drydock-cli 3.0.56__tar.gz → 3.0.58__tar.gz

{drydock_cli-3.0.56/drydock_cli.egg-info → drydock_cli-3.0.58}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: drydock-cli
-Version: 3.0.56
+Version: 3.0.58
 Summary: Drydock — a local, provider-agnostic terminal coding agent for local LLMs
 Author: Frank Bobe III
 License-Expression: Apache-2.0
@@ -23,9 +23,10 @@ Dynamic: license-file
 # ⚓ Drydock
 
 A local-first, provider-agnostic **terminal coding agent** for your own LLM.
-No accounts, no telemetry, no cloud — the only network call it makes is to
-the model endpoint you configure. Primary target: **Gemma-4-26B-A4B** served
-by llama.cpp on a single workstation.
+No accounts, no telemetry, no cloud — the only outbound calls are to the model
+endpoint you configure and (optionally) the web-search tools you invoke.
+Primary target: **dense Gemma-4-31B** (QAT, 64K) served by llama.cpp on a
+single workstation.
 
 > **v3 — clean-room rebuild.** Drydock is being rebuilt as an original,
 > Apache-2.0 codebase owned end to end (no upstream fork). Every release is
@@ -44,9 +45,31 @@ your box.
 
 Shipping. Published on PyPI as **`drydock-cli`** (v3.x). The Textual TUI is the
 default surface: a scrolling transcript with streamed assistant text, collapsible
-tool cards, a live nautical activity line, and a multi-line prompt. The agent
-loop, OpenAI-compatible provider, two-tier compaction, and the core tools
-(Read/Write/Edit/Bash/Glob/Grep) are in, with Gemma reliability hardening verified.
+tool cards, collapsible reasoning ("thinking") cards, a live nautical activity
+line, and a multi-line prompt. The agent loop, OpenAI-compatible provider,
+two-tier compaction, and the full agentic toolset (below) are in, with Gemma
+reliability hardening verified hands-on.
+
+## Capabilities
+
+A full agentic CLI harness — every tool below is clean-room and dependency-free
+(nothing beyond `openai` + `textual`), and the model calls them autonomously:
+
+- **Files & shell** — `Read` (with a structure index for huge files), `Write`,
+  `Edit`, `Bash`, `Glob`, `Grep`.
+- **Version control** — `GitStatus`, `GitDiff`, `GitLog`, `GitCommit`
+  (structured + truncated; commit is local and reversible).
+- **Internet** — `WebSearch` + `WebFetch` (DuckDuckGo; offline-safe).
+- **Knowledge base (GraphRAG)** — build a local entity-graph index from your
+  docs/code with `/graphrag build <path>`; the agent retrieves from it via the
+  read-only `Knowledge` tool.
+- **Multi-agent** — `Dispatch` runs several read-only sub-agents in parallel;
+  `task` runs one — each in a fresh context, for focused investigation.
+- **MCP** — connect to Model Context Protocol servers (`~/.drydock/mcp.json`);
+  their tools appear as `mcp__<server>__<tool>`. List them with `/mcp`.
+- **Skills** — reusable `/<name>` commands authored as markdown in
+  `~/.drydock/skills/` (or `<project>/.drydock/skills/`); `$ARGS` substitution.
+- **Loops** — `/loop <count> <prompt>` runs a prompt iteratively (Esc stops).
 
