autocontext 0.4.0__tar.gz → 0.4.2__tar.gz

{autocontext-0.4.0 → autocontext-0.4.2}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: autocontext
-Version: 0.4.0
+Version: 0.4.2
 Summary: autocontext control plane for iterative strategy evolution.
 Project-URL: Homepage, https://github.com/greyhaven-ai/autocontext
 Project-URL: Repository, https://github.com/greyhaven-ai/autocontext
@@ -60,7 +60,7 @@ The intended use is to hand the harness a real task in plain language, let it so
 pip install autocontext
 ```
 
-The current PyPI release line is `autocontext==0.4.0`.
+The current PyPI release line is `autocontext==0.4.2`.
 The PyPI package name is now `autocontext`. The CLI entrypoint remains `autoctx`.
 
 ## Working Directory
@@ -82,13 +82,13 @@ The Python package is the full control-plane surface in this repo. It currently
 
 - generation-loop execution via `autoctx run`
 - plain-language simulation via `autoctx simulate`
+- plain-language investigation via `autoctx investigate`
 - local training workflows via `autoctx export-training-data` and `autoctx train`
 - scenario creation and materialization via `autoctx new-scenario`
 - HTTP API and MCP server surfaces via `autoctx serve` and `autoctx mcp-serve`
 
 Some newer operator-facing surfaces are currently TypeScript-first:
 
-- `autoctx investigate`
 - `autoctx analyze`
 - the interactive terminal UI via `npx autoctx tui`
 
@@ -105,7 +105,7 @@ source .venv/bin/activate
 uv sync --group dev
 ```
 
-Use the repo-level `.env.example` as the reference for available `AUTOCONTEXT_*` settings.
+Use the repo-level `.env.example` as the reference for available `AUTOCONTEXT_*` settings and supported provider-native credential aliases such as `ANTHROPIC_API_KEY`.
 
 `operator-in-the-loop` is a runnable scenario family for escalation and clarification experiments. Use it when you want executable operator-loop simulations, judgment evaluation, and live-agent escalation workflow testing.
 
@@ -120,7 +120,28 @@ Run with Anthropic:
 
 ```bash
 AUTOCONTEXT_AGENT_PROVIDER=anthropic \
-AUTOCONTEXT_ANTHROPIC_API_KEY=... \
+ANTHROPIC_API_KEY=... \
+uv run autoctx solve --description "improve customer-support replies for billing disputes" --gens 3
+```
+
+`ANTHROPIC_API_KEY` is the preferred Anthropic credential env var. `AUTOCONTEXT_ANTHROPIC_API_KEY` remains supported as a compatibility alias.
+
+Run with Claude CLI (`claude -p` via a local authenticated Claude Code runtime):
+
+```bash
+AUTOCONTEXT_AGENT_PROVIDER=claude-cli \
+AUTOCONTEXT_CLAUDE_MODEL=sonnet \
+AUTOCONTEXT_CLAUDE_TIMEOUT=300 \
+uv run autoctx solve --description "improve customer-support replies for billing disputes" --gens 3
+```
+
+For longer `autoctx judge` / `autoctx improve` prompts on `claude-cli`, use `--timeout <seconds>` or set `AUTOCONTEXT_CLAUDE_TIMEOUT`.
+
+Run with Codex CLI (`codex exec` via a local authenticated Codex runtime):
+
+```bash
+AUTOCONTEXT_AGENT_PROVIDER=codex \
+AUTOCONTEXT_CODEX_MODEL=o4-mini \
 uv run autoctx solve --description "improve customer-support replies for billing disputes" --gens 3
 ```
 
