teich 0.2.2__tar.gz → 0.2.3__tar.gz

{teich-0.2.2 → teich-0.2.3}/DOCS.md

@@ -3,6 +3,7 @@
 Teich documentation now lives in focused pages under [`docs/`](docs/index.md).
 
 - [Generation](docs/generation.md): create new Codex, Pi, Claude Code, Hermes, or chat datasets, or extract existing local sessions with `teich extract`.
+- [CLI Reference](docs/cli.md): command usage for `init`, `generate`, `extract`, `convert`, `anonymize`, `studio`, and `pool`.
 - [Teich Studio](docs/studio.md): configure projects, manage prompts, run batches, and save interactive sessions from a browser.
 - [Preparing Data](docs/prepare-data.md): load local files, folders, Hugging Face datasets, `datasets.Dataset` objects, and source mixes.
 - [Training](docs/training.md): use Teich with TRL / Unsloth and apply response-only labels with `mask_data()`.

{teich-0.2.2 → teich-0.2.3}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: teich
-Version: 0.2.2
+Version: 0.2.3
 Summary: Turn coding agent traces into auditable supervised fine-tuning data
 License: Apache-2.0
 License-File: LICENSE
@@ -151,8 +151,13 @@ If the agent store is somewhere other than the default home-directory location,
 
 ```bash
 teich extract claude --sessions-dir /path/to/.claude --out data
+teich extract claude --sessions-dir /path/to/.claude/projects --out data
+teich extract codex --sessions-dir /path/to/.codex --out data
 teich extract codex --sessions-dir /path/to/.codex/sessions --out data
 teich extract pi --sessions-dir /path/to/.pi --out data
+teich extract pi --sessions-dir /path/to/.pi/agent/sessions --out data
+teich extract pi --sessions-dir /path/to/.pi/sessions --out data
+teich extract hermes --sessions-dir /path/to/.hermes --out data
 teich extract hermes --sessions-dir /path/to/.hermes/state.db --out data
 ```
 
@@ -168,6 +173,7 @@ teich convert data --out teich-training.jsonl
 
 | Use case | Start here |
 | --- | --- |
+| Find command examples and options | [CLI Reference](docs/cli.md) |
 | Configure and steer runs in a browser | [Teich Studio](docs/studio.md) |
 | Generate Codex, Pi, Claude Code, Hermes, or chat data | [Generation](docs/generation.md) |
 | Load local files, folders, Hugging Face datasets, or `datasets.Dataset` objects | [Preparing Data](docs/prepare-data.md) |

{teich-0.2.2 → teich-0.2.3}/README.md

@@ -118,8 +118,13 @@ If the agent store is somewhere other than the default home-directory location,
 
 ```bash
 teich extract claude --sessions-dir /path/to/.claude --out data
+teich extract claude --sessions-dir /path/to/.claude/projects --out data
+teich extract codex --sessions-dir /path/to/.codex --out data
 teich extract codex --sessions-dir /path/to/.codex/sessions --out data
 teich extract pi --sessions-dir /path/to/.pi --out data
+teich extract pi --sessions-dir /path/to/.pi/agent/sessions --out data
+teich extract pi --sessions-dir /path/to/.pi/sessions --out data
+teich extract hermes --sessions-dir /path/to/.hermes --out data
 teich extract hermes --sessions-dir /path/to/.hermes/state.db --out data
 ```
 
@@ -133,9 +138,10 @@ teich convert data --out teich-training.jsonl
 
 ## What Teich Supports
 
-| Use case | Start here |
-| --- | --- |
-| Configure and steer runs in a browser | [Teich Studio](docs/studio.md) |
+| Use case | Start here |
+| --- | --- |
+| Find command examples and options | [CLI Reference](docs/cli.md) |
+| Configure and steer runs in a browser | [Teich Studio](docs/studio.md) |
 | Generate Codex, Pi, Claude Code, Hermes, or chat data | [Generation](docs/generation.md) |
 | Load local files, folders, Hugging Face datasets, or `datasets.Dataset` objects | [Preparing Data](docs/prepare-data.md) |
 | Train with TRL / Unsloth while keeping response-only labels correct | [Training](docs/training.md) |