 ## Install
 
@@ -55,7 +78,7 @@ pip install drydock-cli
 drydock
 ```
 
-Requires Python 3.12+. From source instead:
+Requires Python 3.11+. From source instead:
 `git clone https://github.com/fbobe321/drydock-v3.git && cd drydock-v3 && pip install -e .`
 
 On first launch with no config, Drydock probes localhost for a running local
@@ -76,8 +99,11 @@ commands to do the work, showing each as a collapsible tool card.
 - A live activity line shows progress while it works:
   `◡ Keelhauling…  (12s · ↓ 6.2k tokens · thinking with high effort)`
 - Submit while it's working and the prompt **queues** (drains in order)
-- Slash commands: `/model [name]` · `/cwd [path]` · `/undo` (revert last write)
-  · `/back` (rewind last turn) · `/status` · `/clear` · `/help` · `/quit`
+- Slash commands: `/model` · `/cwd` · `/undo` (revert last write) · `/back`
+  (rewind last turn) · `/status` · `/compact` (shrink context) · `/graphrag`
+  (build/query a knowledge base) · `/skills` (list your `/<name>` skills) ·
+  `/loop` (repeat a prompt) · `/mcp` (list MCP servers) · `/clear` · `/help` ·
+  `/quit`
 
 It honors `AGENTS.md` / `DRYDOCK.md` in the working directory for project
 conventions.
@@ -96,22 +122,26 @@ blocked:
   `raise` outside an except, and refuses to write git conflict-marker content.
 
 Point it at a local OpenAI-compatible endpoint (e.g. llama.cpp's `server-cuda`
-serving Gemma 4 26B).
+serving Gemma-4-31B). The web tools (`WebSearch`/`WebFetch`) are read-only and
+degrade cleanly offline; the release scanner allowlists only the search backend.
 
 ## Model server (reference setup)
 
 Drydock is provider-agnostic, but it's tuned and measured against this rig:
 
-- **Model:** Gemma-4-26B-A4B-it (26B MoE, ~4B active/token), Unsloth `Q3_K_M`
-  GGUF, served by `ghcr.io/ggml-org/llama.cpp:server-cuda` with `--jinja`.
-- **GPUs:** 2× NVIDIA RTX 4060 Ti 16GB. The Q3_K_M weights (~13 GB) fit on a
-  **single 16 GB card**, so each GPU runs a **full, independent instance** —
-  two cards give two parallel instances for throughput; the model is **not**
-  tensor-split or "pooled" across both.
+- **Model:** dense **Gemma-4-31B** (QAT `Q4_K_XL` GGUF), served by
+  `ghcr.io/ggml-org/llama.cpp:server-cuda` with `--jinja`. Swapped from the
+  26B-A4B MoE, whose ~4B active params caused fatal agentic tool-loops; the
+  dense 31B is loop-free (slower, but it finishes).
+- **GPUs:** 2× NVIDIA RTX 4060 Ti 16GB, **tensor-split** across both cards
+  (`--tensor-split 1,1`) so the 31B weights fit.
 - **Context:** 64k (`-c 65536`) with `q8_0` KV-cache quantization
-  (`-ctk q8_0 -ctv q8_0`) — 64k @ q8 fits roughly the same VRAM as 32k @ f16.
-- **Throughput:** ~**64 tok/s** decode, ~94 tok/s prompt (per instance, Q3_K_M).
-- **Minimum:** any single 16 GB+ CUDA card runs it.
+  (`-ctk q8_0 -ctv q8_0`); set `context_limit` in `~/.drydock/config.toml` to
+  match your server's `-c`.
+- **Throughput:** ~15 tok/s decode (tensor-split 31B). Faster single-GPU
+  options exist if you drop to a smaller model.
+- **Provider-agnostic:** any OpenAI-compatible endpoint (llama.cpp, vLLM,
+  Ollama, LM Studio) works — point `--base-url` at it.
 
 ## Principles
 

{drydock_cli-3.0.56 → drydock_cli-3.0.58}/README.md

@@ -1,9 +1,10 @@
 # ⚓ Drydock
 
 A local-first, provider-agnostic **terminal coding agent** for your own LLM.
-No accounts, no telemetry, no cloud — the only network call it makes is to
-the model endpoint you configure. Primary target: **Gemma-4-26B-A4B** served
-by llama.cpp on a single workstation.
+No accounts, no telemetry, no cloud — the only outbound calls are to the model
+endpoint you configure and (optionally) the web-search tools you invoke.
+Primary target: **dense Gemma-4-31B** (QAT, 64K) served by llama.cpp on a
+single workstation.
 
 > **v3 — clean-room rebuild.** Drydock is being rebuilt as an original,
 > Apache-2.0 codebase owned end to end (no upstream fork). Every release is
@@ -22,9 +23,31 @@ your box.
 
 Shipping. Published on PyPI as **`drydock-cli`** (v3.x). The Textual TUI is the
 default surface: a scrolling transcript with streamed assistant text, collapsible
-tool cards, a live nautical activity line, and a multi-line prompt. The agent
-loop, OpenAI-compatible provider, two-tier compaction, and the core tools
-(Read/Write/Edit/Bash/Glob/Grep) are in, with Gemma reliability hardening verified.
+tool cards, collapsible reasoning ("thinking") cards, a live nautical activity
+line, and a multi-line prompt. The agent loop, OpenAI-compatible provider,
+two-tier compaction, and the full agentic toolset (below) are in, with Gemma
+reliability hardening verified hands-on.
+
+## Capabilities
+
+A full agentic CLI harness — every tool below is clean-room and dependency-free
+(nothing beyond `openai` + `textual`), and the model calls them autonomously:
+
+- **Files & shell** — `Read` (with a structure index for huge files), `Write`,
+  `Edit`, `Bash`, `Glob`, `Grep`.
+- **Version control** — `GitStatus`, `GitDiff`, `GitLog`, `GitCommit`
+  (structured + truncated; commit is local and reversible).
+- **Internet** — `WebSearch` + `WebFetch` (DuckDuckGo; offline-safe).
+- **Knowledge base (GraphRAG)** — build a local entity-graph index from your
+  docs/code with `/graphrag build <path>`; the agent retrieves from it via the
+  read-only `Knowledge` tool.
+- **Multi-agent** — `Dispatch` runs several read-only sub-agents in parallel;
+  `task` runs one — each in a fresh context, for focused investigation.
+- **MCP** — connect to Model Context Protocol servers (`~/.drydock/mcp.json`);
+  their tools appear as `mcp__<server>__<tool>`. List them with `/mcp`.
+- **Skills** — reusable `/<name>` commands authored as markdown in
+  `~/.drydock/skills/` (or `<project>/.drydock/skills/`); `$ARGS` substitution.
+- **Loops** — `/loop <count> <prompt>` runs a prompt iteratively (Esc stops).
 
 ## Install
 
@@ -33,7 +56,7 @@ pip install drydock-cli
 drydock
 ```
 
