drydock-cli 3.0.86__tar.gz → 3.0.88__tar.gz

{drydock_cli-3.0.86/drydock_cli.egg-info → drydock_cli-3.0.88}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: drydock-cli
-Version: 3.0.86
+Version: 3.0.88
 Summary: Drydock — a local, provider-agnostic terminal coding agent for local LLMs
 Author: Frank Bobe III
 License-Expression: Apache-2.0

{drydock_cli-3.0.86 → drydock_cli-3.0.88}/drydock/agent.py

@@ -63,6 +63,7 @@ class AgentState:
     messages: list = field(default_factory=list)
     total_input_tokens: int = 0
     total_output_tokens: int = 0
+    last_input_tokens: int = 0  # prompt tokens the SERVER counted on the last call
     turn_count: int = 0
     current_effort: str = ""  # "high"/"low" of the in-flight LLM call (for the UI)
 
@@ -210,6 +211,7 @@ def run(
 
         state.total_input_tokens += assistant_turn.input_tokens
         state.total_output_tokens += assistant_turn.output_tokens
+        state.last_input_tokens = assistant_turn.input_tokens
         yield TurnDone(assistant_turn.input_tokens, assistant_turn.output_tokens)
 
         # No tool calls = conversation complete — UNLESS the model emitted a

{drydock_cli-3.0.86 → drydock_cli-3.0.88}/drydock/compaction.py

@@ -192,9 +192,16 @@ def emergency_compact(messages: list, context_limit: int = 131072) -> list:
 
 
 def maybe_compact(state, config: dict) -> None:
-    """Compact state.messages if approaching context limit."""
+    """Compact state.messages if approaching context limit.
+
+    Uses the MAX of the char-based estimate and the SERVER's real prompt-token
+    count from the last call. estimate_tokens (chars/3.5) undercounts token-dense
+    content — code, compiler/build output, tracebacks tokenize to ~3 chars/token
+    — so on such tasks the real context (what the gauge shows) can pass 60% while
+    the estimate is still under it, and compaction fires late. The real count
+    catches that."""
     limit = config.get("context_limit", 131072)
-    current = estimate_tokens(state.messages)
+    current = max(estimate_tokens(state.messages), getattr(state, "last_input_tokens", 0))
 
     if current > limit * 0.60:
         state.messages = compact(state.messages, limit)

{drydock_cli-3.0.86 → drydock_cli-3.0.88}/drydock/tools/__init__.py

@@ -85,7 +85,7 @@ SCHEMAS = [
     },
     {
         "name": "ViewImage",
-        "description": "Look at an image file with your vision — use this when you need to SEE a screenshot, mockup, diagram, photo, or rendered output (.png/.jpg/.jpeg/.gif/.webp/.bmp). The image is shown to you so you can describe it, read text from it, or debug it. (Reading an image with the Read tool gives binary garbage; use ViewImage instead.)",
+        "description": "Look at an image file with your vision — use this when you need to SEE a screenshot, mockup, diagram, photo, or rendered output (.png/.jpg/.jpeg/.gif/.webp/.bmp). The image is shown to you so you can describe it, read text from it, or debug it. To READ TEXT OR DATA out of an image — a scanned document, invoice, receipt, form, or a screenshot with text — use ViewImage FIRST (you can read it directly); prefer it over OCR tools like tesseract/pdftotext, which are often unreliable or not installed. (Reading an image with the Read tool gives binary garbage; use ViewImage instead.)",
         "input_schema": {
             "type": "object",
             "properties": {

{drydock_cli-3.0.86 → drydock_cli-3.0.88}/drydock/tuning.py

@@ -95,7 +95,9 @@ _DEFAULT_SYSTEM_PROMPT = (
     ".jpeg/.gif/.webp/.bmp on disk — a screenshot, mockup, diagram, or photo), it "
     "is attached and you can SEE it; you can also call the `ViewImage` tool to "
     "look at any image you find yourself (Read gives binary garbage for images — "
-    "use ViewImage). Describe it, read text from it, or debug with it — never "
+    "use ViewImage). To read text/data out of a scanned document, invoice, "
+    "receipt, or screenshot, use ViewImage directly rather than OCR tools like "
+    "tesseract. Describe it, read text from it, or debug with it — never "
     "claim you cannot view images."
 )
 