@@ -132,7 +153,9 @@ AUTOCONTEXT_PI_COMMAND=pi \
 uv run autoctx solve --description "improve customer-support replies for billing disputes" --gens 3
 ```
 
-`autoctx simulate` now follows the effective architect-role runtime surface, so `AUTOCONTEXT_ARCHITECT_PROVIDER` and other role-routing overrides apply to live simulation generation as well.
+`autoctx simulate` now follows the effective architect-role runtime surface, so `AUTOCONTEXT_ARCHITECT_PROVIDER`, other role-routing overrides, and per-call `--provider <name>` overrides all apply to live simulation generation.
+
+`autoctx investigate` now ships as a first-class Python CLI surface as well. It uses the architect runtime for investigation-spec synthesis and the analyst runtime for hypothesis generation, so role-routing overrides apply there too.
 
 Run with Pi RPC (remote Pi agent via HTTP):
 
@@ -172,6 +195,9 @@ uv run autoctx mcp-serve
 ```bash
 uv run autoctx solve --description "improve customer-support replies for billing disputes" --gens 3
 uv run autoctx simulate --description "simulate deploying a web service with rollback"
+uv run autoctx simulate --description "simulate deploying a web service with rollback" --provider claude-cli
+uv run autoctx investigate --description "why did conversion drop after Tuesday's release"
+uv run autoctx queue add --task-prompt "Write a 1-line fact about primes" --rubric "correct" --threshold 0.8 --rounds 2
 uv run autoctx simulate --replay deploy_sim --variables threshold=0.9
 uv run autoctx list
 uv run autoctx status <run_id>
@@ -186,10 +212,19 @@ uv run autoctx mcp-serve
 uv run autoctx wait <condition_id> --json
 ```
 
+Saved custom scenarios under `knowledge/_custom_scenarios/` can be rerun and benchmarked by name once their `spec.json` has been persisted, so the `new-scenario` / `solve` workflow lines up with the named `run` and `benchmark` surfaces.
+
 Useful variants:
 
 ```bash
-AUTOCONTEXT_AGENT_PROVIDER=anthropic AUTOCONTEXT_ANTHROPIC_API_KEY=... \
+AUTOCONTEXT_AGENT_PROVIDER=anthropic ANTHROPIC_API_KEY=... \
+uv run autoctx solve --description "improve customer-support replies for billing disputes" --gens 3
+
+AUTOCONTEXT_AGENT_PROVIDER=anthropic \
+ANTHROPIC_API_KEY=sk-ant-primary \
+AUTOCONTEXT_COMPETITOR_PROVIDER=openai-compatible \
+AUTOCONTEXT_COMPETITOR_API_KEY=sk-role \
+AUTOCONTEXT_COMPETITOR_BASE_URL=http://localhost:8000/v1 \
 uv run autoctx solve --description "improve customer-support replies for billing disputes" --gens 3
 
 AUTOCONTEXT_AGENT_PROVIDER=deterministic AUTOCONTEXT_RLM_ENABLED=true \
@@ -284,7 +319,7 @@ autocontext exposes:
 - [Agent integration guide](docs/agent-integration.md) — CLI-first integration for external agents, MCP fallback, JSON output reference
 - [Sandbox modes](docs/sandbox.md)
 - [MLX host training](docs/mlx-training.md)
-- [TypeScript package guide](../ts/README.md) — `investigate`, `analyze`, and interactive TUI surfaces
+- [TypeScript package guide](../ts/README.md) — `analyze`, mission control, and interactive TUI surfaces
 - [Demo data notes](demo_data/README.md)
 - [Copy-paste examples](../examples/README.md)
 - [Change history](../CHANGELOG.md)

{autocontext-0.4.0 → autocontext-0.4.2}/README.md

@@ -10,7 +10,7 @@ The intended use is to hand the harness a real task in plain language, let it so
 pip install autocontext
 ```
 
-The current PyPI release line is `autocontext==0.4.0`.
+The current PyPI release line is `autocontext==0.4.2`.
 The PyPI package name is now `autocontext`. The CLI entrypoint remains `autoctx`.
 
 ## Working Directory
@@ -32,13 +32,13 @@ The Python package is the full control-plane surface in this repo. It currently
 
 - generation-loop execution via `autoctx run`
 - plain-language simulation via `autoctx simulate`
+- plain-language investigation via `autoctx investigate`
 - local training workflows via `autoctx export-training-data` and `autoctx train`
 - scenario creation and materialization via `autoctx new-scenario`
 - HTTP API and MCP server surfaces via `autoctx serve` and `autoctx mcp-serve`
 
 Some newer operator-facing surfaces are currently TypeScript-first:
 
-- `autoctx investigate`
 - `autoctx analyze`
 - the interactive terminal UI via `npx autoctx tui`
 
@@ -55,7 +55,7 @@ source .venv/bin/activate
 uv sync --group dev
 ```
 
