pretext-sim 0.3.2__tar.gz → 0.3.3__tar.gz

{pretext_sim-0.3.2 → pretext_sim-0.3.3}/.github/workflows/ci.yml

@@ -15,7 +15,7 @@ jobs:
       fail-fast: false
       matrix:
         os: [ubuntu-latest]
-        python: ["3.10", "3.11", "3.12", "3.13"]
+        python: ["3.11", "3.12", "3.13"]
 
     steps:
       - uses: actions/checkout@v7

{pretext_sim-0.3.2 → pretext_sim-0.3.3}/.gitignore

@@ -10,3 +10,6 @@ dist/
 .crush/
 .empryo/
 .coverage
+
+.env.*
+!.env.example

{pretext_sim-0.3.2 → pretext_sim-0.3.3}/AGENTS.md

@@ -54,7 +54,9 @@ resolves.
 
 `doctor` runs its probes through `Client.list_models` and `Client.ping_json` so they
 stay testable via `MockTransport`; `--full` is the only part that calls `generate`.
-`batch` mirrors `generate`'s endpoint flags, skips `_`-prefixed files, isolates
+`batch` runs profiles in a bounded thread pool (default 4) and one `Client` per
+profile, because `Client.last_usage` is per-instance. Printing stays on the calling
+thread. `batch` mirrors `generate`'s endpoint flags, skips `_`-prefixed files, isolates
 per-profile failures, and exits 1 if any profile failed. Keep the summary on stderr
 and never abort the run on the first failure.
 
@@ -71,6 +73,8 @@ and never abort the run on the first failure.
   `Simulation` forbids extra keys and requires `escalation_path` stages to be
   exactly `1..n`, so a model that invents a `stage 2` gap fails instead of
   producing a broken table.
+- `_read_response` validates both halves of an export inside one `try`; a corrupt
+  `profile` key must surface as `ProfileError`, not a raw pydantic traceback.
 - **Errors are `PretextError` subclasses** (`ConfigError`, `ProfileError`,
   `APIError`). The CLI catches them, prints to stderr, exits 1. Do not let raw
   `httpx` or `ValidationError` escape to the user, and do not catch them per
@@ -100,6 +104,8 @@ and never abort the run on the first failure.
 - Empty model content is an error, not an empty simulation. Overlong artifacts are
   trimmed to 300 chars in a `mode="before"` validator; constraints run before
   `mode="after"` validators, so trim early or the run fails.
+- `--out` creates parent directories through `_write_out`; use it rather than calling
+  `write_text` directly so a missing directory never tracebacks.
 - `--dry-run` still resolves settings, so it needs a key; it prints the prompt to
   stdout via `typer.echo` (unwrapped) and exits 0. `validate` needs no key.
 - File output with the default rich format renders into a `Console.capture()`
@@ -127,7 +133,7 @@ the social preview (404 on every version), so it is uploaded in Settings → Gen
 ## Repo hygiene
 
 CI (`.github/workflows/ci.yml`) runs ruff, pytest, and a dry-run smoke on Python
-3.10-3.13; the local equivalents are `ruff check src tests` and `pytest`. Test
+3.11-3.13; the local equivalents are `ruff check src tests` and `pytest`. Test
 files are exempt from E501, source files are not.
 
 Releasing means bumping `__version__` in `__init__.py` and `version` in
@@ -150,6 +156,9 @@ without HTTP. Cover the failure path (retry exhaustion, schema violation, bad
 key) whenever you touch the client or schema. `tests/test_models.py` owns
 precedence and validation edge cases.
 
+`--env-file` swaps the dotenv source; note a `PRETEXT_BASE_URL` in that file beats a
+`--provider` preset, which is what trips up mixed setups.
+
 `--from-file` bypasses the API entirely for rendering work. It accepts our own
 export, a bare simulation, or a chat body, and needs no key; the profile is taken
 from the export or must be passed explicitly. `tests/fixtures/responses/dana.json`

{pretext_sim-0.3.2 → pretext_sim-0.3.3}/CHANGELOG.md

@@ -4,6 +4,18 @@ All notable changes. This project follows semantic versioning.
 