-Requires Python 3.12+. From source instead:
+Requires Python 3.11+. From source instead:
 `git clone https://github.com/fbobe321/drydock-v3.git && cd drydock-v3 && pip install -e .`
 
 On first launch with no config, Drydock probes localhost for a running local
@@ -54,8 +77,11 @@ commands to do the work, showing each as a collapsible tool card.
 - A live activity line shows progress while it works:
   `◡ Keelhauling…  (12s · ↓ 6.2k tokens · thinking with high effort)`
 - Submit while it's working and the prompt **queues** (drains in order)
-- Slash commands: `/model [name]` · `/cwd [path]` · `/undo` (revert last write)
-  · `/back` (rewind last turn) · `/status` · `/clear` · `/help` · `/quit`
+- Slash commands: `/model` · `/cwd` · `/undo` (revert last write) · `/back`
+  (rewind last turn) · `/status` · `/compact` (shrink context) · `/graphrag`
+  (build/query a knowledge base) · `/skills` (list your `/<name>` skills) ·
+  `/loop` (repeat a prompt) · `/mcp` (list MCP servers) · `/clear` · `/help` ·
+  `/quit`
 
 It honors `AGENTS.md` / `DRYDOCK.md` in the working directory for project
 conventions.
@@ -74,22 +100,26 @@ blocked:
   `raise` outside an except, and refuses to write git conflict-marker content.
 
 Point it at a local OpenAI-compatible endpoint (e.g. llama.cpp's `server-cuda`
-serving Gemma 4 26B).
+serving Gemma-4-31B). The web tools (`WebSearch`/`WebFetch`) are read-only and
+degrade cleanly offline; the release scanner allowlists only the search backend.
 