teich-0.2.3/docs/cli.md

@@ -0,0 +1,111 @@
+# CLI Reference
+
+Use `teich --help` for the command list and `teich COMMAND --help` for command-specific options.
+
+## Project Setup
+
+Create a starter project:
+
+```bash
+teich init my-project
+cd my-project
+```
+
+This writes `config.yaml` and `prompts.jsonl`. Edit both, set an API key such as `TEICH_API_KEY`, `OPENROUTER_API_KEY`, or `OPENAI_API_KEY`, then run generation.
+
+## Generate
+
+Run prompts from a config file:
+
+```bash
+teich generate -c config.yaml
+```
+
+Useful options:
+
+- `--config`, `-c`: config path, default `config.yaml`
+- `--output`, `-o`: override `output.traces_dir`
+- `--concurrency`, `-j`: number of prompts to run in parallel
+- `--resume`: skip prompts that already have completed outputs
+
+Generated runs write raw traces, converted JSONL rows, and a dataset `README.md` under `output/` by default. Agent providers also write workspace snapshots under `sandbox/`; failed or interrupted traces go under `failures/`.
+
+## Extract
+
+Extract existing local sessions:
+
+```bash
+teich extract claude --model fable-5 --out data
+```
+
+The first argument is the provider: `claude`, `codex`, `pi`, or `hermes`.
+
+Useful options:
+
+- `--sessions-dir`, `-s`: explicit agent root, native session folder, `state.db`, or JSONL file. Can be passed more than once.
+- `--model`, `-m`: substring match against model metadata. For example, `fable-5` matches `claude-fable-5`.
+- `--output`, `--out`, `-o`: staged dataset directory, default `data`
+- `--no-anon`, `--no-anonymize`: skip automatic anonymization
+- `--private`: if you choose the Hugging Face upload prompt, create the dataset as private
+
+Default stores:
+
+```text
+claude  CLAUDE_CONFIG_DIR/projects, CLAUDE_HOME/projects, ~/.claude/projects
+codex   CODEX_HOME/sessions, ~/.codex/sessions
+pi      PI_SESSION_DIR, PI_CODING_AGENT_DIR/sessions, ~/.pi/agent/sessions, ~/.pi/sessions
+hermes  HERMES_STATE_DB, HERMES_HOME/state.db, ~/.hermes/state.db
+```
+
+Explicit paths can point at either the agent root or the native store:
+
+```bash
+teich extract claude --sessions-dir /path/to/.claude --out data
+teich extract claude --sessions-dir /path/to/.claude/projects --out data
+teich extract codex --sessions-dir /path/to/.codex --out data
+teich extract codex --sessions-dir /path/to/.codex/sessions --out data
+teich extract pi --sessions-dir /path/to/.pi --out data
+teich extract pi --sessions-dir /path/to/.pi/agent/sessions --out data
+teich extract pi --sessions-dir /path/to/.pi/sessions --out data
+teich extract hermes --sessions-dir /path/to/.hermes --out data
+teich extract hermes --sessions-dir /path/to/.hermes/state.db --out data
+```
+
+Extraction anonymizes staged traces by default. Review the staged files before upload; `--no-anon` keeps raw values unchanged.
+
+## Convert
+
+Convert raw or extracted traces into normalized Teich JSONL:
+
+```bash
+teich convert data --out teich-training.jsonl
+```
+
+Each output line contains `prompt`, `messages`, `tools`, and `metadata`. Use this when your training stack can consume OpenAI-style message rows without importing Teich. Use `prepare_data()` and `mask_data()` when you want tokenizer-specific rendering and exact response-only labels.
+
+## Anonymize
+
+Scrub common secrets and local usernames:
+
+```bash
+teich anonymize output --output output_anonymized
+teich anonymize data --in-place
+```
+
+Anonymization replaces API keys, email addresses, and home-directory usernames with deterministic dummy values while preserving embedded base64 media payloads. It is a best-effort pass; review data before publishing.
+
+## Studio
+
+Launch the browser UI:
+
+```bash
+teich studio
+teich studio ./my-project
+teich studio --host 127.0.0.1 --port 8420 --no-open
+```
+
+Studio edits the same `config.yaml` and `prompts.jsonl` files used by `teich generate`.
+
+## Pool
+
+`teich pool upload` is reserved for a future Teich community pool backend. For Hugging Face dataset uploads today, use `teich generate` or `teich extract` and confirm the upload prompt.

{teich-0.2.2 → teich-0.2.3}/docs/data-format.md

@@ -2,6 +2,12 @@
 
 Teich normalizes supported sources into structured training examples.
 