 ## Unreleased
 
+- `batch` runs profiles concurrently (`-j`, default 4)
+- `--env-file` selects another dotenv, for running two providers side by side
+- Upstream provider error detail is surfaced instead of "Provider returned error"
+- Corpus entries for two OpenRouter models; socket-level transport tests
+
+- A corrupt `profile` inside a saved response reports field errors instead of a
+  traceback
+- `doctor` suggests close model names, and shows the first schema errors verbatim
+- `batch` reads `.yml` as well as `.yaml`
+- `--out` creates missing parent directories
+- Section parsing lives in one helper instead of two copies
+
 - Distribution renamed `pretext-sim` (the `pretext` name belongs to PreTeXt); the
   command stays `pretext`
 - PyPI publish workflow on release, via trusted publishing
@@ -19,8 +31,14 @@ All notable changes. This project follows semantic versioning.
 - `--dry-run --out` writes the prompt instead of ignoring `--out`
 - Verified against Venice (venice-uncensored-1-2, qwen3-235b-a22b-instruct)
 
+## 0.3.3
+
+- Correct install instructions: `uvx` needs `--from pretext-sim`
+
 ## 0.3.2
 
+- Dropped Python 3.10 (end of life) so the lockfile can use a patched `filelock`
+
 - PyPI upload runs inside the release workflow
 
 ## 0.3.1

{pretext_sim-0.3.2 → pretext_sim-0.3.3}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.5
 Name: pretext-sim
-Version: 0.3.2
+Version: 0.3.3
 Summary: Generate labeled social-engineering simulations for security training
 Author: pretext contributors
 License: MIT
@@ -11,12 +11,11 @@ Classifier: Environment :: Console
 Classifier: Intended Audience :: Information Technology
 Classifier: License :: OSI Approved :: MIT License
 Classifier: Programming Language :: Python :: 3
-Classifier: Programming Language :: Python :: 3.10
 Classifier: Programming Language :: Python :: 3.11
 Classifier: Programming Language :: Python :: 3.12
 Classifier: Programming Language :: Python :: 3.13
 Classifier: Topic :: Security
-Requires-Python: >=3.10
+Requires-Python: >=3.11
 Requires-Dist: httpx>=0.27
 Requires-Dist: pydantic>=2.6
 Requires-Dist: pyyaml>=6.0
@@ -40,7 +39,7 @@ Narrative, opening message, escalation path, and the red flags defenders should
 have caught, generated, validated, then rendered where you need it.
 
 [![ci](https://github.com/sjdsjadjajdjsa/pretext/actions/workflows/ci.yml/badge.svg)](https://github.com/sjdsjadjajdjsa/pretext/actions/workflows/ci.yml)
-![python](https://img.shields.io/badge/python-3.10%20%7C%203.11%20%7C%203.12%20%7C%203.13-blue.svg)
+![python](https://img.shields.io/badge/python-3.11%20%7C%203.12%20%7C%203.13-blue.svg)
 ![license](https://img.shields.io/badge/license-MIT-blue.svg)
 
 <img src="docs/demo.png" alt="pretext rendering a simulation in the terminal" width="820">
@@ -82,12 +81,12 @@ complete Markdown export; [`docs/demo.png`](docs/demo.png) is the terminal view.
 ## Install
 
 ```sh
-uvx pretext-sim generate persona.yaml     # no install
-pipx install pretext-sim                  # or a permanent command
+uvx --from pretext-sim pretext generate persona.yaml     # no install
+pipx install pretext-sim                                 # or a permanent command
 ```
 
-Python 3.10+. The distribution is `pretext-sim`; the command is `pretext`. For the
-source tree: `uv sync --extra dev` then `uv run pretext`.
+Python 3.11+. The distribution is `pretext-sim`; the command is `pretext`, so `uvx`
+needs `--from`. For the source tree: `uv sync --extra dev` then `uv run pretext`.
 
 ## Quickstart
 
@@ -146,20 +145,22 @@ rejected so typos surface early. Fixture personas live in
 | `schema` | Print the JSON Schema model output must satisfy (`--all` adds the profile) |
 | `config` | Show resolved settings, key masked |
 