-Use the repo-level `.env.example` as the reference for available `AUTOCONTEXT_*` settings.
+Use the repo-level `.env.example` as the reference for available `AUTOCONTEXT_*` settings and supported provider-native credential aliases such as `ANTHROPIC_API_KEY`.
 
 `operator-in-the-loop` is a runnable scenario family for escalation and clarification experiments. Use it when you want executable operator-loop simulations, judgment evaluation, and live-agent escalation workflow testing.
 
@@ -70,7 +70,28 @@ Run with Anthropic:
 
 ```bash
 AUTOCONTEXT_AGENT_PROVIDER=anthropic \
-AUTOCONTEXT_ANTHROPIC_API_KEY=... \
+ANTHROPIC_API_KEY=... \
+uv run autoctx solve --description "improve customer-support replies for billing disputes" --gens 3
+```
+
+`ANTHROPIC_API_KEY` is the preferred Anthropic credential env var. `AUTOCONTEXT_ANTHROPIC_API_KEY` remains supported as a compatibility alias.
+
+Run with Claude CLI (`claude -p` via a local authenticated Claude Code runtime):
+
+```bash
+AUTOCONTEXT_AGENT_PROVIDER=claude-cli \
+AUTOCONTEXT_CLAUDE_MODEL=sonnet \
+AUTOCONTEXT_CLAUDE_TIMEOUT=300 \
+uv run autoctx solve --description "improve customer-support replies for billing disputes" --gens 3
+```
+
+For longer `autoctx judge` / `autoctx improve` prompts on `claude-cli`, use `--timeout <seconds>` or set `AUTOCONTEXT_CLAUDE_TIMEOUT`.
+
+Run with Codex CLI (`codex exec` via a local authenticated Codex runtime):
+
+```bash
+AUTOCONTEXT_AGENT_PROVIDER=codex \
+AUTOCONTEXT_CODEX_MODEL=o4-mini \
 uv run autoctx solve --description "improve customer-support replies for billing disputes" --gens 3
 ```
 
@@ -82,7 +103,9 @@ AUTOCONTEXT_PI_COMMAND=pi \
 uv run autoctx solve --description "improve customer-support replies for billing disputes" --gens 3
 ```
 
-`autoctx simulate` now follows the effective architect-role runtime surface, so `AUTOCONTEXT_ARCHITECT_PROVIDER` and other role-routing overrides apply to live simulation generation as well.
+`autoctx simulate` now follows the effective architect-role runtime surface, so `AUTOCONTEXT_ARCHITECT_PROVIDER`, other role-routing overrides, and per-call `--provider <name>` overrides all apply to live simulation generation.
+
+`autoctx investigate` now ships as a first-class Python CLI surface as well. It uses the architect runtime for investigation-spec synthesis and the analyst runtime for hypothesis generation, so role-routing overrides apply there too.
 
 Run with Pi RPC (remote Pi agent via HTTP):
 
@@ -122,6 +145,9 @@ uv run autoctx mcp-serve
 ```bash
 uv run autoctx solve --description "improve customer-support replies for billing disputes" --gens 3
 uv run autoctx simulate --description "simulate deploying a web service with rollback"
+uv run autoctx simulate --description "simulate deploying a web service with rollback" --provider claude-cli
+uv run autoctx investigate --description "why did conversion drop after Tuesday's release"
+uv run autoctx queue add --task-prompt "Write a 1-line fact about primes" --rubric "correct" --threshold 0.8 --rounds 2
 uv run autoctx simulate --replay deploy_sim --variables threshold=0.9
 uv run autoctx list
 uv run autoctx status <run_id>
@@ -136,10 +162,19 @@ uv run autoctx mcp-serve
 uv run autoctx wait <condition_id> --json
 ```
 
+Saved custom scenarios under `knowledge/_custom_scenarios/` can be rerun and benchmarked by name once their `spec.json` has been persisted, so the `new-scenario` / `solve` workflow lines up with the named `run` and `benchmark` surfaces.
+
 Useful variants:
 
