drydock-cli 0.7.3__tar.gz → 0.8.0__tar.gz

{drydock_cli-0.7.3 → drydock_cli-0.8.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: drydock-cli
-Version: 0.7.3
+Version: 0.8.0
 Summary: Drydock — nautical CLI coding agent
 Project-URL: Homepage, https://github.com/fbobe321/drydock
 Project-URL: Repository, https://github.com/fbobe321/drydock
@@ -24,6 +24,7 @@ Classifier: Topic :: Utilities
 Requires-Python: >=3.12
 Requires-Dist: agent-client-protocol==0.8.1
 Requires-Dist: anyio>=4.12.0
+Requires-Dist: beautifulsoup4>=4.12.0
 Requires-Dist: cachetools>=5.5.0
 Requires-Dist: cryptography<=46.0.3,>=44.0.0
 Requires-Dist: gitpython>=3.1.46

{drydock_cli-0.7.3 → drydock_cli-0.8.0}/drydock/cli/cli.py

@@ -196,7 +196,7 @@ def run_cli(args: argparse.Namespace) -> None:
                 entrypoint_metadata=EntrypointMetadata(
                     agent_entrypoint="cli",
                     agent_version=__version__,
-                    client_name="vibe_cli",
+                    client_name="drydock_cli",
                     client_version=__version__,
                 ),
             )

{drydock_cli-0.7.3 → drydock_cli-0.8.0}/drydock/cli/textual_ui/external_editor.py

@@ -16,7 +16,7 @@ class ExternalEditor:
 
     def edit(self, initial_content: str = "") -> str | None:
         editor = self.get_editor()
-        fd, filepath = tempfile.mkstemp(suffix=".md", prefix="vibe_")
+        fd, filepath = tempfile.mkstemp(suffix=".md", prefix="drydock_")
         try:
             with os.fdopen(fd, "w") as f:
                 f.write(initial_content)

{drydock_cli-0.7.3 → drydock_cli-0.8.0}/drydock/core/agent_loop.py

@@ -482,6 +482,7 @@ class AgentLoop:
             bash_count = 0  # Track consecutive bash calls without an edit
             search_replace_failures = 0  # Track failed search_replace attempts
             files_explored: list[str] = []  # Track files read for context state
+            files_modified: list[str] = []  # Track files edited for blast radius
             context_summary_injected = False
             while not should_break_loop:
                 # Loop protection: prevent infinite tool-call loops
@@ -551,6 +552,14 @@ class AgentLoop:
                                 if tc.function.name in ("search_replace", "write_file"):
                                     has_made_edit = True
                                     bash_count = 0  # Reset on successful edit
+                                    # Track blast radius
+                                    try:
+                                        edit_args = json.loads(tc.function.arguments or "{}")
+                                        edit_path = edit_args.get("file_path", edit_args.get("path", ""))
+                                        if edit_path and edit_path not in files_modified:
+                                            files_modified.append(edit_path)
+                                    except (json.JSONDecodeError, AttributeError):
+                                        pass
                                 if tc.function.name == "read_file":
                                     try:
                                         args = json.loads(tc.function.arguments or "{}")
@@ -569,6 +578,15 @@ class AgentLoop:
                             if "search_replace" in content.lower() and "error" in content.lower():
                                 search_replace_failures += 1
 
+                # Blast radius check: warn when touching 5+ files
+                if len(files_modified) >= 5 and len(files_modified) % 5 == 0:
+                    self._inject_system_note(
+                        f"BLAST RADIUS WARNING: You have modified {len(files_modified)} files. "
+                        f"Files: {', '.join(files_modified[-5:])}. "
+                        f"This is a large change. Are you sure all these edits are needed? "
+                        f"Consider stopping to verify your approach."
+                    )
+
                 # Bash abuse detection: if model keeps using bash instead of
                 # search_replace/read_file, redirect it to the proper tools
                 if not has_made_edit and bash_count >= 10 and bash_count % 5 == 0:
@@ -579,14 +597,26 @@ class AgentLoop:
                         "Call search_replace NOW with your fix."
                     )
 