@@ -123,8 +125,10 @@ _GEMMA_SYSTEM_PROMPT = (
     "(.png/.jpg/.jpeg/.gif/.webp/.bmp that exists on disk — a screenshot, mockup, "
     "diagram, or photo), that image is attached and you can SEE it. You can also "
     "call the `ViewImage` tool to look at any image you find yourself (Read gives "
-    "binary garbage for images — use ViewImage). Describe it, read text from it, "
-    "or use it to debug — never claim you cannot view images."
+    "binary garbage for images — use ViewImage). To read TEXT or DATA out of a "
+    "scanned document, invoice, receipt, form, or a screenshot, use ViewImage "
+    "directly — don't reach for OCR tools like tesseract first. Describe it, read "
+    "text from it, or use it to debug — never claim you cannot view images."
 )
 
 

{drydock_cli-3.0.86 → drydock_cli-3.0.88/drydock_cli.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: drydock-cli
-Version: 3.0.86
+Version: 3.0.88
 Summary: Drydock — a local, provider-agnostic terminal coding agent for local LLMs
 Author: Frank Bobe III
 License-Expression: Apache-2.0

{drydock_cli-3.0.86 → drydock_cli-3.0.88}/pyproject.toml

@@ -7,7 +7,7 @@ build-backend = "setuptools.build_meta"
 # PyPI distribution name is drydock-cli (the established install name, continued
 # from the retired v2 fork); the import package + CLI command stay `drydock`.
 name = "drydock-cli"
-version = "3.0.86"
+version = "3.0.88"
 description = "Drydock — a local, provider-agnostic terminal coding agent for local LLMs"
 readme = "README.md"
 requires-python = ">=3.11"

drydock_cli-3.0.88/tests/test_compact_command.py

@@ -0,0 +1,91 @@
+"""The /compact slash command (advertised in /help) must actually compact:
+wire it to compaction.compact and report tokens freed. Regression for the gap
+where /help listed /compact but handle_command never implemented it."""
+from __future__ import annotations
+
+from drydock import cli
+from drydock.agent import AgentState
+from drydock.compaction import estimate_tokens
+
+
+def _state_with_big_tool_result() -> AgentState:
+    state = AgentState()
+    state.messages = [
+        {"role": "user", "content": "do a thing"},
+        {
+            "role": "assistant",
+            "content": "",
+            "tool_calls": [{"id": "c1", "name": "Bash", "input": {"command": "cat big"}}],
+        },
+        {"role": "tool", "tool_call_id": "c1", "content": "Z" * 8000},
+        {"role": "assistant", "content": "done"},
+    ]
+    return state
+
+
+def test_compact_command_is_handled():
+    """/compact returns True (handled), not False (falls through to the agent)."""
+    state = AgentState()
+    assert cli.handle_command("/compact", state, {"context_limit": 65536}) is True
+
+
+def test_compact_shrinks_a_bloated_conversation(capsys):
+    state = _state_with_big_tool_result()
+    before = estimate_tokens(state.messages)
+
+    cli.handle_command("/compact", state, {"context_limit": 65536})
+
+    after = estimate_tokens(state.messages)
+    assert after < before  # the 8000-char tool result was truncated
+    out = capsys.readouterr().out.lower()
+    assert "compact" in out and "freed" in out
+
+
+def test_compact_on_empty_conversation_is_a_noop(capsys):
+    state = AgentState()
+    assert cli.handle_command("/compact", state, {"context_limit": 65536}) is True
+    assert "nothing to compact" in capsys.readouterr().out.lower()
+
+
+def test_maybe_compact_uses_real_token_count_not_just_estimate():
+    """maybe_compact must trigger on the SERVER's real prompt-token count, not
+    only the chars/3.5 estimate (which undercounts dense code/build output).
+    Regression: build task hit 64% real ctx without compacting because the
+    estimate read ~55%."""
+    from drydock.agent import AgentState
+    from drydock.compaction import maybe_compact, estimate_tokens
+
+    st = AgentState()
+    # A few small messages: the char estimate is tiny (well under 60% of 65536)…
+    st.messages = [
+        {"role": "user", "content": "hi"},
+        {"role": "assistant", "content": "hello", "tool_calls": []},
+    ]
+    assert estimate_tokens(st.messages) < 0.60 * 65536      # estimate says "fine"
+    cfg = {"context_limit": 65536}
+    # …but the server reported a large real prompt (dense content) → must compact.
+    st.last_input_tokens = 45000                            # 68% of 65536
+    before = list(st.messages)
+    maybe_compact(st, cfg)
+    # compact() ran (it targets 60%); with only tiny messages it can't shrink, but
+    # the point is it was INVOKED on the real count — verify the trigger path by
+    # checking a case with droppable content:
+    st.messages = before + [
+        {"role": "assistant", "content": "", "tool_calls": [{"id": "a", "name": "Bash", "input": {"command": "x"}}]},
+        {"role": "tool", "tool_call_id": "a", "name": "Bash", "content": "Z" * 200000},  # huge old tool result
+    ]
+    st.last_input_tokens = 45000
+    maybe_compact(st, cfg)
+    # the huge old tool result should be trimmed by compact()
+    assert estimate_tokens(st.messages) < estimate_tokens(before + [{"role": "tool", "content": "Z" * 200000}])
+
+
+def test_maybe_compact_low_real_count_no_trigger():
+    from drydock.agent import AgentState
+    from drydock.compaction import maybe_compact
+    st = AgentState()
+    st.messages = [{"role": "user", "content": "small"}]
+    st.last_input_tokens = 1000
+    before = list(st.messages)
+    maybe_compact(st, {"context_limit": 65536})
+    assert st.messages == before      # nothing to do, unchanged