+To write this normalized format as standalone JSONL from raw or extracted traces, run:
+
+```bash
+teich convert data --out teich-training.jsonl
+```
+
 Core fields:
 
 - `prompt`: initial task description

{teich-0.2.2 → teich-0.2.3}/docs/generation.md

@@ -63,6 +63,8 @@ teich extract claude --sessions-dir /path/to/.claude/projects --out data
 teich extract codex --sessions-dir /path/to/.codex --out data
 teich extract codex --sessions-dir /path/to/.codex/sessions --out data
 teich extract pi --sessions-dir /path/to/.pi --out data
+teich extract pi --sessions-dir /path/to/.pi/agent/sessions --out data
+teich extract pi --sessions-dir /path/to/.pi/sessions --out data
 teich extract hermes --sessions-dir /path/to/.hermes --out data
 teich extract hermes --sessions-dir /path/to/.hermes/state.db --out data
 ```

{teich-0.2.2 → teich-0.2.3}/docs/index.md

@@ -5,6 +5,7 @@ Start with the README for a quick overview, then use these pages for implementat
 ## Guides
 
 - [Teich Studio](studio.md): configure projects, manage prompts, run batches, and save interactive sessions from a browser.
+- [CLI Reference](cli.md): command usage for `init`, `generate`, `extract`, `convert`, `anonymize`, `studio`, and `pool`.
 - [Generation](generation.md): create new Codex, Pi, Claude Code, Hermes, or chat datasets.
 - [Preparing Data](prepare-data.md): load local files, folders, Hugging Face datasets, `datasets.Dataset` objects, and source mixes.
 - [Training](training.md): use Teich with TRL / Unsloth and apply response-only labels with `mask_data()`.
@@ -32,6 +33,12 @@ Already have local agent sessions:
 teich extract claude --model fable-5 -> optional HF upload -> prepare_data()
 ```
 
