devagent-ai 0.8.3__tar.gz → 0.8.4__tar.gz

{devagent_ai-0.8.3/devagent_ai.egg-info → devagent_ai-0.8.4}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: devagent-ai
-Version: 0.8.3
+Version: 0.8.4
 Summary: Evidence-driven local autonomous software engineering agent
 Author: Tom Ha
 Maintainer: Tom Ha
@@ -152,6 +152,39 @@ reviewer     → independently review final diff
 
 The deterministic harness remains responsible for safety, tool execution, verification validity, acceptance adjudication, final status, reporting, and source-control publication regardless of which model handles a role.
 
+#### Example: use multiple AI models in one DevAgent run
+
+You do not have to use one AI model for every reasoning step. If you believe different models are better suited to different engineering roles, configure a default model plus any role-specific overrides. Roles that are not explicitly configured fall back to the default model.
+
+```bash
+# Keep credentials in environment variables; DevAgent does not store the keys.
+export OPENAI_API_KEY=...
+export GEMINI_API_KEY=...
+export ANTHROPIC_API_KEY=...
+export XAI_API_KEY=...
+
+# Default/fallback model.
+devagent setup --provider openai --model YOUR_OPENAI_MODEL
+
+# Optional per-role models.
+devagent setup --role investigator --provider gemini --model YOUR_GEMINI_MODEL
+devagent setup --role planner --provider anthropic --model YOUR_CLAUDE_MODEL
+devagent setup --role implementer --provider openai --model YOUR_OPENAI_MODEL
+devagent setup --role reviewer --provider xai --model YOUR_GROK_MODEL
+
+# Inspect routing and optionally probe every configured cloud model.
+devagent models
+devagent doctor --live
+
+# Run normally; saved role routing is applied automatically.
+cd my-repo
+devagent "Fix the checkout race condition and add regression coverage."
+```
+
+For example, a user may choose a fast or lower-cost model for repository investigation, a different model for planning, a preferred coding model for implementation, and another provider for independent review. This can be useful for cost, latency, provider diversity, or model-strength preferences, but it does not guarantee a better result. DevAgent still requires the same repository evidence, acceptance gates, deterministic verification, and publication rules.
+
+When using saved role routing, run the task without run-level `--provider`, `--model`, or `--base-url` overrides. Supplying those flags explicitly selects one provider/model for that run instead of the saved per-role routing.
+
 ## Run
 
 From the application repository:

{devagent_ai-0.8.3 → devagent_ai-0.8.4}/README.md

@@ -121,6 +121,39 @@ reviewer     → independently review final diff
 
 The deterministic harness remains responsible for safety, tool execution, verification validity, acceptance adjudication, final status, reporting, and source-control publication regardless of which model handles a role.
 
+#### Example: use multiple AI models in one DevAgent run
+
+You do not have to use one AI model for every reasoning step. If you believe different models are better suited to different engineering roles, configure a default model plus any role-specific overrides. Roles that are not explicitly configured fall back to the default model.
+
+```bash
+# Keep credentials in environment variables; DevAgent does not store the keys.
+export OPENAI_API_KEY=...
+export GEMINI_API_KEY=...
+export ANTHROPIC_API_KEY=...
+export XAI_API_KEY=...
+
+# Default/fallback model.
+devagent setup --provider openai --model YOUR_OPENAI_MODEL
+
+# Optional per-role models.
+devagent setup --role investigator --provider gemini --model YOUR_GEMINI_MODEL
+devagent setup --role planner --provider anthropic --model YOUR_CLAUDE_MODEL
+devagent setup --role implementer --provider openai --model YOUR_OPENAI_MODEL
+devagent setup --role reviewer --provider xai --model YOUR_GROK_MODEL
+
+# Inspect routing and optionally probe every configured cloud model.
+devagent models
+devagent doctor --live
+
+# Run normally; saved role routing is applied automatically.
+cd my-repo
+devagent "Fix the checkout race condition and add regression coverage."
+```
+
+For example, a user may choose a fast or lower-cost model for repository investigation, a different model for planning, a preferred coding model for implementation, and another provider for independent review. This can be useful for cost, latency, provider diversity, or model-strength preferences, but it does not guarantee a better result. DevAgent still requires the same repository evidence, acceptance gates, deterministic verification, and publication rules.
+
+When using saved role routing, run the task without run-level `--provider`, `--model`, or `--base-url` overrides. Supplying those flags explicitly selects one provider/model for that run instead of the saved per-role routing.
+
 ## Run
 
 From the application repository:

{devagent_ai-0.8.3 → devagent_ai-0.8.4}/devagent/__init__.py

@@ -1,3 +1,3 @@
 """DevAgent: evidence-driven local software engineering automation."""
 