 ## Model server (reference setup)
 
 Drydock is provider-agnostic, but it's tuned and measured against this rig:
 
-- **Model:** Gemma-4-26B-A4B-it (26B MoE, ~4B active/token), Unsloth `Q3_K_M`
-  GGUF, served by `ghcr.io/ggml-org/llama.cpp:server-cuda` with `--jinja`.
-- **GPUs:** 2× NVIDIA RTX 4060 Ti 16GB. The Q3_K_M weights (~13 GB) fit on a
-  **single 16 GB card**, so each GPU runs a **full, independent instance** —
-  two cards give two parallel instances for throughput; the model is **not**
-  tensor-split or "pooled" across both.
+- **Model:** dense **Gemma-4-31B** (QAT `Q4_K_XL` GGUF), served by
+  `ghcr.io/ggml-org/llama.cpp:server-cuda` with `--jinja`. Swapped from the
+  26B-A4B MoE, whose ~4B active params caused fatal agentic tool-loops; the
+  dense 31B is loop-free (slower, but it finishes).
+- **GPUs:** 2× NVIDIA RTX 4060 Ti 16GB, **tensor-split** across both cards
+  (`--tensor-split 1,1`) so the 31B weights fit.
 - **Context:** 64k (`-c 65536`) with `q8_0` KV-cache quantization
-  (`-ctk q8_0 -ctv q8_0`) — 64k @ q8 fits roughly the same VRAM as 32k @ f16.
-- **Throughput:** ~**64 tok/s** decode, ~94 tok/s prompt (per instance, Q3_K_M).
-- **Minimum:** any single 16 GB+ CUDA card runs it.
+  (`-ctk q8_0 -ctv q8_0`); set `context_limit` in `~/.drydock/config.toml` to
+  match your server's `-c`.
+- **Throughput:** ~15 tok/s decode (tensor-split 31B). Faster single-GPU
+  options exist if you drop to a smaller model.
+- **Provider-agnostic:** any OpenAI-compatible endpoint (llama.cpp, vLLM,
+  Ollama, LM Studio) works — point `--base-url` at it.
 
 ## Principles
 

{drydock_cli-3.0.56 → drydock_cli-3.0.58}/drydock/tools/__init__.py

@@ -11,9 +11,17 @@ import difflib
 import glob as _glob
 import signal
 import subprocess
+import threading
 import time
 from pathlib import Path
 
+# Hard ceiling on bytes read from a command's output. communicate() buffers ALL
+# stdout in RAM before we ever truncate for context, so a runaway/infinite-output
+# command (`yes`, `cat /dev/urandom`, a massive build log) could balloon memory
+# to gigabytes within the timeout. We stream with a byte cap and kill the command
+# once it's hit — bounding RAM regardless of how much it tries to produce.
+_MAX_BASH_OUTPUT_BYTES = 256 * 1024  # 256 KB — plenty of context, safe for RAM
+
 from drydock.tool_registry import ToolDef, register
 from drydock.guards import (
     conflict_marker_refusal,
@@ -728,30 +736,59 @@ def tool_bash(params: dict, config: dict) -> str:
             text=True, cwd=config.get("cwd"), start_new_session=True,
         )
         config.setdefault("_abort", {})["proc"] = proc
+        # Read output in a daemon thread with a HARD byte cap, so memory can't
+        # balloon on a runaway-output command. The thread stops (and we kill the
+        # process) once the cap is hit; the main loop polls cancel + timeout.
+        chunks: list[str] = []
+        total = [0]
+        capped = threading.Event()
+
+        def _drain():
+            assert proc.stdout is not None
+            while True:
+                block = proc.stdout.read(8192)
+                if not block:
+                    break
+                chunks.append(block)
+                total[0] += len(block)
+                if total[0] >= _MAX_BASH_OUTPUT_BYTES:
+                    capped.set()
+                    break
+
+        reader = threading.Thread(target=_drain, daemon=True)
+        reader.start()
         start = time.monotonic()
-        while True:
-            try:
-                out, _ = proc.communicate(timeout=0.5)
+        while reader.is_alive():
+            reader.join(0.3)
+            if capped.is_set():
+                kill_process_group(proc)  # stop it producing more
                 break
-            except subprocess.TimeoutExpired:
-                if cancel is not None and cancel.is_set():
-                    kill_process_group(proc)
-                    proc.communicate()
-                    return "[stopped by user]"
-                if time.monotonic() - start > timeout:
-                    kill_process_group(proc)
-                    proc.communicate()
-                    bigger = min(timeout * 4, 1800)
-                    msg = (
-                        f"Error: command timed out after {timeout}s. If it is "
-                        f"legitimately slow (a big query, build, download, or "
-                        f"test run), retry with a larger timeout — pass "
-                        f"timeout: {bigger}. Otherwise it may be hung."
-                    )
-                    if _is_network_command(cmd):
-                        msg += _OFFLINE_HINT
-                    return msg
-        output = out or ""
+            if cancel is not None and cancel.is_set():
+                kill_process_group(proc)
+                proc.wait()
+                return "[stopped by user]"
+            if time.monotonic() - start > timeout:
+                kill_process_group(proc)
+                proc.wait()
+                bigger = min(timeout * 4, 1800)
+                msg = (
+                    f"Error: command timed out after {timeout}s. If it is "
+                    f"legitimately slow (a big query, build, download, or "
+                    f"test run), retry with a larger timeout — pass "
+                    f"timeout: {bigger}. Otherwise it may be hung."
+                )
+                if _is_network_command(cmd):
+                    msg += _OFFLINE_HINT
+                return msg
+        proc.wait()
+        output = "".join(chunks)
+        if capped.is_set():
+            return (
+                output[:_MAX_BASH_OUTPUT_BYTES]
+                + f"\n[output truncated at {_MAX_BASH_OUTPUT_BYTES // 1024} KB — "
+                "the command produced more; redirect to a file and inspect it in "
+                "pieces (head/tail/grep) instead of dumping it all]"
+            )
         if proc.returncode != 0:
             output += f"\n[exit code: {proc.returncode}]"
             # Offline environments make downloads fail forever; the model tends