-Flags on `generate`:
+Flags on `generate` (endpoint, `--env-file`, `--sections` and `--format` also apply
+to `batch`):
 
 | Flag | Effect |
 | ---- | ------ |
 | `--channel` | Force the opening channel (email, phone, sms, chat, in_person, linkedin) |
 | `--extra` | Extra direction for the model |
 | `--format` | `rich` (default), `markdown`, `json` |
-| `--out` | Write to a file; plain text unless a format is chosen |
+| `--out` | Write to a file, creating parents; plain text unless a format is chosen |
 | `--sections` | Render only some of `pretext,opening,escalation,red_flags` |
 | `--from-file` | Replay a saved response; no API call, no key |
 | `--dry-run` | Print resolved config and prompt, send nothing; `--out` writes the prompt |
 | `--usage` | Token counts on stderr |
 | `--base-url`, `--model`, `--api-key` | Override endpoint settings |
 | `--temperature`, `--max-tokens`, `--timeout`, `--retries`, `--seed` | Sampling and transport |
+| `--env-file` | Read a different dotenv file (default `./.env`) |
 | `--no-color` | Plain terminal output |
 
 Exit codes: `0` success, `1` bad profile/config/API/schema failure, `2` usage error,
@@ -187,7 +188,8 @@ continuing past failures:
 pretext batch profiles/ --out-dir packet/ --format markdown
 ```
 
-Progress and a `wrote 7/9` summary go to stderr, the failure reasons are listed, and
+Profiles run four at a time by default (`-j/--concurrency`, max 8); progress and a
+`wrote 7/9` summary go to stderr, the failure reasons are listed, and
 the exit code is `1` if anything failed. Endpoint flags are the same as `generate`;
 `--sections`, `--format` and `--usage` apply to the run as a whole.
 
@@ -197,7 +199,7 @@ First hit wins:
 
 1. Flags (`--base-url`, `--model`, `--api-key`, `--provider`)
 2. Environment: the same names under `PRETEXT_` (`PRETEXT_API_KEY`, `PRETEXT_MODEL`, ...)
-3. `.env` in the working directory
+3. `.env` in the working directory, or the file given to `--env-file`
 4. `OPENAI_API_KEY` for the key, then `https://api.openai.com/v1` and `gpt-4o-mini`
 
 The key is never printed and never written to output.

{pretext_sim-0.3.2 → pretext_sim-0.3.3}/README.md

@@ -8,7 +8,7 @@ Narrative, opening message, escalation path, and the red flags defenders should
 have caught, generated, validated, then rendered where you need it.
 
 [![ci](https://github.com/sjdsjadjajdjsa/pretext/actions/workflows/ci.yml/badge.svg)](https://github.com/sjdsjadjajdjsa/pretext/actions/workflows/ci.yml)
-![python](https://img.shields.io/badge/python-3.10%20%7C%203.11%20%7C%203.12%20%7C%203.13-blue.svg)
+![python](https://img.shields.io/badge/python-3.11%20%7C%203.12%20%7C%203.13-blue.svg)
 ![license](https://img.shields.io/badge/license-MIT-blue.svg)
 
 <img src="docs/demo.png" alt="pretext rendering a simulation in the terminal" width="820">
@@ -50,12 +50,12 @@ complete Markdown export; [`docs/demo.png`](docs/demo.png) is the terminal view.
 ## Install
 
 ```sh
-uvx pretext-sim generate persona.yaml     # no install
-pipx install pretext-sim                  # or a permanent command
+uvx --from pretext-sim pretext generate persona.yaml     # no install
+pipx install pretext-sim                                 # or a permanent command
 ```
 
-Python 3.10+. The distribution is `pretext-sim`; the command is `pretext`. For the
-source tree: `uv sync --extra dev` then `uv run pretext`.
+Python 3.11+. The distribution is `pretext-sim`; the command is `pretext`, so `uvx`
+needs `--from`. For the source tree: `uv sync --extra dev` then `uv run pretext`.
 
 ## Quickstart
 
@@ -114,20 +114,22 @@ rejected so typos surface early. Fixture personas live in
 | `schema` | Print the JSON Schema model output must satisfy (`--all` adds the profile) |
 | `config` | Show resolved settings, key masked |
 
-Flags on `generate`:
+Flags on `generate` (endpoint, `--env-file`, `--sections` and `--format` also apply
+to `batch`):
 
 | Flag | Effect |
 | ---- | ------ |
 | `--channel` | Force the opening channel (email, phone, sms, chat, in_person, linkedin) |
 | `--extra` | Extra direction for the model |
 | `--format` | `rich` (default), `markdown`, `json` |
-| `--out` | Write to a file; plain text unless a format is chosen |
+| `--out` | Write to a file, creating parents; plain text unless a format is chosen |
 | `--sections` | Render only some of `pretext,opening,escalation,red_flags` |
 | `--from-file` | Replay a saved response; no API call, no key |
 | `--dry-run` | Print resolved config and prompt, send nothing; `--out` writes the prompt |
 | `--usage` | Token counts on stderr |
 | `--base-url`, `--model`, `--api-key` | Override endpoint settings |
 | `--temperature`, `--max-tokens`, `--timeout`, `--retries`, `--seed` | Sampling and transport |
+| `--env-file` | Read a different dotenv file (default `./.env`) |
 | `--no-color` | Plain terminal output |
 
 Exit codes: `0` success, `1` bad profile/config/API/schema failure, `2` usage error,
@@ -155,7 +157,8 @@ continuing past failures:
 pretext batch profiles/ --out-dir packet/ --format markdown
 ```
 