 ```bash
-AUTOCONTEXT_AGENT_PROVIDER=anthropic AUTOCONTEXT_ANTHROPIC_API_KEY=... \
+AUTOCONTEXT_AGENT_PROVIDER=anthropic ANTHROPIC_API_KEY=... \
+uv run autoctx solve --description "improve customer-support replies for billing disputes" --gens 3
+
+AUTOCONTEXT_AGENT_PROVIDER=anthropic \
+ANTHROPIC_API_KEY=sk-ant-primary \
+AUTOCONTEXT_COMPETITOR_PROVIDER=openai-compatible \
+AUTOCONTEXT_COMPETITOR_API_KEY=sk-role \
+AUTOCONTEXT_COMPETITOR_BASE_URL=http://localhost:8000/v1 \
 uv run autoctx solve --description "improve customer-support replies for billing disputes" --gens 3
 
 AUTOCONTEXT_AGENT_PROVIDER=deterministic AUTOCONTEXT_RLM_ENABLED=true \
@@ -234,7 +269,7 @@ autocontext exposes:
 - [Agent integration guide](docs/agent-integration.md) — CLI-first integration for external agents, MCP fallback, JSON output reference
 - [Sandbox modes](docs/sandbox.md)
 - [MLX host training](docs/mlx-training.md)
-- [TypeScript package guide](../ts/README.md) — `investigate`, `analyze`, and interactive TUI surfaces
+- [TypeScript package guide](../ts/README.md) — `analyze`, mission control, and interactive TUI surfaces
 - [Demo data notes](demo_data/README.md)
 - [Copy-paste examples](../examples/README.md)
 - [Change history](../CHANGELOG.md)

{autocontext-0.4.0 → autocontext-0.4.2}/docs/agent-integration.md

@@ -223,7 +223,7 @@ Configure which LLM provider autocontext uses via environment variables:
 ```bash
 # Anthropic (default)
 AUTOCONTEXT_AGENT_PROVIDER=anthropic \
-AUTOCONTEXT_ANTHROPIC_API_KEY=sk-ant-... \
+ANTHROPIC_API_KEY=sk-ant-... \
 autoctx run --scenario my_task --json
 
 # OpenAI-compatible
@@ -267,17 +267,35 @@ AUTOCONTEXT_AGENT_PROVIDER=pi-rpc \
 AUTOCONTEXT_PI_RPC_ENDPOINT=http://localhost:3284 \
 AUTOCONTEXT_PI_RPC_API_KEY=your-key \
 autoctx run --scenario my_task --json
+
+# Role-scoped override: competitor uses a separate gateway/key
+AUTOCONTEXT_AGENT_PROVIDER=anthropic \
+ANTHROPIC_API_KEY=sk-ant-primary \
+AUTOCONTEXT_COMPETITOR_PROVIDER=openai-compatible \
+AUTOCONTEXT_COMPETITOR_API_KEY=sk-role \
+AUTOCONTEXT_COMPETITOR_BASE_URL=http://localhost:8000/v1 \
+autoctx run --scenario my_task --json
 ```
 
+`ANTHROPIC_API_KEY` is the preferred Anthropic credential env var. `AUTOCONTEXT_ANTHROPIC_API_KEY` remains supported as a compatibility alias.
+
 Key environment variables:
 