-                # search_replace keeps failing — suggest re-reading the file
-                if search_replace_failures >= 3 and search_replace_failures % 2 == 1:
-                    self._inject_system_note(
-                        f"Your search_replace has failed {search_replace_failures} times. "
-                        "The text you're searching for doesn't match the file. STOP guessing. "
-                        "Use read_file with offset/limit to see the EXACT current text, "
-                        "then copy it precisely into search_replace old_str."
-                    )
+                # 3-Strike Rule: search_replace keeps failing
+                if search_replace_failures >= 3:
+                    if search_replace_failures == 3:
+                        # Strike 3: STOP and ask user
+                        self.messages.append(LLMMessage(
+                            role=Role.user,
+                            content=(
+                                f"3-STRIKE RULE: Your search_replace has failed {search_replace_failures} times. "
+                                "You have exhausted your attempts. STOP trying the same approach. "
+                                "Either: (1) use read_file to get the EXACT text first, "
+                                "(2) try a completely different file, or "
+                                "(3) ask the user for help with /consult."
+                            )
+                        ))
+                    elif search_replace_failures % 3 == 0:
+                        # Every 3 more failures, escalate
+                        self.messages.append(LLMMessage(
+                            role=Role.user,
+                            content=f"STOP: {search_replace_failures} failed edits. You are stuck. Ask the user for help."
+                        ))
 
                 # Context budget warning: after 7 tool turns without an edit,
                 # warn that context is being consumed without progress
@@ -647,7 +677,11 @@ class AgentLoop:
                         should_break_loop = True
                         nudge_text = None
                     if nudge_text:
-                        self._inject_system_note(nudge_text)
+                        # Inject as user message so the model actually sees it
+                        # (inject_system_note buries it in old tool results)
+                        self.messages.append(
+                            LLMMessage(role=Role.user, content=nudge_text)
+                        )
                         logger.info("Model gave text without editing — nudging (attempt %d)", text_without_action)
 
                 # If model has been investigating for too long without making an edit,

{drydock_cli-0.7.3 → drydock_cli-0.8.0}/drydock/core/auth/github.py

@@ -11,7 +11,7 @@ import keyring.errors
 
 GITHUB_CLIENT_ID = "Ov23liJ7sk5kFDMEyvDT"
 
-_SERVICE_NAME = "vibe"
+_SERVICE_NAME = "drydock"
 _KEYRING_USERNAME = "github_token"
 _DEVICE_CODE_URL = "https://github.com/login/device/code"
 _TOKEN_URL = "https://github.com/login/oauth/access_token"

{drydock_cli-0.7.3 → drydock_cli-0.8.0}/drydock/core/config/_settings.py

@@ -267,7 +267,7 @@ DEFAULT_PROVIDERS = [
 
 DEFAULT_MODELS = [
     ModelConfig(
-        name="mistral-vibe-cli-latest",
+        name="drydock-cli-latest",
         provider="mistral",
         alias="devstral-2",
         input_price=0.4,

{drydock_cli-0.7.3 → drydock_cli-0.8.0}/drydock/core/logger.py

@@ -9,7 +9,7 @@ from drydock.core.paths import LOG_DIR, LOG_FILE
 
 LOG_DIR.path.mkdir(parents=True, exist_ok=True)
 
-logger = logging.getLogger("vibe")
+logger = logging.getLogger("drydock")
 
 
 class StructuredLogFormatter(logging.Formatter):

{drydock_cli-0.7.3 → drydock_cli-0.8.0}/drydock/core/programmatic.py

@@ -21,7 +21,7 @@ from drydock.core.types import (
 )
 from drydock.core.utils import ConversationLimitException
 
-_DEFAULT_CLIENT_METADATA = ClientMetadata(name="vibe_programmatic", version=__version__)
+_DEFAULT_CLIENT_METADATA = ClientMetadata(name="drydock_programmatic", version=__version__)
 
 
 def run_programmatic(

{drydock_cli-0.7.3 → drydock_cli-0.8.0}/drydock/core/tools/builtins/search_replace.py

@@ -119,12 +119,15 @@ class SearchReplace(
     ) -> AsyncGenerator[ToolStreamEvent | SearchReplaceResult, None]:
         file_path, search_replace_blocks = self._prepare_and_validate_args(args)
 