-Progress and a `wrote 7/9` summary go to stderr, the failure reasons are listed, and
+Profiles run four at a time by default (`-j/--concurrency`, max 8); progress and a
+`wrote 7/9` summary go to stderr, the failure reasons are listed, and
 the exit code is `1` if anything failed. Endpoint flags are the same as `generate`;
 `--sections`, `--format` and `--usage` apply to the run as a whole.
 
@@ -165,7 +168,7 @@ First hit wins:
 
 1. Flags (`--base-url`, `--model`, `--api-key`, `--provider`)
 2. Environment: the same names under `PRETEXT_` (`PRETEXT_API_KEY`, `PRETEXT_MODEL`, ...)
-3. `.env` in the working directory
+3. `.env` in the working directory, or the file given to `--env-file`
 4. `OPENAI_API_KEY` for the key, then `https://api.openai.com/v1` and `gpt-4o-mini`
 
 The key is never printed and never written to output.

{pretext_sim-0.3.2 → pretext_sim-0.3.3}/pyproject.toml

@@ -1,9 +1,9 @@
 [project]
 name = "pretext-sim"
-version = "0.3.2"
+version = "0.3.3"
 description = "Generate labeled social-engineering simulations for security training"
 readme = "README.md"
-requires-python = ">=3.10"
+requires-python = ">=3.11"
 license = { text = "MIT" }
 authors = [{ name = "pretext contributors" }]
 keywords = ["security", "security-awareness", "phishing", "red-team", "cli", "training"]
@@ -13,7 +13,6 @@ classifiers = [
     "Intended Audience :: Information Technology",
     "License :: OSI Approved :: MIT License",
     "Programming Language :: Python :: 3",
-    "Programming Language :: Python :: 3.10",
     "Programming Language :: Python :: 3.11",
     "Programming Language :: Python :: 3.12",
     "Programming Language :: Python :: 3.13",
@@ -54,7 +53,7 @@ addopts = "-q"
 source = ["pretext"]
 
 [tool.mypy]
-python_version = "3.10"
+python_version = "3.11"
 files = ["src"]
 warn_unused_ignores = true
 warn_redundant_casts = true
@@ -67,7 +66,7 @@ ignore_missing_imports = true
 
 [tool.ruff]
 line-length = 100
-target-version = "py310"
+target-version = "py311"
 
 [tool.ruff.lint]
 ignore = ["B008"]

{pretext_sim-0.3.2 → pretext_sim-0.3.3}/scripts/capture_corpus.py

@@ -1,6 +1,6 @@
 """Capture model outputs into the regression corpus.
 
-Usage: uv run python scripts/capture_corpus.py MODEL [MODEL ...]
+Usage: [PRETEXT_ENV_FILE=.env.other] uv run python scripts/capture_corpus.py MODEL ...
 
 Needs a configured endpoint and key (see README). The corpus is checked in so the
 suite stays offline; re-run this after a prompt change and read the diff.
@@ -9,6 +9,7 @@ suite stays offline; re-run this after a prompt change and read the diff.
 from __future__ import annotations
 
 import json
+import os
 import sys
 from pathlib import Path
 
@@ -28,8 +29,13 @@ def main(models: list[str]) -> int:
     CORPUS.mkdir(parents=True, exist_ok=True)
     profile = load_profile(PROFILE)
 
+    env_file = os.environ.get("PRETEXT_ENV_FILE")
     for model in models:
-        settings = resolve_settings(model=model, max_tokens=12000)
+        settings = resolve_settings(
+            model=model,
+            max_tokens=12000,
+            dotenv=Path(env_file) if env_file else Path.cwd() / ".env",
+        )
         simulation = Client(settings).generate(profile)
         target = CORPUS / f"{model.replace('/', '__')}.json"
         target.write_text(