 | Variable | Purpose |
 |---|---|
 | `AUTOCONTEXT_AGENT_PROVIDER` | Agent provider: `anthropic`, `openai-compatible`, `ollama`, `vllm`, `pi`, `pi-rpc`, `deterministic` |
+| `AUTOCONTEXT_AGENT_API_KEY` | Global agent API key override (or use provider-native env vars such as `ANTHROPIC_API_KEY`) |
+| `AUTOCONTEXT_AGENT_BASE_URL` | Global base URL for OpenAI-compatible agent endpoints |
+| `AUTOCONTEXT_COMPETITOR_API_KEY` / `AUTOCONTEXT_COMPETITOR_BASE_URL` | Optional competitor-specific credential and endpoint override |
+| `AUTOCONTEXT_ANALYST_API_KEY` / `AUTOCONTEXT_ANALYST_BASE_URL` | Optional analyst-specific credential and endpoint override |
+| `AUTOCONTEXT_COACH_API_KEY` / `AUTOCONTEXT_COACH_BASE_URL` | Optional coach-specific credential and endpoint override |
+| `AUTOCONTEXT_ARCHITECT_API_KEY` / `AUTOCONTEXT_ARCHITECT_BASE_URL` | Optional architect-specific credential and endpoint override |
 | `AUTOCONTEXT_JUDGE_PROVIDER` | Judge provider (defaults to `anthropic`) |
 | `AUTOCONTEXT_JUDGE_API_KEY` | API key for the judge provider |
 | `AUTOCONTEXT_JUDGE_BASE_URL` | Base URL for OpenAI-compatible judge endpoints |
 | `AUTOCONTEXT_JUDGE_MODEL` | Override judge model name |
+| `AUTOCONTEXT_CLAUDE_MODEL` | Claude CLI model alias (default: `sonnet`) |
+| `AUTOCONTEXT_CLAUDE_TIMEOUT` | Claude CLI execution timeout in seconds (default: 120) |
 | `AUTOCONTEXT_MODEL_COMPETITOR` | Override competitor agent model |
 | `AUTOCONTEXT_DB_PATH` | SQLite database path |
 | `AUTOCONTEXT_PI_COMMAND` | Path to Pi CLI binary (default: `pi`) |
@@ -530,7 +548,7 @@ Add to your project's `.claude/settings.json`:
       "args": ["run", "--directory", "/path/to/autocontext", "autoctx", "mcp-serve"],
       "env": {
         "AUTOCONTEXT_AGENT_PROVIDER": "anthropic",
-        "AUTOCONTEXT_ANTHROPIC_API_KEY": "sk-ant-..."
+        "ANTHROPIC_API_KEY": "sk-ant-..."
       }
     }
   }

{autocontext-0.4.0 → autocontext-0.4.2}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "hatchling.build"
 
 [project]
 name = "autocontext"
-version = "0.4.0"
+version = "0.4.2"
 description = "autocontext control plane for iterative strategy evolution."
 readme = "README.md"
 license = { text = "Apache-2.0" }

{autocontext-0.4.0 → autocontext-0.4.2}/src/autocontext/__init__.py

@@ -4,4 +4,4 @@ from autocontext.sdk import AutoContext
 
 __all__ = ["AutoContext", "__version__"]
 
-__version__ = "0.4.0"
+__version__ = "0.4.2"

{autocontext-0.4.0 → autocontext-0.4.2}/src/autocontext/agents/feedback_loops.py

@@ -15,10 +15,11 @@ Key types:
 
 from __future__ import annotations
 
+import json
 import statistics
 from typing import Any
 
-from pydantic import BaseModel, Field, computed_field
+from pydantic import BaseModel, Field, computed_field, field_validator
 
 # ---------------------------------------------------------------------------
 # AC-336: Analyst quality scoring
@@ -39,6 +40,17 @@ class AnalystRating(BaseModel):
     def overall(self) -> float:
         return round(statistics.mean([self.actionability, self.specificity, self.correctness]), 2)
 
+    @field_validator("rationale", mode="before")
+    @classmethod
+    def _coerce_rationale(cls, value: Any) -> str:
+        if value is None:
+            return ""
+        if isinstance(value, str):
+            return value
+        if isinstance(value, dict | list):
+            return json.dumps(value, sort_keys=True)
+        return str(value)
+
     def to_dict(self) -> dict[str, Any]:
         return self.model_dump()
 