-        # Warn (but don't block) when editing test files
+        # BLOCK editing test files — the bug is in source code, not tests
         file_str = str(file_path)
         if ("/tests/" in file_str or "/test_" in file_str or
                 file_str.endswith("_test.py") or "/testing/" in file_str):
-            import logging
-            logging.getLogger(__name__).warning("search_replace targeting test file: %s", file_str)
+            raise ToolError(
+                f"BLOCKED: You are trying to edit a test file ({file_path.name}). "
+                f"The bug is in LIBRARY SOURCE code, not tests. "
+                f"Use grep to find the corresponding source file and edit that instead."
+            )
 
         # Injection guard: scan replacement content for suspicious patterns
         from drydock.core.tools.injection_guard import check_content_for_injection

drydock_cli-0.8.0/drydock/core/tools/builtins/websearch.py

@@ -0,0 +1,133 @@
+from __future__ import annotations
+
+from collections.abc import AsyncGenerator
+import os
+import urllib.parse
+from typing import ClassVar, final
+
+from pydantic import BaseModel, Field
+from bs4 import BeautifulSoup
+
+from drydock.core.tools.base import (
+    BaseTool,
+    BaseToolConfig,
+    BaseToolState,
+    InvokeContext,
+    ToolError,
+    ToolPermission,
+)
+from drydock.core.tools.ui import ToolCallDisplay, ToolResultDisplay, ToolUIData
+from drydock.core.types import ToolStreamEvent
+
+
+class WebSearchSource(BaseModel):
+    title: str
+    url: str
+
+
+class WebSearchArgs(BaseModel):
+    query: str = Field(min_length=1)
+
+
+class WebSearchResult(BaseModel):
+    answer: str
+    sources: list[WebSearchSource] = Field(default_factory=list)
+
+
+class WebSearchConfig(BaseToolConfig):
+    permission: ToolPermission = ToolPermission.ASK
+    timeout: int = Field(default=30, description="HTTP timeout in seconds.")
+
+
+class WebSearch(
+    BaseTool[WebSearchArgs, WebSearchResult, WebSearchConfig, BaseToolState],
+    ToolUIData[WebSearchArgs, WebSearchResult],
+):
+    description: ClassVar[str] = "Search the web directly without an API key to find current information."
+
+    @classmethod
+    def is_available(cls) -> bool:
+        # Always available! No API key required.
+        return True
+
+    @final
+    async def run(
+        self, args: WebSearchArgs, ctx: InvokeContext | None = None
+    ) -> AsyncGenerator[ToolStreamEvent | WebSearchResult, None]:
+        
+        import httpx
+
+        # Grab your proxy settings
+        proxy_url = os.getenv("HTTPS_PROXY") or os.getenv("https_proxy") or os.getenv("HTTP_PROXY") or os.getenv("http_proxy")
+        
+        headers = {
+            "User-Agent": "Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/120.0.0.0 Safari/537.36",
+            "Accept": "text/html,application/xhtml+xml,application/xml;q=0.9,*/*;q=0.8",
+        }
+
+        try:
+            # Respect DRYDOCK_INSECURE flag for SSL bypass
+            verify_ssl = os.environ.get("DRYDOCK_INSECURE") != "1"
+            async with httpx.AsyncClient(proxy=proxy_url, verify=not verify_ssl if verify_ssl else False, headers=headers) as client:
+                response = await client.post(
+                    "https://html.duckduckgo.com/html/",
+                    data={"q": args.query},
+                    timeout=self.config.timeout
+                )
+                response.raise_for_status()
+        except Exception as e:
+            raise ToolError(f"Failed to fetch search results: {str(e)}")
+
+        # Parse the DuckDuckGo HTML results
+        soup = BeautifulSoup(response.text, "html.parser")
+        results = soup.find_all("div", class_="result")
+
+        text_parts = []
+        sources = []
+
+        # Extract the top 5 results
+        for res in results[:5]:
+            title_tag = res.find("a", class_="result__url")
+            snippet_tag = res.find("a", class_="result__snippet")
+            
+            if title_tag and snippet_tag:
+                title = title_tag.text.strip()
+                snippet = snippet_tag.text.strip()
+                raw_url = title_tag.get("href", "")
+                
+                # Clean up DuckDuckGo's redirect URLs
+                if "uddg=" in raw_url:
+                    clean_url = urllib.parse.unquote(raw_url.split("uddg=")[1].split("&")[0])
+                else:
+                    clean_url = raw_url
+
+                sources.append(WebSearchSource(title=title, url=clean_url))
+                text_parts.append(f"Source: {title}\nURL: {clean_url}\nInformation: {snippet}\n")
+
+        if not text_parts:
+            answer = "No relevant information found on the web."
+        else:
+            # We bundle the snippets into a single text block for your local model to read
+            answer = "Here is the raw information retrieved from the web:\n\n" + "\n".join(text_parts)
+
+        yield WebSearchResult(answer=answer, sources=sources)
+
+    @classmethod
+    def get_call_display(cls, event: ToolCallEvent) -> ToolCallDisplay:
+        if event.args is None or not isinstance(event.args, WebSearchArgs):
+            return ToolCallDisplay(summary="websearch")
+        return ToolCallDisplay(summary=f"Searching DuckDuckGo for: '{event.args.query}'")
+
+    @classmethod
+    def get_result_display(cls, event: ToolResultEvent) -> ToolResultDisplay:
+        if not isinstance(event.result, WebSearchResult):
+            return ToolResultDisplay(
+                success=False, message=event.error or event.skip_reason or "No result"
+            )
+        return ToolResultDisplay(
+            success=True, message=f"{len(event.result.sources)} web sources retrieved for the local model."
+        )
+
+    @classmethod
+    def get_status_text(cls) -> str:
+        return "Scraping the web..."