{pretext_sim-0.3.2 → pretext_sim-0.3.3}/src/pretext/__init__.py

@@ -15,7 +15,7 @@ from .models import (
     resolve_settings,
 )
 
-__version__ = "0.3.2"
+__version__ = "0.3.3"
 
 __all__ = [
     "APIError",

{pretext_sim-0.3.2 → pretext_sim-0.3.3}/src/pretext/cli.py

@@ -2,9 +2,11 @@
 
 from __future__ import annotations
 
+import difflib
 import json
 import sys
-from enum import Enum
+from concurrent.futures import ThreadPoolExecutor, as_completed
+from enum import StrEnum
 from importlib import resources
 from pathlib import Path
 
@@ -52,7 +54,7 @@ def root(
     """Labeled social-engineering simulations for authorized training."""
 
 
-class OutputFormat(str, Enum):
+class OutputFormat(StrEnum):
     rich = "rich"
     json = "json"
     markdown = "markdown"
@@ -77,20 +79,41 @@ def _read_response(path: Path) -> tuple[Profile | None, Simulation]:
     if not isinstance(document, dict):
         raise ProfileError(f"response {path} must be a JSON object")
 
-    profile = None
-    if isinstance(document.get("profile"), dict):
-        profile = Profile.model_validate(document["profile"])
+    try:
+        profile = (
+            Profile.model_validate(document["profile"])
+            if isinstance(document.get("profile"), dict)
+            else None
+        )
 
-    raw = document.get("simulation", document)
-    if isinstance(document.get("choices"), list):
-        raw = extract_json(_content_of(document))
+        raw = document.get("simulation", document)
+        if isinstance(document.get("choices"), list):
+            raw = extract_json(_content_of(document))
 
-    try:
         return profile, Simulation.model_validate(raw)
     except ValidationError as exc:
         raise ProfileError(_format_validation(exc, prefix=f"invalid response {path}")) from exc
 
 
+def _write_out(path: Path, body: str) -> None:
+    path.parent.mkdir(parents=True, exist_ok=True)
+    path.write_text(body, encoding="utf-8")
+
+
+def _parse_sections(raw: str | None, err) -> tuple[str, ...]:
+    if raw is None:
+        return render.SECTIONS
+    chosen = tuple(part.strip() for part in raw.split(",") if part.strip())
+    unknown = [part for part in chosen if part not in render.SECTIONS]
+    if unknown or not chosen:
+        err.print(
+            f"[bold red]error[/bold red] unknown section(s): {', '.join(unknown) or 'none'}; "
+            f"choose from {', '.join(render.SECTIONS)}"
+        )
+        raise typer.Exit(2)
+    return chosen
+
+
 def _warnings(settings, profile=None) -> list[str]:
     warnings: list[str] = []
     if settings.base_url.scheme != "https":
@@ -123,6 +146,9 @@ def generate(
         "-p",
         help=f"Preset base URL: {', '.join(PROVIDERS)}. --base-url wins.",
     ),
+    env_file: Path = typer.Option(
+        None, "--env-file", help="Dotenv file to read instead of ./.env."
+    ),
     channel: Channel | None = typer.Option(
         None, "--channel", "-c", help="Force the opening channel."
     ),
@@ -155,16 +181,7 @@ def generate(
     out = render.console(no_color=no_color)
     err = render.console(no_color=no_color, stderr=True)
 
-    selected: tuple[str, ...] = render.SECTIONS
-    if sections is not None:
-        selected = tuple(part.strip() for part in sections.split(",") if part.strip())
-        unknown = [part for part in selected if part not in render.SECTIONS]
-        if unknown or not selected:
-            err.print(
-                f"[bold red]error[/bold red] unknown section(s): {', '.join(unknown) or 'none'}; "
-                f"choose from {', '.join(render.SECTIONS)}"
-            )
-            raise typer.Exit(2)
+    selected = _parse_sections(sections, err)
 
     if profile is None and response_file is None:
         err.print("[bold red]error[/bold red] pass a PROFILE or --from-file")
@@ -192,7 +209,7 @@ def generate(
             retries=retries,
             max_tokens=max_tokens,
             seed=seed,
-            dotenv=Path.cwd() / ".env",
+            dotenv=env_file or Path.cwd() / ".env",
         )
     except PretextError as exc:
         err.print(f"[bold red]error[/bold red] {exc}")
@@ -210,7 +227,7 @@ def generate(
             err.print()
             typer.echo(prompt)
         else:
-            out_file.write_text(prompt + "\n", encoding="utf-8")
+            _write_out(out_file, prompt + "\n")
             err.print(f"wrote prompt to {out_file}")
         raise typer.Exit(0)
 