-__version__ = "0.8.3"
+__version__ = "0.8.4"

{devagent_ai-0.8.3 → devagent_ai-0.8.4}/devagent/cli.py

@@ -4,10 +4,11 @@ import argparse
 import importlib.util
 import json
 import os
+import re
 import shutil
 import sys
 from pathlib import Path
-from typing import Sequence
+from typing import Callable, Sequence
 
 from devagent import __version__
 from devagent.config import (
@@ -37,6 +38,84 @@ _LIVE_PROBE_SCHEMA = {
     "additionalProperties": False,
 }
 
+_PROGRESS_STAGES: dict[str, tuple[int, str]] = {
+    "DISCOVER": (1, "DISCOVER / UNDERSTAND"),
+    "UNDERSTAND": (1, "DISCOVER / UNDERSTAND"),
+    "TASK_SPEC": (2, "REQUIREMENTS / PLAN"),
+    "BASELINE": (2, "REQUIREMENTS / PLAN"),
+    "PLAN": (2, "REQUIREMENTS / PLAN"),
+    "GATHER_CONTEXT": (2, "REQUIREMENTS / PLAN"),
+    "REPRODUCE": (2, "REQUIREMENTS / PLAN"),
+    "IMPLEMENT": (3, "IMPLEMENT"),
+    "VERIFY_TARGETED": (4, "VERIFY / REPAIR IF NEEDED"),
+    "VERIFY_BROAD": (4, "VERIFY / REPAIR IF NEEDED"),
+    "REVIEW": (5, "INDEPENDENT REVIEW"),
+    "QUALITY_CHECK": (6, "FINAL VERIFICATION"),
+    "FINAL_VERIFY": (6, "FINAL VERIFICATION"),
+}
+_STATE_LINE = re.compile(r"^\[([A-Z_]+)\]$")
+
+
+class _ProgressStatus:
+    """Render stable user-facing milestones while keeping full diagnostics opt-in."""
+
+    def __init__(
+        self,
+        sink: Callable[[str], None] = print,
+        *,
+        verbose: bool = False,
+    ) -> None:
+        self.sink = sink
+        self.verbose = verbose
+        self._last_stage = 0
+        self._plan_seen = False
+        self._implement_seen = False
+        self._review_seen = False
+
+    def __call__(self, message: str) -> None:
+        if self.verbose:
+            self.sink(message)
+            return
+
+        match = _STATE_LINE.fullmatch(message)
+        if match is None:
+            return
+        state = match.group(1)
+
+        if state == "DIAGNOSE":
+            self.sink("      ↳ DIAGNOSE")
+            return
+        if state == "PLAN":
+            if self._plan_seen:
+                self.sink("      ↳ REPLAN")
+                return
+            self._plan_seen = True
+        if state == "IMPLEMENT":
+            if self._implement_seen:
+                label = "APPLY REVIEW FIXES" if self._review_seen else "APPLY CORRECTION"
+                self.sink(f"      ↳ {label}")
+                return
+            self._implement_seen = True
+        if state == "REVIEW":
+            self._review_seen = True
+
+        stage = _PROGRESS_STAGES.get(state)
+        if stage is None:
+            return
+        number, label = stage
+        if number <= self._last_stage:
+            return
+        self._last_stage = number
+        self.sink(f"[{number}/7] {label}")
+
+    def report(self) -> None:
+        if self.verbose:
+            self.sink("[ENGINEERING_REPORT]")
+            return
+        if self._last_stage < 7:
+            self._last_stage = 7
+            self.sink("[7/7] ENGINEERING REPORT")
+
 