drydock_cli-0.8.0/drydock/skills/investigate/SKILL.md

@@ -0,0 +1,67 @@
+---
+name: investigate
+description: Systematic debugging with 3-strike rule. Investigate → Analyze → Hypothesize → Fix. Stops after 3 failed hypotheses.
+user-invocable: true
+allowed-tools:
+  - bash
+  - grep
+  - read_file
+  - search_replace
+  - write_file
+  - ask_user_question
+---
+
+# Investigate
+
+Systematic debugging workflow. You are a senior debugger.
+
+## Iron Law
+
+**Never fix without root cause.** If you don't understand WHY the bug happens, you don't fix it.
+
+## Workflow
+
+### Phase 1: INVESTIGATE
+- Read the error message / bug report carefully
+- Identify the module and function where the bug manifests
+- grep for the relevant code paths
+- Read the specific functions involved
+
+### Phase 2: ANALYZE
+- Trace the data flow from input to the point of failure
+- Identify what value is wrong and where it diverges from expected
+- Check edge cases: null, empty, boundary values
+
+### Phase 3: HYPOTHESIZE
+State your hypothesis clearly:
+```
+HYPOTHESIS: The bug is caused by [X] in [file:line] because [reason].
+EXPECTED: [what should happen]
+ACTUAL: [what happens instead]
+FIX: [what change to make]
+```
+
+### Phase 4: IMPLEMENT
+- Make the minimal fix
+- Read back the changed code to verify
+- If the fix doesn't work, go back to Phase 3
+
+## 3-Strike Rule
+
+Track your failed hypotheses:
+- Strike 1: "Hypothesis 1 failed because [reason]. Trying different approach."
+- Strike 2: "Hypothesis 2 failed because [reason]. One more attempt."
+- Strike 3: **STOP.** Ask the user: "I've tried 3 approaches and none worked. Here's what I've learned: [summary]. What should I try next?"
+
+Do NOT continue past 3 strikes. Ask for help.
+
+## Scope Lock
+
+Once you identify the module, **stay in that module**. Do not wander to other parts of the codebase unless the bug clearly crosses module boundaries.
+
+## Blast Radius Check
+
+Before applying a fix, count the files you're about to change:
+- 1-2 files: proceed
+- 3-4 files: mention it to the user first
+- 5+ files: **STOP and ask.** "This fix touches {N} files. Should I proceed?"