@@ -254,7 +271,7 @@ def generate(
             with recording.capture() as captured:
                 render.render_simulation(simulation, prof, recording, sections=selected)
             body = captured.get()
-        out_file.write_text(body, encoding="utf-8")
+        _write_out(out_file, body)
         err.print(f"wrote {out_file}")
         return
 
@@ -289,13 +306,16 @@ def example(
     if out_file is None:
         typer.echo(text, nl=False)
     else:
-        out_file.write_text(text, encoding="utf-8")
+        _write_out(out_file, text)
         typer.echo(f"wrote {out_file}")
 
 
 @app.command("config")
 def show_config(
     provider: str = typer.Option(None, "--provider", "-p", help=f"Preset: {', '.join(PROVIDERS)}."),
+    env_file: Path = typer.Option(
+        None, "--env-file", help="Dotenv file to read instead of ./.env."
+    ),
     base_url: str = typer.Option(None, "--base-url", "-u"),
     model: str = typer.Option(None, "--model", "-m"),
     api_key: str = typer.Option(None, "--api-key", "-k"),
@@ -310,7 +330,7 @@ def show_config(
             base_url=base_url,
             model=model,
             api_key=api_key,
-            dotenv=Path.cwd() / ".env",
+            dotenv=env_file or Path.cwd() / ".env",
         )
     except ConfigError as exc:
         err.print(f"[bold red]error[/bold red] {exc}")
@@ -347,32 +367,33 @@ def batch(
         "-p",
         help=f"Preset base URL: {', '.join(PROVIDERS)}. --base-url wins.",
     ),
+    env_file: Path = typer.Option(
+        None, "--env-file", help="Dotenv file to read instead of ./.env."
+    ),
     temperature: float = typer.Option(None, "--temperature", "-t", min=0.0, max=2.0),
     max_tokens: int = typer.Option(None, "--max-tokens", min=256, max=16000),
     timeout: float = typer.Option(None, "--timeout", min=1.0, max=600.0),
     retries: int = typer.Option(None, "--retries", min=0, max=8),
     seed: int = typer.Option(None, "--seed"),
     usage: bool = typer.Option(False, "--usage", help="Print token totals to stderr."),
+    concurrency: int = typer.Option(
+        4, "--concurrency", "-j", min=1, max=8, help="Profiles in flight at once."
+    ),
     no_color: bool = typer.Option(False, "--no-color"),
 ) -> None:
     """Generate one exercise per profile in a directory."""
 
     err = render.console(no_color=no_color, stderr=True)
-    selected: tuple[str, ...] = render.SECTIONS
-    if sections is not None:
-        selected = tuple(part.strip() for part in sections.split(",") if part.strip())
-        unknown = [part for part in selected if part not in render.SECTIONS]
-        if unknown or not selected:
-            err.print(f"[bold red]error[/bold red] unknown section(s): {', '.join(unknown)}")
-            raise typer.Exit(2)
+    selected = _parse_sections(sections, err)
 
     if not directory.is_dir():
         err.print(f"[bold red]error[/bold red] {directory} is not a directory")
         raise typer.Exit(1)
 
-    profiles = sorted(p for p in directory.glob("*.yaml") if not p.name.startswith("_"))
+    candidates = list(directory.glob("*.yaml")) + list(directory.glob("*.yml"))
+    profiles = sorted(p for p in candidates if not p.name.startswith("_"))
     if not profiles:
-        err.print(f"[bold red]error[/bold red] no *.yaml profiles in {directory}")
+        err.print(f"[bold red]error[/bold red] no *.yaml or *.yml profiles in {directory}")
         raise typer.Exit(1)
 
     try:
@@ -386,32 +407,26 @@ def batch(
             retries=retries,
             max_tokens=max_tokens,
             seed=seed,
-            dotenv=Path.cwd() / ".env",
+            dotenv=env_file or Path.cwd() / ".env",
         )
     except PretextError as exc:
         err.print(f"[bold red]error[/bold red] {exc}")
         raise typer.Exit(1) from exc
 
     out_dir.mkdir(parents=True, exist_ok=True)
-    client = Client(settings)
     suffix = {"markdown": "md", "json": "json", "rich": "txt"}[fmt.value]
     failures: list[tuple[str, str]] = []
     totals = {"prompt_tokens": 0, "completion_tokens": 0, "total_tokens": 0}
+    done = 0
 