{drydock_cli-3.0.56 → drydock_cli-3.0.58/drydock_cli.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: drydock-cli
-Version: 3.0.56
+Version: 3.0.58
 Summary: Drydock — a local, provider-agnostic terminal coding agent for local LLMs
 Author: Frank Bobe III
 License-Expression: Apache-2.0
@@ -23,9 +23,10 @@ Dynamic: license-file
 # ⚓ Drydock
 
 A local-first, provider-agnostic **terminal coding agent** for your own LLM.
-No accounts, no telemetry, no cloud — the only network call it makes is to
-the model endpoint you configure. Primary target: **Gemma-4-26B-A4B** served
-by llama.cpp on a single workstation.
+No accounts, no telemetry, no cloud — the only outbound calls are to the model
+endpoint you configure and (optionally) the web-search tools you invoke.
+Primary target: **dense Gemma-4-31B** (QAT, 64K) served by llama.cpp on a
+single workstation.
 
 > **v3 — clean-room rebuild.** Drydock is being rebuilt as an original,
 > Apache-2.0 codebase owned end to end (no upstream fork). Every release is
@@ -44,9 +45,31 @@ your box.
 
 Shipping. Published on PyPI as **`drydock-cli`** (v3.x). The Textual TUI is the
 default surface: a scrolling transcript with streamed assistant text, collapsible
-tool cards, a live nautical activity line, and a multi-line prompt. The agent
-loop, OpenAI-compatible provider, two-tier compaction, and the core tools
-(Read/Write/Edit/Bash/Glob/Grep) are in, with Gemma reliability hardening verified.
+tool cards, collapsible reasoning ("thinking") cards, a live nautical activity
+line, and a multi-line prompt. The agent loop, OpenAI-compatible provider,
+two-tier compaction, and the full agentic toolset (below) are in, with Gemma
+reliability hardening verified hands-on.
+
+## Capabilities
+
+A full agentic CLI harness — every tool below is clean-room and dependency-free
+(nothing beyond `openai` + `textual`), and the model calls them autonomously:
+
+- **Files & shell** — `Read` (with a structure index for huge files), `Write`,
+  `Edit`, `Bash`, `Glob`, `Grep`.
+- **Version control** — `GitStatus`, `GitDiff`, `GitLog`, `GitCommit`
+  (structured + truncated; commit is local and reversible).
+- **Internet** — `WebSearch` + `WebFetch` (DuckDuckGo; offline-safe).
+- **Knowledge base (GraphRAG)** — build a local entity-graph index from your
+  docs/code with `/graphrag build <path>`; the agent retrieves from it via the
+  read-only `Knowledge` tool.
+- **Multi-agent** — `Dispatch` runs several read-only sub-agents in parallel;
+  `task` runs one — each in a fresh context, for focused investigation.
+- **MCP** — connect to Model Context Protocol servers (`~/.drydock/mcp.json`);
+  their tools appear as `mcp__<server>__<tool>`. List them with `/mcp`.
+- **Skills** — reusable `/<name>` commands authored as markdown in
+  `~/.drydock/skills/` (or `<project>/.drydock/skills/`); `$ARGS` substitution.
+- **Loops** — `/loop <count> <prompt>` runs a prompt iteratively (Esc stops).
 