 def _top_parser() -> argparse.ArgumentParser:
     parser = argparse.ArgumentParser(prog="devagent", description="Evidence-driven local software engineering agent")
@@ -53,7 +132,11 @@ def _top_parser() -> argparse.ArgumentParser:
     parser.add_argument("--provider", choices=_PROVIDER_CHOICES)
     parser.add_argument("--model")
     parser.add_argument("--base-url")
-    parser.add_argument("--verbose", action="store_true", help="Show state transitions")
+    parser.add_argument(
+        "--verbose",
+        action="store_true",
+        help="Show internal state transitions and diagnostics instead of concise progress stages",
+    )
     parser.add_argument("--no-isolation", action="store_true", help="Work in place instead of creating a local detached worktree")
     parser.add_argument(
         "--publish",
@@ -337,11 +420,14 @@ def _run(argv: Sequence[str]) -> int:
         from devagent.orchestrator import DevAgent
         from devagent.report import recommendations_for, render_report
 
+        progress = _ProgressStatus(print, verbose=args.verbose)
         result = DevAgent(
             model_provider,
             isolate=not args.no_isolation,
-            verbose=args.verbose,
-            status=print,
+            # Internal status events are always emitted. The progress reporter keeps normal
+            # CLI output concise and passes the full state/diagnostic stream only in --verbose.
+            verbose=True,
+            status=progress,
             base_commit=publication_plan.base_commit if publication_plan else None,
         ).run(args.repo, requirement)
 
@@ -361,6 +447,7 @@ def _run(argv: Sequence[str]) -> int:
 
         # The full engineering report is intentionally emitted before any Git commit/push.
         result.recommendations = recommendations_for(result)
+        progress.report()
         print(render_report(result))
 
         if publish_requested:

{devagent_ai-0.8.3 → devagent_ai-0.8.4/devagent_ai.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: devagent-ai
-Version: 0.8.3
+Version: 0.8.4
 Summary: Evidence-driven local autonomous software engineering agent
 Author: Tom Ha
 Maintainer: Tom Ha
@@ -152,6 +152,39 @@ reviewer     → independently review final diff
 
 The deterministic harness remains responsible for safety, tool execution, verification validity, acceptance adjudication, final status, reporting, and source-control publication regardless of which model handles a role.
 
+#### Example: use multiple AI models in one DevAgent run
+
+You do not have to use one AI model for every reasoning step. If you believe different models are better suited to different engineering roles, configure a default model plus any role-specific overrides. Roles that are not explicitly configured fall back to the default model.
+
+```bash
+# Keep credentials in environment variables; DevAgent does not store the keys.
+export OPENAI_API_KEY=...
+export GEMINI_API_KEY=...
+export ANTHROPIC_API_KEY=...
+export XAI_API_KEY=...
+
+# Default/fallback model.
+devagent setup --provider openai --model YOUR_OPENAI_MODEL
+
+# Optional per-role models.
+devagent setup --role investigator --provider gemini --model YOUR_GEMINI_MODEL
+devagent setup --role planner --provider anthropic --model YOUR_CLAUDE_MODEL
+devagent setup --role implementer --provider openai --model YOUR_OPENAI_MODEL
+devagent setup --role reviewer --provider xai --model YOUR_GROK_MODEL
+
+# Inspect routing and optionally probe every configured cloud model.
+devagent models
+devagent doctor --live
+
+# Run normally; saved role routing is applied automatically.
+cd my-repo
+devagent "Fix the checkout race condition and add regression coverage."
+```
+
+For example, a user may choose a fast or lower-cost model for repository investigation, a different model for planning, a preferred coding model for implementation, and another provider for independent review. This can be useful for cost, latency, provider diversity, or model-strength preferences, but it does not guarantee a better result. DevAgent still requires the same repository evidence, acceptance gates, deterministic verification, and publication rules.
+
+When using saved role routing, run the task without run-level `--provider`, `--model`, or `--base-url` overrides. Supplying those flags explicitly selects one provider/model for that run instead of the saved per-role routing.
+
 ## Run
 
 From the application repository:

{devagent_ai-0.8.3 → devagent_ai-0.8.4}/devagent_ai.egg-info/SOURCES.txt

@@ -48,6 +48,7 @@ tests/test_benchmark_catalog.py
 tests/test_browser_verification.py
 tests/test_capability_discovery.py
 tests/test_cli.py
+tests/test_cli_progress.py
 tests/test_developer_review_report.py
 tests/test_discovery_memory.py
 tests/test_e2e_fake_provider.py