@@ -90,8 +102,7 @@ class ToolUsageTracker:
     def __init__(self, known_tools: list[str]) -> None:
         self._tools = known_tools
         self._records: dict[str, ToolUsageRecord] = {
-            name: ToolUsageRecord(tool_name=name, used_in_gens=[], last_used=0, total_refs=0)
-            for name in known_tools
+            name: ToolUsageRecord(tool_name=name, used_in_gens=[], last_used=0, total_refs=0) for name in known_tools
         }
 
     def record_generation(self, generation: int, strategy_text: str) -> None:
@@ -110,10 +121,7 @@ class ToolUsageTracker:
 
     def to_dict(self) -> dict[str, Any]:
         return {
-            "records": {
-                name: record.to_dict()
-                for name, record in sorted(self._records.items())
-            },
+            "records": {name: record.to_dict() for name, record in sorted(self._records.items())},
         }
 
     @classmethod

{autocontext-0.4.0 → autocontext-0.4.2}/src/autocontext/agents/llm_client.py

@@ -529,7 +529,11 @@ def build_client_from_settings(
 ) -> LanguageModelClient:
     """Construct a LanguageModelClient from AppSettings."""
     if settings.agent_provider == "anthropic":
-        api_key = settings.anthropic_api_key or os.getenv("ANTHROPIC_API_KEY", "")
+        api_key = (
+            settings.anthropic_api_key
+            or os.getenv("ANTHROPIC_API_KEY")
+            or os.getenv("AUTOCONTEXT_ANTHROPIC_API_KEY", "")
+        )
         if not api_key:
             raise ValueError(
                 "AUTOCONTEXT_ANTHROPIC_API_KEY or ANTHROPIC_API_KEY is required "
@@ -564,6 +568,30 @@ def build_client_from_settings(
             model=settings.agent_default_model,
         )
         return ProviderBridgeClient(provider, use_provider_default_model=True)
+    if settings.agent_provider == "claude-cli":
+        from autocontext.agents.provider_bridge import RuntimeBridgeClient
+        from autocontext.runtimes.claude_cli import ClaudeCLIConfig, ClaudeCLIRuntime
+
+        claude_config = ClaudeCLIConfig(
+            model=settings.claude_model,
+            tools=settings.claude_tools,
+            permission_mode=settings.claude_permission_mode,
+            session_persistence=settings.claude_session_persistence,
+            timeout=settings.claude_timeout,
+        )
+        return RuntimeBridgeClient(ClaudeCLIRuntime(claude_config))
+    if settings.agent_provider == "codex":
+        from autocontext.agents.provider_bridge import RuntimeBridgeClient
+        from autocontext.runtimes.codex_cli import CodexCLIConfig, CodexCLIRuntime
+
+        codex_config = CodexCLIConfig(
+            model=settings.codex_model,
+            approval_mode=settings.codex_approval_mode,
+            timeout=settings.codex_timeout,
+            workspace=settings.codex_workspace,
+            quiet=settings.codex_quiet,
+        )
+        return RuntimeBridgeClient(CodexCLIRuntime(codex_config))
     if settings.agent_provider == "pi":
         from autocontext.agents.provider_bridge import RuntimeBridgeClient
         from autocontext.providers.scenario_routing import resolve_pi_model
@@ -585,19 +613,20 @@ def build_client_from_settings(
             if handoff is not None:
                 resolved_model = handoff.checkpoint_path
 
-        config = PiCLIConfig(
+        pi_config = PiCLIConfig(
             pi_command=settings.pi_command,
             timeout=settings.pi_timeout,
             workspace=settings.pi_workspace,
             model=resolved_model,
         )
-        return RuntimeBridgeClient(PiCLIRuntime(config))
+        return RuntimeBridgeClient(PiCLIRuntime(pi_config))
     if settings.agent_provider == "pi-rpc":
         from autocontext.agents.provider_bridge import RuntimeBridgeClient
         from autocontext.runtimes.pi_rpc import PiRPCConfig, PiRPCRuntime
 
         rpc_config = PiRPCConfig(
             pi_command=settings.pi_command,
+            timeout=settings.pi_timeout,
             session_persistence=settings.pi_rpc_session_persistence,
         )
         return RuntimeBridgeClient(PiRPCRuntime(rpc_config))