drydock_cli-0.8.0/drydock/skills/review/SKILL.md

@@ -0,0 +1,67 @@
+---
+name: review
+description: Automated code review. Two-pass (critical then informational), scope drift detection, fix-first.
+user-invocable: true
+allowed-tools:
+  - bash
+  - grep
+  - read_file
+  - search_replace
+---
+
+# Code Review
+
+You are a senior code reviewer. Review the current changes.
+
+## Two-Pass Review
+
+### Pass 1: CRITICAL (must fix before merge)
+- Security vulnerabilities (injection, XSS, auth bypass)
+- Data loss risks (missing transactions, race conditions)
+- Breaking changes to public APIs
+- Missing error handling that could crash
+- Tests missing for new functionality
+
+### Pass 2: INFORMATIONAL (nice to have)
+- Code style and naming
+- Performance suggestions
+- Refactoring opportunities
+- Documentation gaps
+
+## Workflow
+
+1. Run `git diff` to see what changed
+2. Read each changed file
+3. For each issue found:
+   - **Critical**: Fix it immediately with search_replace
+   - **Judgment call**: Ask the user with options
+   - **Informational**: Note it but don't block
+
+## Scope Drift Detection
+
+If the diff contains changes unrelated to the stated goal:
+- Flag: "These files seem unrelated to the task: [list]"
+- Ask: "Should these be in a separate commit?"
+
+## Fix-First Philosophy
+
+- Mechanical issues (typos, missing imports, obvious bugs): fix silently
+- Style issues: fix if trivial, note if subjective
+- Architecture decisions: always ask
+
+## Output Format
+
+```
+## Review Summary
+
+### Critical (must fix)
+- [ ] [file:line] Issue description
+
+### Informational
+- [file:line] Suggestion
+
+### Scope
+- Files changed: N
+- Lines added: N, removed: N
+- Test coverage: [assessed/not assessed]
+```

drydock_cli-0.8.0/drydock/skills/ship/SKILL.md

@@ -0,0 +1,65 @@
+---
+name: ship
+description: Automated shipping pipeline. Test → Review → Commit → Push → PR. Non-interactive.
+user-invocable: true
+allowed-tools:
+  - bash
+  - grep
+  - read_file
+  - write_file
+  - search_replace
+---
+
+# Ship
+
+Automated shipping pipeline. You are a release engineer.
+
+## Verification Gate
+
+**Never claim completion without fresh test evidence.**
+
+## Pipeline
+
+### Step 1: Pre-flight checks
+```bash
+git status
+git diff --stat
+```
+- If working tree is dirty, ask what to include
+- If on main/master, ask to create a branch first
+
+### Step 2: Run tests
+```bash
+python3 -m pytest tests/ -q --tb=short 2>&1 | tail -20
+```
+- If tests fail: **STOP.** Fix the failures first.
+- If no test runner found, note it and continue.
+
+### Step 3: Review diff
+- Run `/review` mentally (check for critical issues)
+- If >5 files changed, confirm with user
+
+### Step 4: Commit
+- Write a clear commit message following conventional commits:
+  - `feat:` for new features
+  - `fix:` for bug fixes
+  - `refactor:` for code changes that aren't fixes or features
+  - `docs:` for documentation
+  - `test:` for test changes
+- Each commit should be bisectable (tests pass at every commit)
+
+### Step 5: Push and PR
+```bash
+git push origin HEAD
+```
+- If `gh` CLI is available, create a PR:
+```bash
+gh pr create --title "..." --body "..."
+```
+
+## Rules
+
+- Never force push to main/master
+- Never skip tests
+- Never commit secrets, tokens, or credentials
+- Always include what changed and why in the commit message

{drydock_cli-0.7.3 → drydock_cli-0.8.0}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "drydock-cli"
-version = "0.7.3"
+version = "0.8.0"
 description = "Drydock — nautical CLI coding agent"
 readme = "README.md"
 requires-python = ">=3.12"
@@ -30,6 +30,7 @@ classifiers = [
 dependencies = [
     "agent-client-protocol==0.8.1",
     "anyio>=4.12.0",
+    "beautifulsoup4>=4.12.0",
     "cachetools>=5.5.0",
     "cryptography>=44.0.0,<=46.0.3",
     "gitpython>=3.1.46",
@@ -77,6 +78,9 @@ include = ["drydock/"]
 [tool.hatch.build.targets.wheel.force-include]
 "drydock/skills/create-presentation" = "drydock/skills/create-presentation"
 "drydock/skills/deep-research" = "drydock/skills/deep-research"
+"drydock/skills/investigate" = "drydock/skills/investigate"
+"drydock/skills/review" = "drydock/skills/review"
+"drydock/skills/ship" = "drydock/skills/ship"
 
 
 [project.scripts]

drydock_cli-0.8.0/scripts/backup.sh

@@ -0,0 +1,84 @@
+#!/bin/bash
+# Backup DryDock project to NAS
+# Cron: 0 3 * * * /data3/drydock/scripts/backup.sh >> /data3/drydock/logs/backup.log 2>&1
+
+set -euo pipefail
+
+DRYDOCK_SRC="/data3/drydock"
+BENCH_DIR="/data3/swe_bench_runs"
+NAS_MOUNT="/mnt/nas_backups"
+BACKUP_DIR="$NAS_MOUNT/drydock"
+DATE=$(date +%Y%m%d)
+LOGDIR="$DRYDOCK_SRC/logs"
+
+mkdir -p "$LOGDIR"
+
+log() { echo "[$(date '+%Y-%m-%d %H:%M:%S')] $*"; }
+
+# Check NAS is mounted
+if ! mountpoint -q "$NAS_MOUNT" 2>/dev/null; then
+    log "NAS not mounted. Attempting mount..."
+    sudo mount "$NAS_MOUNT" 2>/dev/null || {
+        log "ERROR: Cannot mount NAS at $NAS_MOUNT. Skipping backup."
+        exit 1
+    }
+fi
+
+mkdir -p "$BACKUP_DIR"
+
+log "Starting backup..."
+
+# Backup 1: DryDock source code (excluding __pycache__, .git objects)
+rsync -rlpt --delete --no-group --no-owner \
+    --exclude='__pycache__/' \
+    --exclude='.pytest_cache/' \
+    --exclude='dist/' \
+    --exclude='*.egg-info/' \
+    --exclude='.git/objects/' \
+    --exclude='logs/' \
+    "$DRYDOCK_SRC/" "$BACKUP_DIR/source/"
+
+log "Source backed up."
+
+# Backup 2: SWE-bench results and state (not the repos — too large)
+rsync -rlpt --delete --no-group --no-owner \
+    --exclude='repos/' \
+    --exclude='tasks/*/repo/' \
+    --exclude='__pycache__/' \
+    "$BENCH_DIR/continuous_bench_state.json" \
+    "$BENCH_DIR/auto_improve_state.json" \
+    "$BENCH_DIR/harness.py" \
+    "$BENCH_DIR/continuous_bench.sh" \
+    "$BENCH_DIR/analyze_batch.py" \
+    "$BACKUP_DIR/bench/" 2>/dev/null || true
+
+# Backup results directories (small — just JSON)
+rsync -rlpt --no-group --no-owner \
+    --include='*/' \
+    --include='results.json' \
+    --exclude='*' \
+    "$BENCH_DIR/results/" "$BACKUP_DIR/bench/results/" 2>/dev/null || true
+
+log "Bench state backed up."
+
+# Backup 3: Config files
+mkdir -p "$BACKUP_DIR/config/drydock_config" "$BACKUP_DIR/config/drydock_home"
+rsync -rlpt --no-group --no-owner "$HOME/.config/drydock/" "$BACKUP_DIR/config/drydock_config/" 2>/dev/null || true
+rsync -rlpt --no-group --no-owner --exclude='logs/' "$HOME/.drydock/" "$BACKUP_DIR/config/drydock_home/" 2>/dev/null || true
+
+log "Config backed up."
+
+# Backup 4: Crontab
+crontab -l > "$BACKUP_DIR/crontab_$(hostname).txt" 2>/dev/null || true
+
+# Write backup manifest
+cat > "$BACKUP_DIR/MANIFEST.txt" << EOF
+DryDock Backup
+Date: $(date)
+Host: $(hostname)
+Source: $DRYDOCK_SRC
+Bench: $BENCH_DIR
+Version: $(python3 -c "import sys; sys.path.insert(0,'$DRYDOCK_SRC'); from drydock import __version__; print(__version__)" 2>/dev/null || echo "unknown")
+EOF
+
+log "Backup complete to $BACKUP_DIR"