{devagent_ai-0.8.3 → devagent_ai-0.8.4}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "devagent-ai"
-version = "0.8.3"
+version = "0.8.4"
 description = "Evidence-driven local autonomous software engineering agent"
 readme = "README.md"
 requires-python = ">=3.10"

devagent_ai-0.8.4/tests/test_cli_progress.py

@@ -0,0 +1,91 @@
+from __future__ import annotations
+
+from devagent.cli import _ProgressStatus
+
+
+def test_default_progress_reports_only_stable_engineering_milestones() -> None:
+    output: list[str] = []
+    progress = _ProgressStatus(output.append)
+
+    events = [
+        "[PREFLIGHT]",
+        "[WORKTREE] source: /tmp/repo",
+        "[DISCOVER]",
+        "[DISCOVER] repository files: 120",
+        "[UNDERSTAND]",
+        "[RETRIEVAL] selected: src/service.py",
+        "[TASK_SPEC]",
+        "[BASELINE]",
+        "[PLAN]",
+        "[GATHER_CONTEXT]",
+        "[REPRODUCE]",
+        "[IMPLEMENT]",
+        "[VERIFY_TARGETED]",
+        "[DIAGNOSE]",
+        "[PLAN]",
+        "[GATHER_CONTEXT]",
+        "[IMPLEMENT]",
+        "[VERIFY_TARGETED]",
+        "[VERIFY_BROAD]",
+        "[REVIEW]",
+        "[QUALITY_CHECK]",
+        "[FINAL_VERIFY]",
+        "[REPORT]",
+    ]
+    for event in events:
+        progress(event)
+    progress.report()
+
+    assert output == [
+        "[1/7] DISCOVER / UNDERSTAND",
+        "[2/7] REQUIREMENTS / PLAN",
+        "[3/7] IMPLEMENT",
+        "[4/7] VERIFY / REPAIR IF NEEDED",
+        "      ↳ DIAGNOSE",
+        "      ↳ REPLAN",
+        "      ↳ APPLY CORRECTION",
+        "[5/7] INDEPENDENT REVIEW",
+        "[6/7] FINAL VERIFICATION",
+        "[7/7] ENGINEERING REPORT",
+    ]
+
+
+def test_review_correction_is_named_without_replaying_completed_stages() -> None:
+    output: list[str] = []
+    progress = _ProgressStatus(output.append)
+
+    for event in (
+        "[DISCOVER]",
+        "[TASK_SPEC]",
+        "[PLAN]",
+        "[IMPLEMENT]",
+        "[VERIFY_TARGETED]",
+        "[REVIEW]",
+        "[IMPLEMENT]",
+        "[VERIFY_TARGETED]",
+        "[REVIEW]",
+        "[FINAL_VERIFY]",
+    ):
+        progress(event)
+    progress.report()
+
+    assert "      ↳ APPLY REVIEW FIXES" in output
+    assert output.count("[3/7] IMPLEMENT") == 1
+    assert output.count("[4/7] VERIFY / REPAIR IF NEEDED") == 1
+    assert output.count("[5/7] INDEPENDENT REVIEW") == 1
+
+
+def test_verbose_progress_preserves_internal_state_and_diagnostics() -> None:
+    output: list[str] = []
+    progress = _ProgressStatus(output.append, verbose=True)
+
+    events = [
+        "[DISCOVER]",
+        "[WORKTREE] source: /tmp/repo",
+        "[RETRIEVAL] selected: src/service.py",
+    ]
+    for event in events:
+        progress(event)
+    progress.report()
+
+    assert output == [*events, "[ENGINEERING_REPORT]"]

{devagent_ai-0.8.3 → devagent_ai-0.8.4}/tests/test_production_v040.py

@@ -62,7 +62,7 @@ def test_release_version_is_semver_and_consistent() -> None:
     version = match.group(1)
     assert re.fullmatch(r"\d+\.\d+\.\d+", version)
     assert version == __version__
-    assert version == "0.8.3"
+    assert version == "0.8.4"
 
 
 def test_release_workflow_requires_green_exact_main_revision() -> None: