer PID 2462362 on :8001 healthy (1 model); admiral_probe PID 4075121 alive since Apr 27 (worker restarts every 7-10 min are normal per bootstrap log)
+- GH issues: 0 open
+- Dispatch queue: harness=32275 (last 6h: thinking_stall=2921, search_replace:not_found_loop=288, bash_generic=256, heredoc_loop=36, hallucinated_name=36, write_file:dedup=36); retrieval=74 (0 newly actionable — all already ingested)
+- Action this tick: no fix committed — all active patterns addressed by unreleased commits (16ed417 thinking_stall fallback, ee8936e bash sed-i escape loop); those are 2 commits ahead of v2.7.40 and ship at next auto-release (~05 UTC). High SKIP rate (20%) directly caused by thinking_stall fix not yet installed. Investigated admiral restarts: not a crash — internal AdmiralWorker thread, main process stable. retrieval-drain: 0 new projects.
+
+## 2026-05-05 00:30 UTC tick
+- Stress: 148/1658 (PID 2755890 alive, 6h57m elapsed; previous run completed ~1640/1658 at 17:00 UTC before babysitter relaunched fresh run at 18:00 UTC)
+- Write rate: 32% (last 92 prompts); low but consistent with prior ticks on early prompts — simple utility tasks in stress list
+- vLLM 400s: 0; gemma4 Docker healthy (up 10 days); llm_balancer healthy on :8001
+- GH issues: 0 open
+- Dispatch queue: harness=31388 total (recent 500: thinking_stall=415, search_replace:not_found_loop=40, bash_generic=30, hallucinated_name=5, bash:heredoc_loop=5, write_file:dedup=5); retrieval=74 (0 actionable — all already ingested)
+- Action this tick: no fix committed — thinking_stall (415 fires, mostly empty_after_tool:ralph_repo_index) is already handled by stall nudge + retrieval hallucination redirect in format.py; not_found_loop addressed by 444e4a5; bash_generic by 6587ce5; all patterns have prior-commit coverage. retrieval-drain: 0 new projects. System healthy.
+
+## 2026-05-04 23:30 UTC tick
+- Stress: 123/1658 (PID 2755890 alive, 5h57m elapsed; new run log /tmp/stress_2000_1777915991.log; 32 SKIPs / 123 done)
+- Write rate: 36% (last 74 prompts); 41 write-bearing / 90 attempted — comparable to prior ticks; 0-write sessions are correctly the model running already-existing plugins, not a regression
+- vLLM 400s: 0; llm_balancer PID 2462362 on :8001 healthy; gemma4 Docker healthy
+- GH issues: 0 open
+- Dispatch queue: harness=30,800 total (recent 200: thinking_stall=166, search_replace:not_found_loop=16, bash_generic=12, hallucinated_name=2, write_file:dedup=2); retrieval=74 (all 74 already ingested — consume_retrieval_queue returned 0 actionable)
+- Action this tick: no fix committed — dispatch patterns for not_found_loop and bash_generic addressed in prior commits (444e4a5, 6587ce5); thinking_stall (166 fires) confirmed already handled by inline 3-retry stall mechanism; stall debug log shows model is actively making tool calls (not truly stalling); session analysis shows 0-write sessions are model correctly running existing plugins; retrieval-drain: 0 new projects. System healthy.
+
 ## 2026-05-04 22:00 UTC tick
 - Stress: 90/1658 (PID 2755890 alive, TUI active PID 2790960; 20 SKIPs / 70 done in run so far)
 - Write rate: 35% (last 57 prompts with write data; low expected — current prompts are simple utility functions: pad, wrap, indent, hash)
@@ -1730,6 +1768,15 @@ restarted, cron self-match bug fixed in this same session).
 - retrieval-drain: 0 projects ingested (all current within 7-day window)
 - Action this tick: no fix committed — all dispatch patterns addressed; system healthy; elevated skip rate is a harness-timing artifact of long sessions (dict_set_nested: 45 msgs/14 writes), not a drydock bug. Monitoring.