 ## Install
 
@@ -55,7 +78,7 @@ pip install drydock-cli
 drydock
 ```
 
-Requires Python 3.12+. From source instead:
+Requires Python 3.11+. From source instead:
 `git clone https://github.com/fbobe321/drydock-v3.git && cd drydock-v3 && pip install -e .`
 
 On first launch with no config, Drydock probes localhost for a running local
@@ -76,8 +99,11 @@ commands to do the work, showing each as a collapsible tool card.
 - A live activity line shows progress while it works:
   `◡ Keelhauling…  (12s · ↓ 6.2k tokens · thinking with high effort)`
 - Submit while it's working and the prompt **queues** (drains in order)
-- Slash commands: `/model [name]` · `/cwd [path]` · `/undo` (revert last write)
-  · `/back` (rewind last turn) · `/status` · `/clear` · `/help` · `/quit`
+- Slash commands: `/model` · `/cwd` · `/undo` (revert last write) · `/back`
+  (rewind last turn) · `/status` · `/compact` (shrink context) · `/graphrag`
+  (build/query a knowledge base) · `/skills` (list your `/<name>` skills) ·
+  `/loop` (repeat a prompt) · `/mcp` (list MCP servers) · `/clear` · `/help` ·
+  `/quit`
 
 It honors `AGENTS.md` / `DRYDOCK.md` in the working directory for project
 conventions.
@@ -96,22 +122,26 @@ blocked:
   `raise` outside an except, and refuses to write git conflict-marker content.
 
 Point it at a local OpenAI-compatible endpoint (e.g. llama.cpp's `server-cuda`
-serving Gemma 4 26B).
+serving Gemma-4-31B). The web tools (`WebSearch`/`WebFetch`) are read-only and
+degrade cleanly offline; the release scanner allowlists only the search backend.
 
 ## Model server (reference setup)
 
 Drydock is provider-agnostic, but it's tuned and measured against this rig:
 
-- **Model:** Gemma-4-26B-A4B-it (26B MoE, ~4B active/token), Unsloth `Q3_K_M`
-  GGUF, served by `ghcr.io/ggml-org/llama.cpp:server-cuda` with `--jinja`.
-- **GPUs:** 2× NVIDIA RTX 4060 Ti 16GB. The Q3_K_M weights (~13 GB) fit on a
-  **single 16 GB card**, so each GPU runs a **full, independent instance** —
-  two cards give two parallel instances for throughput; the model is **not**
-  tensor-split or "pooled" across both.
+- **Model:** dense **Gemma-4-31B** (QAT `Q4_K_XL` GGUF), served by
+  `ghcr.io/ggml-org/llama.cpp:server-cuda` with `--jinja`. Swapped from the
+  26B-A4B MoE, whose ~4B active params caused fatal agentic tool-loops; the
+  dense 31B is loop-free (slower, but it finishes).
+- **GPUs:** 2× NVIDIA RTX 4060 Ti 16GB, **tensor-split** across both cards
+  (`--tensor-split 1,1`) so the 31B weights fit.
 - **Context:** 64k (`-c 65536`) with `q8_0` KV-cache quantization
-  (`-ctk q8_0 -ctv q8_0`) — 64k @ q8 fits roughly the same VRAM as 32k @ f16.
-- **Throughput:** ~**64 tok/s** decode, ~94 tok/s prompt (per instance, Q3_K_M).
-- **Minimum:** any single 16 GB+ CUDA card runs it.
+  (`-ctk q8_0 -ctv q8_0`); set `context_limit` in `~/.drydock/config.toml` to
+  match your server's `-c`.
+- **Throughput:** ~15 tok/s decode (tensor-split 31B). Faster single-GPU
+  options exist if you drop to a smaller model.
+- **Provider-agnostic:** any OpenAI-compatible endpoint (llama.cpp, vLLM,
+  Ollama, LM Studio) works — point `--base-url` at it.
 
 ## Principles
 

{drydock_cli-3.0.56 → drydock_cli-3.0.58}/drydock_cli.egg-info/SOURCES.txt

@@ -34,6 +34,7 @@ drydock_cli.egg-info/requires.txt
 drydock_cli.egg-info/top_level.txt
 tests/test_approval.py