+Want normalized JSONL for another trainer:
+
+```text
+teich extract claude --out data -> teich convert data --out teich-training.jsonl
+```
+
 Prefer a browser:
 
 ```text

{teich-0.2.2 → teich-0.2.3}/docs/pipeline.md

@@ -9,6 +9,7 @@ flowchart TD
     A["Raw source<br/>HF repo, local traces, Dataset, source mix, or prompts"] --> B{"Need generation?"}
     B -->|"yes"| C["teich generate"]
     B -->|"existing local sessions"| X["teich extract"]
+    B -->|"want standalone JSONL"| Y["teich convert"]
     B -->|"no"| D["prepare_data"]
     C --> C1["Run provider<br/>codex, pi, claude-code, hermes, or chat"]
     C1 --> C2["Write raw/native traces or structured chat rows"]
@@ -21,6 +22,8 @@ flowchart TD
     X4 -->|"yes"| X5["Upload JSONL + README"]
     X4 -->|"no"| D
     X5 --> D
+    X3 --> Y
+    Y --> Y1["Write normalized Teich JSONL<br/>prompt, messages, tools, metadata"]
     D --> E["Resolve and load sources"]
     E --> F["Convert traces to messages + tools"]
     F --> G["Render tokenizer chat template"]
@@ -163,6 +166,7 @@ flowchart TD
 ```text
 prepare_data keeps human-readable text plus typed span metadata.
 mask_data converts the selected spans into exact token-level labels after trainer tokenization.
+teich convert writes normalized message JSONL without tokenizer rendering or token labels.
 ```
 
 This lets Teich stay compatible with TRL / Unsloth trainer flows while still controlling exactly what the model learns.

{teich-0.2.2 → teich-0.2.3}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "teich"
-version = "0.2.2"
+version = "0.2.3"
 description = "Turn coding agent traces into auditable supervised fine-tuning data"
 readme = "README.md"
 license = {text = "Apache-2.0"}

{teich-0.2.2 → teich-0.2.3}/src/teich/__init__.py

@@ -1,6 +1,6 @@
 from __future__ import annotations
 
-__version__ = "0.2.2"
+__version__ = "0.2.3"
 
 from .audit import SFTAuditReport, audit_sft_dataset
 from .config import Config, load_config

{teich-0.2.2 → teich-0.2.3}/src/teich/cli.py

@@ -15,6 +15,7 @@ from rich.console import Console, Group
 from rich.live import Live
 from rich.panel import Panel
 from rich.table import Table
+from typer.core import TyperCommand, TyperGroup
 
 from .anonymize import anonymize_path
 from .config import Config
@@ -36,19 +37,141 @@ from .trace_readme import write_traces_readme
 from .tool_schema import snapshot_configured_tools
 
 console = Console()
+ROOT_EXTRA_HELP = """\
+Common workflows:
+  teich init my-project
+  teich generate -c config.yaml
+  teich extract claude --model fable-5 --out data
+  teich convert data --out teich-training.jsonl
+  teich studio
+
+For command-specific options and examples:
+  teich generate --help
+  teich extract --help
+  teich convert --help
+  teich studio --help
+"""
+EXTRACT_HELP = "Extract existing local agent sessions into a staged Teich dataset folder."
+EXTRACT_EXTRA_HELP = """\
+Common examples:
+  teich extract claude --model fable-5 --out data
+  teich extract claude --sessions-dir /path/to/.claude --out data
+  teich extract claude --sessions-dir /path/to/.claude/projects --out data
+  teich extract codex --sessions-dir /path/to/.codex --out data
+  teich extract codex --sessions-dir /path/to/.codex/sessions --out data
+  teich extract pi --sessions-dir /path/to/.pi --out data
+  teich extract pi --sessions-dir /path/to/.pi/agent/sessions --out data
+  teich extract pi --sessions-dir /path/to/.pi/sessions --out data
+  teich extract hermes --sessions-dir /path/to/.hermes --out data
+  teich extract hermes --sessions-dir /path/to/.hermes/state.db --out data
+
+Default stores:
+  claude  CLAUDE_CONFIG_DIR/projects, CLAUDE_HOME/projects, ~/.claude/projects
+  codex   CODEX_HOME/sessions, ~/.codex/sessions
+  pi      PI_SESSION_DIR, PI_CODING_AGENT_DIR/sessions, ~/.pi/agent/sessions, ~/.pi/sessions
+  hermes  HERMES_STATE_DB, HERMES_HOME/state.db, ~/.hermes/state.db
+
+After extraction:
+  teich convert data --out teich-training.jsonl
+  This writes normalized Teich JSONL rows with prompt, messages, tools, and metadata for trainers that do not import Teich.
+"""
+GENERATE_EXTRA_HELP = """\
+Typical generated project flow:
+  teich init my-project
+  cd my-project
+  edit prompts.jsonl and config.yaml
+  teich generate -c config.yaml
+  teich generate -c config.yaml --resume
+
+Outputs:
+  output/    raw traces, converted JSONL rows, and README.md
+  sandbox/   workspace snapshots for agent providers
+  failures/  failed or interrupted traces excluded from resume/upload
+
+Use teich extract instead when you already have local Claude, Codex, Pi, or Hermes sessions.
+"""
+CONVERT_EXTRA_HELP = """\
+Examples:
+  teich convert data --out teich-training.jsonl
+  teich convert output/session.jsonl --out session.training.jsonl
+
+Output format:
+  Newline-delimited JSON rows with prompt, messages, tools, and metadata.
+  This is useful when your trainer can consume OpenAI-style message rows without importing Teich.
+
+Use prepare_data() and mask_data() when you want tokenizer-specific rendering and exact response-only labels.
+"""
+ANONYMIZE_EXTRA_HELP = """\
+Examples:
+  teich anonymize output --output output_anonymized
+  teich anonymize data --in-place
+
+What is scrubbed:
+  API keys, email addresses, and home-directory usernames.