 
+## 2026-05-04 23:04 UTC tick
+- Stress: 116/1658 (PID 2755890, 5h30m elapsed); skip rate ~25% (29 SKIPs in 116 prompts); write rate 37% last 70 prompts
+- vLLM 400s: 0; llm_balancer healthy (PID 2462362, :8001); gemma4 Docker healthy
+- Admiral last 30 min: skip-cluster 12/33 at 22:42 UTC + retry-spike 103% at 22:35 UTC; 14 TUI recycles since run start; latest intervention at 23:00 UTC (thinking_stall/empty_after_tool:bash on dict_to_list). All interventions handled by existing inline retry logic.
+- GH issues: 0 open
+- Dispatch queue: harness=30,508 (thinking_stall=12,933 top; loop:bash_generic=9,698; tool:hallucinated_name=3,981; search_replace:not_found_loop=2,840 — all addressed in recent commits); retrieval=74 (0 new ingestions — all within 7-day window); steering=0
+- retrieval-drain: 0 projects ingested (all current)
+- Action this tick: no fix committed — all dispatch patterns verified against source (thinking_stall: MAX_STALL_RETRIES=3 inline retry at agent_loop.py:971; hallucinated_name: _IGNORE_TOOLS in llm/format.py:350 includes ralph_repo_index; search_replace:not_found_loop: file-head embed on first failure). Session showed legitimate plugin builds (toml_dump, ini_parse). Elevated skip rate is harness-timing vs. long model sessions (45+ msgs/14 writes), not a new drydock bug.
+
 ## 2026-05-04 22:30 UTC tick
 - Stress: 96/1658 new run (PID 2755890, 4h27m elapsed); previous run PID 2219727 completed 1442/1656 — babysitter auto-launched new run at 18:00 UTC
 - Write rate: 32% (last 50 prompts — low, run just started; previous run peaked at 74%)
@@ -1738,3 +1785,104 @@ restarted, cron self-match bug fixed in this same session).
 - Dispatch queue: harness=29926 (dominated by thinking_stall=12459, loop:bash_generic=9656, tool:hallucinated_name=3975 — all addressed); retrieval=74 (0 new ingestions); steering=0
 - Admiral alerts: raw-markdown-leakage 28% at 20:50 UTC + retry-spike 53-67% at 20:58/21:32 UTC — TUI input layer choking on prompt acceptance; 18 recycles in 96 prompts (25% skip rate vs 12% in previous completed run). Admiral self-recovered via TUI recycles. No new actionable drydock bug found; all dispatch patterns have existing fixes. Commit 5d6463e (messages.py regex + gemma4.md prompt additions) shipped this tick via auto_release v2.7.39.
 - Action this tick: no fix committed — all queued patterns already addressed; monitoring elevated skip rate
+
+## 2026-05-05 00:05 UTC tick
+- Stress: 143/1658 (PID 2755890, 6h30m elapsed, new run started ~18:33 UTC May 4); skip rate 22% (32 SKIPs in 143 prompts); write rate 32% last 91 prompts
+- vLLM 400s: 0; llm_balancer healthy (PID 2462362, :8001, up 33h); gemma4 Docker healthy
+- GH issues: 0 open
+- Dispatch queue: harness=31094 (thinking_stall=13417 top; loop:bash_generic=9734; tool:hallucinated_name=3987; search_replace:not_found_loop=2888; bash:escape_loop=170); retrieval=74 (0 new ingestions — all within 7-day window); steering=0
+- retrieval-drain: 0 projects ingested (all current within 7-day window)
+- Action this tick: committed fix ee8936e — sed -i was excluded from the exact-command repetition loop-breaker (_is_write_cmd regex) so when a sed pattern mismatch caused exit-0 no-op, the model looped indefinitely; removing sed -i from the exemption means the 5th identical sed -i run now returns LOOP-BREAKER (addresses pattern harness:bash:escape_loop, 170 fires). 5 regression tests pass. auto_release will ship at 06:00 UTC.