-    for index, path in enumerate(profiles, start=1):
-        label = f"[cyan]{index}/{len(profiles)}[/cyan] {path.name}"
+    def run_one(path: Path) -> tuple[Path, str | None, dict[str, int]]:
         try:
             prof = load_profile(path)
+            client = Client(settings)
             simulation = client.generate(prof, channel=None, extra=extra)
         except PretextError as exc:
-            failures.append((path.name, str(exc).splitlines()[0]))
-            err.print(f"{label} [red]failed[/red] {failures[-1][1]}")
-            continue
+            return path, str(exc).splitlines()[0], {}
 
-        for key in totals:
-            totals[key] += (client.last_usage or {}).get(key, 0)
-
-        target = out_dir / f"{path.stem}.{suffix}"
         if fmt is OutputFormat.json:
             body = render.to_json(simulation, prof, sections=selected)
         elif fmt is OutputFormat.markdown:
@@ -421,8 +436,23 @@ def batch(
             with recording.capture() as captured:
                 render.render_simulation(simulation, prof, recording, sections=selected)
             body = captured.get()
-        target.write_text(body, encoding="utf-8")
-        err.print(f"{label} [green]ok[/green] {target.name}")
+        _write_out(out_dir / f"{path.stem}.{suffix}", body)
+        return path, None, client.last_usage or {}
+
+    workers = max(1, min(concurrency, len(profiles)))
+    with ThreadPoolExecutor(max_workers=workers) as pool:
+        futures = {pool.submit(run_one, path): path for path in profiles}
+        for future in as_completed(futures):
+            path, error, used = future.result()
+            done += 1
+            label = f"[cyan]{done}/{len(profiles)}[/cyan] {path.name}"
+            if error is not None:
+                failures.append((path.name, error))
+                err.print(f"{label} [red]failed[/red] {error}")
+                continue
+            for key in totals:
+                totals[key] += used.get(key, 0)
+            err.print(f"{label} [green]ok[/green] {path.stem}.{suffix}")
 
     err.print()
     err.print(f"wrote {len(profiles) - len(failures)}/{len(profiles)} to {out_dir}")
@@ -452,6 +482,9 @@ def doctor(
         "-p",
         help=f"Preset base URL: {', '.join(PROVIDERS)}. --base-url wins.",
     ),
+    env_file: Path = typer.Option(
+        None, "--env-file", help="Dotenv file to read instead of ./.env."
+    ),
     full: bool = typer.Option(False, "--full", help="Also run one real simulation."),
     no_color: bool = typer.Option(False, "--no-color"),
 ) -> None:
@@ -470,7 +503,7 @@ def doctor(
             base_url=base_url,
             model=model,
             api_key=api_key,
-            dotenv=Path.cwd() / ".env",
+            dotenv=env_file or Path.cwd() / ".env",
         )
         record("configuration", "pass", f"{settings.base_url} · {settings.model}")
     except PretextError as exc:
@@ -492,7 +525,7 @@ def doctor(
         elif settings.model in models:
             record("model", "pass", f"{settings.model} is served")
         else:
-            near = [m for m in models if m.split("/")[-1] in settings.model.split("/")[-1]][:3]
+            near = difflib.get_close_matches(settings.model, models, n=3, cutoff=0.5)
             hint = f"; did you mean {', '.join(near)}" if near else ""
             record("model", "fail", f"{settings.model} is not in the model list{hint}")
     except PretextError as exc:
@@ -519,7 +552,7 @@ def doctor(
                 f"{len(simulation.escalation_path)} stages, {len(simulation.red_flags)} flags",
             )
         except PretextError as exc:
-            record("schema", "fail", str(exc).splitlines()[0])
+            record("schema", "fail", "; ".join(str(exc).splitlines()[:3]))
 
     _render_checks(results, out)
     if any(status == "fail" for _, status, _ in results):