 tests/test_back_command.py
+tests/test_bash_output_bounding.py
 tests/test_bash_process_group.py
 tests/test_bash_safety.py
 tests/test_bash_timeout_network.py

{drydock_cli-3.0.56 → drydock_cli-3.0.58}/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 = "3.0.56"
+version = "3.0.58"
 description = "Drydock — a local, provider-agnostic terminal coding agent for local LLMs"
 readme = "README.md"
 requires-python = ">=3.11"

drydock_cli-3.0.58/tests/test_bash_output_bounding.py

@@ -0,0 +1,37 @@
+"""Critical VRAM/RAM protection (PRD §3.B): tool_bash must bound a command's
+output DURING capture so a runaway/infinite-output command can't balloon memory
+before the context-truncation cap applies."""
+from __future__ import annotations
+
+import time
+
+from drydock.tools import tool_bash, _MAX_BASH_OUTPUT_BYTES
+
+
+def test_infinite_output_is_capped_and_fast(tmp_path):
+    cfg = {"cwd": str(tmp_path)}
+    t0 = time.monotonic()
+    out = tool_bash({"command": "yes"}, cfg)   # infinite stream
+    elapsed = time.monotonic() - t0
+    assert "truncated at" in out                      # capped, not dumped
+    assert len(out) < _MAX_BASH_OUTPUT_BYTES + 500     # bounded memory
+    assert elapsed < 20                                # killed promptly, no hang
+
+
+def test_large_finite_output_is_capped(tmp_path):
+    cfg = {"cwd": str(tmp_path)}
+    # ~1 MB of output → must come back truncated near the cap, not 1 MB.
+    out = tool_bash({"command": "head -c 1000000 /dev/zero | tr '\\0' 'a'"}, cfg)
+    assert len(out) < _MAX_BASH_OUTPUT_BYTES + 500
+
+
+def test_normal_output_untouched(tmp_path):
+    cfg = {"cwd": str(tmp_path)}
+    out = tool_bash({"command": "echo hello world"}, cfg)
+    assert out == "hello world" and "truncated" not in out
+
+
+def test_exit_code_still_reported(tmp_path):
+    cfg = {"cwd": str(tmp_path)}
+    out = tool_bash({"command": "echo oops; exit 3"}, cfg)
+    assert "oops" in out and "exit code: 3" in out

{drydock_cli-3.0.56 → drydock_cli-3.0.58}/tests/test_failure_loop.py

@@ -75,3 +75,26 @@ def test_polling_with_changing_result_not_tripped(monkeypatch):
     st = AgentState()
     list(run("go", st, {"model": "m", "cwd": tempfile.mkdtemp()}, "sys"))
     assert any(m.get("content") == "done" for m in st.messages)
+
+
+def test_infinite_distinct_tool_calls_terminate_at_max_turns(monkeypatch):
+    """PRD §3.D: a model that keeps calling tools (DIFFERENT, succeeding calls)
+    and never yields 'Task Complete' must be forcibly stopped at the hard
+    max-iteration ceiling — protecting VRAM/context from an unbounded loop."""
+    calls = {"n": 0}
+
+    def never_done(**kw):
+        calls["n"] += 1
+        # A unique, succeeding command each turn → the identical-repeat valve
+        # never trips, so only the hard max_turns cap can end it.
+        return iter([AssistantTurn(
+            "", [{"id": str(calls["n"]), "name": "Bash",
+                  "input": {"command": f"echo step {calls['n']}"}}], 1, 1)])
+
+    monkeypatch.setattr(agent_mod, "stream", never_done)
+    st = AgentState()
+    import tempfile
+    # Low cap so the test is fast; the real default is 200.
+    list(run("loop forever", st, {"model": "m", "max_turns": 12,
+                                  "cwd": tempfile.mkdtemp()}, "sys"))
+    assert calls["n"] == 12  # ran exactly to the ceiling, then terminated