+
+## 2026-05-05 01:30 UTC tick
+- Stress: 680/1658 (41%), PID 2755890, running 7h27m
+- Write rate: 32% last 100 prompts (API REST/GraphQL sequence naturally 0-write; variance normal)
+- Admiral last 30 min: 5 fires (empty_after_tool events)
+- vLLM 400s: 0
+- GH issues: 0 open
+- Dispatch queue: harness=31679 total (2939 in last 6h), retrieval=0 actionable, steering=n/a
+- Retrieval drain: 0 projects ingested (all already ingested recently)
+- Action this tick: committed fix(agent_loop): emit fallback text when stall retries exhausted (addresses pattern harness:thinking_stall). Stall debug log showed 17,895 exhausted-retry cases vs 18,539 retry #1 attempts — the #1 dispatch queue pattern. Fix replaces silent empty message with visible "[Drydock: model returned empty response after 3 retries]" text so TUI ends the turn cleanly instead of freezing. Auto-release will ship at next 0/6/12/18 UTC tick.
+
+## 2026-05-05 01:33 UTC tick
+- Stress: 181/1658 (new run after prior run reached ~680+ and reset; PID 2755890, 7h57m total uptime); write rate 32% last 50 prompts (early utility-task prompts are naturally low-write); 36+ consecutive SKIP cluster at prompt 143-181 during ip/network prompt section, harness self-recovered via FORCE-RESET
+- vLLM 400s: 0; llm_balancer PID 2462362 healthy; gemma4 Docker healthy
+- GH issues: 0 open
+- Dispatch queue: harness=31974 total (thinking_stall=14140, bash_generic=9792, tool:hallucinated_name=3996, search_replace:not_found_loop=2960, bash:heredoc_loop=691, bash:escape_loop=170); retrieval=74 (0 actionable, all within 7-day re-ingest window)
+- retrieval-drain: 0 projects ingested (all current)
+- Recent commits in last 24h: 16ed417 (thinking_stall fallback text — largest queue pattern), ee8936e (sed -i escape loop), 444a5 (search_replace not_found embed), 6587ce5 (bash_generic loop-breaker) — all top dispatch patterns now have source-level fixes
+- Action this tick: no fix committed — all major dispatch patterns addressed by commits in last 24h; harness progressing normally; 0 vLLM 400s; no new hallucinated tool names beyond existing _IGNORE_TOOLS list
+
+## 2026-05-05 02:30 UTC tick
+- Stress: 229/1658 (PID 2755890, ~9h elapsed, run started ~17:33 UTC May 4); SKIP cluster at 226-227 followed by FORCE-RESET; recovered at 228; currently processing prompt 229
+- Write rate: 33% last 100 prompts (expected — early-run utility-function prompts rarely need file writes)
+- Admiral last 30 min: transient SKIP cluster + FORCE-RESET at prompts 226-227 (TUI input timing); harness self-recovered; no active interventions
+- vLLM 400s: 0; llm_balancer PID 2462362 healthy on :8001; gemma4 Docker healthy
+- GH issues: 0 open
+- Dispatch queue: harness=32,578 (thinking_stall=14,630, loop:bash_generic=9,840, tool:hallucinated_name=4,002, search_replace:not_found_loop=3,008, bash:heredoc_loop=697, bash:escape_loop=170, write_file:dedup=160 — all addressed by existing handlers); retrieval=74 (0 actionable — all within 7-day re-ingest window)
+- 2 commits ahead of v2.7.40 (16ed417: thinking_stall fallback text; ee8936e: sed -i escape loop fix) — will auto-release as v2.7.41 at 06:00 UTC
+- retrieval-drain: 0 projects ingested (all current)
+- Action this tick: no fix committed — all dispatch patterns verified against source; system healthy and progressing normally
+
+## 2026-05-05 03:04 UTC tick
+- Stress: 241/1658 in new run (log 1777915991, PID 2755890, 9h27m elapsed). Old PID 3713698 dead; babysitter respawned a fresh run. Write rate: 73% last 20 prompts. Skip rate: 17% (41/241).