{drydock_cli-3.0.86 → drydock_cli-3.0.88}/tests/test_viewimage.py

@@ -74,3 +74,15 @@ def test_image_load_error_classifier():
     assert is_image_load_error("invalid_request_error: could not decode image")
     assert not is_image_load_error("rate limit exceeded")
     assert not is_image_load_error("context_length_exceeded")
+
+
+def test_viewimage_nudges_over_ocr_for_documents():
+    """The tool description + both system prompts steer the model to ViewImage
+    (its own vision) for reading text off invoices/receipts/scans, instead of
+    reaching for OCR tools first (observed on the financial-doc tbench task)."""
+    from drydock.tuning import system_prompt_for_model
+    desc = reg.get("ViewImage").schema["description"]
+    assert "OCR" in desc and ("invoice" in desc or "receipt" in desc)
+    for model in ("gemma4", "some-other-model"):
+        p = system_prompt_for_model(model)
+        assert "OCR" in p and "ViewImage" in p

drydock_cli-3.0.86/tests/test_compact_command.py

@@ -1,47 +0,0 @@
-"""The /compact slash command (advertised in /help) must actually compact:
-wire it to compaction.compact and report tokens freed. Regression for the gap
-where /help listed /compact but handle_command never implemented it."""
-from __future__ import annotations
-
-from drydock import cli
-from drydock.agent import AgentState
-from drydock.compaction import estimate_tokens
-
-
-def _state_with_big_tool_result() -> AgentState:
-    state = AgentState()
-    state.messages = [
-        {"role": "user", "content": "do a thing"},
-        {
-            "role": "assistant",
-            "content": "",
-            "tool_calls": [{"id": "c1", "name": "Bash", "input": {"command": "cat big"}}],
-        },
-        {"role": "tool", "tool_call_id": "c1", "content": "Z" * 8000},
-        {"role": "assistant", "content": "done"},
-    ]
-    return state
-
-
-def test_compact_command_is_handled():
-    """/compact returns True (handled), not False (falls through to the agent)."""
-    state = AgentState()
-    assert cli.handle_command("/compact", state, {"context_limit": 65536}) is True
-
-
-def test_compact_shrinks_a_bloated_conversation(capsys):
-    state = _state_with_big_tool_result()
-    before = estimate_tokens(state.messages)
-
-    cli.handle_command("/compact", state, {"context_limit": 65536})
-
-    after = estimate_tokens(state.messages)
-    assert after < before  # the 8000-char tool result was truncated
-    out = capsys.readouterr().out.lower()
-    assert "compact" in out and "freed" in out
-
-
-def test_compact_on_empty_conversation_is_a_noop(capsys):
-    state = AgentState()
-    assert cli.handle_command("/compact", state, {"context_limit": 65536}) is True
-    assert "nothing to compact" in capsys.readouterr().out.lower()