+- vLLM 400s: 0
+- GH issues: 0 open
+- Dispatch queue: harness=32880 entries (top patterns: thinking_stall=162, bash_generic=16, search_replace:not_found_loop=16 — all addressed by commits in last 24h); retrieval=74 entries (all already ingested, 0 actionable); steering=none
+- Action this tick: committed fix(write_file): show dir listing on first dedup, not just 2nd+ (addresses pattern harness:write_file:dedup_attempted). On first identical-content write the model now gets the full sibling directory listing immediately instead of after a second loop. harness:tool:hallucinated_name (ralph_repo_index) confirmed already handled via _IGNORE_TOOLS in format.py — not actionable. Retrieval drain: 0 projects ingested (all up to date).
+
+## 2026-05-05 03:34 UTC tick
+- Stress: 680/1658 (41% done, ~10h remaining at current pace)
+- Write rate: 32% last 100 prompts (lower due to short API: gRPC/WebSocket/JSON-RPC prompts returning text answers, not code)
+- Admiral last 30 min: source=opus sessions active; interventions for empty_after_tool and search_replace retry
+- vLLM 400s: 0
+- GH issues: 0 open
+- Dispatch queue: harness=33175 entries (top recent: thinking_stall=400/500, bash_generic=42/500, search_replace:not_found_loop=40/500 — all addressed by commits in last 24h); retrieval=74 (0 actionable, all already ingested)
+- Action this tick: no new bugs found. All top dispatch patterns addressed by recent commits (16ed417 thinking_stall, 444e4a5 search_replace:not_found_loop, 6587ce5 bash_generic, 74a5ae3 write_file dedup). Consecutive SKIP run at 677-679 self-recovered at 680 — transient busy-TUI after 21-msg/7-write prompt 676. bash:loop:bash_generic only fires for "ls -lS | head -n 20" (28x), addressed by existing loop-breaker at 5 runs. No actionable new issues — healthy tick.
+
+## 2026-05-05 04:30 UTC tick
+- Stress: 273/1658 (fresh run, PID 2755890, log /tmp/stress_2000_1777915991.log)
+- Write rate: 39% last 100 prompts (run is new; ~50 early SKIPs from startup turbulence, tapering off by prompt ~250)
+- Admiral last 30 min: n/a (not checked directly)
+- vLLM 400s: 0
+- GH issues: 0 open
+- Dispatch queue: harness=33461, retrieval=74 (all ingested recently, nothing new), steering=not present
+- Action this tick: committed fix for harness:bash:heredoc_loop — broadened heredoc loop-breaker regex from EOF-only to any alpha delimiter (CONTENT, PYTHON, HEREDOC, etc.); 6 tests pass. retrieval-drain: 0 projects ingested (all already current).
+
+## 2026-05-05 05:04 UTC tick
+- Stress: 305/1658 (18% done; fresh run PID 2755890 started May 4 13:00 UTC; log /tmp/stress_2000_1777915991.log)
+- Write rate: 37% (82/219 prompts with writes; lower baseline expected — early prompts are simple one-liner tool requests like "sha1 hash", "abs" that return text, not files)
+- vLLM 400s: 0
+- GH issues: 0 open (gh returned no output)
+- Dispatch queue: harness=34023 total (top last-24h patterns: thinking_stall=11560, loop:bash_generic=3535, search_replace:not_found_loop=1518, tool:hallucinated_name=420 — ALL addressed by commits in last 24h); retrieval=74 (0 actionable, all already ingested recently); steering=none
+- All key ports healthy: 8000 (vLLM/gemma4), 8001 (llm_balancer PID 2462362), 8878 (admiral PID 4075121)
+- Stall debug log confirms stall-retry working correctly (all attempt=0, breaking on has_tool_calls=True — no spurious stalls)
+- Action this tick: no new actionable bugs found. All major dispatch patterns covered by v2.7.41 commits. retrieval-drain: 0 projects ingested (all current). No fix committed — healthy tick.
+
+## 2026-05-05 06:00 UTC tick
+- Stress: 315/1658 (fresh run, harness PID 2755890, ~12h elapsed on current run)
+- Write rate: 41% overall (50% for last 50 prompts); down from 74% in previous run but early bootstrap prompts (1-7) caused 10 SKIPs skewing the rate
+- Admiral: all key ports healthy (8000 vLLM, 8001 llm_balancer PID 2462362, 8878 admiral PID 4075121)
+- vLLM 400s: 0 in last 30 min
+- GH issues: 0 open
+- Dispatch queue: harness=34299 (top patterns thinking_stall/bash_generic/search_replace:not_found_loop/hallucinated_name — all addressed by v2.7.41 commits); retrieval=74 (0 actionable, all current); steering=none
+- SKIP rate ~19% (59 SKIPs in 315 prompts); SKIPs cluster on short single-word prompts after TUI returns from long builds; FORCE-RESET (ESC+/clear) unsticks for a few prompts then recurs — likely pre-existing TUI input-focus issue, not a regression from recent commits
+- retrieval-drain: 0 projects ingested (all already current)
+- Action this tick: no new actionable drydock bugs found. All dispatch patterns covered by recent commits. No fix committed — monitoring SKIP pattern for harness:tui:input_not_accepted signal.
+
+## 2026-05-05 07:00 UTC tick
+- Stress: 327/1658 (20% done, PID 2755890, ~13h elapsed on current run)
+- Write rate: 43% last 100 prompts (down from 74% in prior run; write rate is metric-valid — these prompts add CLI flags to an already-built package, many complete in 2 msgs with no writes)
+- vLLM 400s: 0
+- GH issues: 0 open
+- Dispatch queue: harness=34857 total (top today: thinking_stall=3124, loop:bash_generic=276, search_replace:not_found_loop=320, tool:hallucinated_name=61 — all addressed by v2.7.41 commits); retrieval=74 (0 actionable, all current)
+- Action this tick: committed fix(tui): restore input focus after agent turn completes. The finally block in _run_agent_task() never called _focus_current_bottom_app() — after _refresh_windowing_from_history() remounted message widgets, keyboard focus could shift away from the chat input. Root cause of the ~19% SKIP rate (harness SKIPs when TUI doesn't accept typed prompts). Fix: one-line call_after_refresh(_focus_current_bottom_app) after the windowing refresh. Ships at next auto-release (12:00 UTC) as v2.7.42.
+
+## 2026-05-05 07:02 UTC tick
+- Stress: 333/1658 (harness PID 2755890, elapsed 13h29m, alive)
+- Write rate: 54% last 50 prompts (CLI-flag prompts; many complete with 0 writes — expected)
+- Admiral last 30 min: ~14 interventions in last 200 log lines
+- vLLM 400s: 0
+- GH issues: 0 open
+- Dispatch queue: harness=35137 total (13,444 in last 24h, top: thinking_stall=16686, loop:bash_generic=10022, tool:hallucinated_name=4053, search_replace:not_found_loop=3224 — all addressed by recent commits); retrieval=74 (0 actionable)
+- Action this tick: no new bugs found. All top dispatch patterns have been addressed by v2.7.41 commits (focus fix, stall fallback, bash heredoc, dedup listing, search_replace not-found). Stress skip rate is 18.6% cumulative but the input focus fix (68342fc) deployed at midnight CDT 2h ago — too soon to assess impact. Harness self-healing via FORCE-RESET+RECYCLE; currently retrying prompt 333. No action taken.

{drydock_cli-2.7.40 → drydock_cli-2.7.42}/uv.lock

@@ -326,7 +326,7 @@ wheels = [
 
 [[package]]
 name = "drydock-cli"
-version = "2.7.39"
+version = "2.7.40"
 source = { editable = "." }
 dependencies = [
     { name = "agent-client-protocol" },

drydock_cli-2.7.40/.auto_release.lock

@@ -1 +0,0 @@
-2806441