+  Embedded base64 media payloads are preserved for conversation context.
+
+This is a best-effort safety pass. Review data before publishing or uploading.
+"""
+INIT_EXTRA_HELP = """\
+Creates:
+  config.yaml    generation settings
+  prompts.jsonl  starter prompt rows
+
+Next:
+  edit config.yaml and prompts.jsonl
+  set TEICH_API_KEY, OPENROUTER_API_KEY, or OPENAI_API_KEY
+  teich generate -c config.yaml
+"""
+STUDIO_EXTRA_HELP = """\
+Examples:
+  teich studio
+  teich studio ./my-project
+  teich studio --host 127.0.0.1 --port 8420 --no-open
+
+Studio edits the same config.yaml and prompts.jsonl files used by teich generate.
+"""
+POOL_EXTRA_HELP = """\
+This namespace is reserved for a future Teich community pool upload backend.
+For Hugging Face dataset uploads today, use teich generate or teich extract and confirm the upload prompt.
+"""
+
+
+class ExtraHelpCommand(TyperCommand):
+    extra_help = ""
+
+    def format_help(self, ctx: typer.Context, formatter: typer.HelpFormatter) -> None:
+        super().format_help(ctx, formatter)
+        if self.extra_help:
+            formatter.write("\n" + self.extra_help)
+
+
+class ExtraHelpGroup(TyperGroup):
+    extra_help = ""
+
+    def format_help(self, ctx: typer.Context, formatter: typer.HelpFormatter) -> None:
+        super().format_help(ctx, formatter)
+        if self.extra_help:
+            formatter.write("\n" + self.extra_help)
+
+
+def _help_command(name: str, extra_help: str) -> type[ExtraHelpCommand]:
+    return type(name, (ExtraHelpCommand,), {"extra_help": extra_help})
+
+
+def _help_group(name: str, extra_help: str) -> type[ExtraHelpGroup]:
+    return type(name, (ExtraHelpGroup,), {"extra_help": extra_help})
+
+
 app = typer.Typer(
     name="teich",
-    help="Generate agent training data using Codex, Pi, Claude Code, Hermes, or chat",
+    help="Generate, extract, convert, prepare, and inspect agent training data.",
     no_args_is_help=True,
+    cls=_help_group("TeichHelpGroup", ROOT_EXTRA_HELP),
 )
-extract_app = typer.Typer(
-    help=(
-        "Extract local agent sessions into Teich output folders. Use a provider subcommand with "
-        "--sessions-dir PATH to read a .claude, .codex, .pi, or .hermes folder outside the default location."
-    )
+pool_app = typer.Typer(
+    help="Reserved Teich community pool commands.",
+    cls=_help_group("PoolHelpGroup", POOL_EXTRA_HELP),
 )
-pool_app = typer.Typer(help="Upload anonymized traces to the Teich community pool.")
-app.add_typer(extract_app, name="extract")
 app.add_typer(pool_app, name="pool")
 NON_DATA_TRACE_DIR_NAMES = {"partials", "failures"}
 UPLOAD_IGNORE_PATTERNS = ["partials/**", "failures/**"]
@@ -356,8 +479,9 @@ def _extract_sources_option() -> list[Path] | None:
         "--sessions-dir",
         "-s",
         help=(
-            "Explicit agent data folder or file. Examples: .claude, .claude/projects, "
-            ".codex, .codex/sessions, .pi, .hermes, or .hermes/state.db. Can be passed more than once."
+            "Explicit agent root, native session folder, state.db file, or JSONL file. "
+            "Examples: .claude, .claude/projects, .codex, .codex/sessions, .pi, "
+            ".pi/agent/sessions, .pi/sessions, .hermes, .hermes/state.db. Can be passed more than once."
         ),
     )
 
@@ -374,55 +498,33 @@ def _extract_no_anonymize_option() -> bool:
     return typer.Option(False, "--no-anon", "--no-anonymize", help="Skip automatic anonymization of extracted traces.")
 
 
-@extract_app.command()
-def codex(
-    output: Path = _extract_output_option(),
-    sessions_dir: list[Path] | None = _extract_sources_option(),
-    model: str | None = _extract_model_option(),
-    no_anon: bool = _extract_no_anonymize_option(),
-    private: bool = typer.Option(False, "--private", help="Create the Hugging Face dataset as private if upload is confirmed."),
-) -> None:
-    """Extract local Codex sessions."""
-    _run_extract_command("codex", output, sessions_dir, model_filter=model, skip_anonymize=no_anon, private=private)
-
-
-@extract_app.command()
-def claude(
-    output: Path = _extract_output_option(),
-    sessions_dir: list[Path] | None = _extract_sources_option(),
-    model: str | None = _extract_model_option(),
-    no_anon: bool = _extract_no_anonymize_option(),
-    private: bool = typer.Option(False, "--private", help="Create the Hugging Face dataset as private if upload is confirmed."),
-) -> None:
-    """Extract local Claude Code sessions."""
-    _run_extract_command("claude", output, sessions_dir, model_filter=model, skip_anonymize=no_anon, private=private)
-
-
-@extract_app.command()
-def pi(
-    output: Path = _extract_output_option(),
-    sessions_dir: list[Path] | None = _extract_sources_option(),
-    model: str | None = _extract_model_option(),
-    no_anon: bool = _extract_no_anonymize_option(),
-    private: bool = typer.Option(False, "--private", help="Create the Hugging Face dataset as private if upload is confirmed."),
-) -> None:
-    """Extract local Pi sessions."""
-    _run_extract_command("pi", output, sessions_dir, model_filter=model, skip_anonymize=no_anon, private=private)
-
-
-@extract_app.command()
-def hermes(
+@app.command(
+    help=EXTRACT_HELP,
+    short_help="Extract existing local agent sessions.",
+    no_args_is_help=True,
+    cls=_help_command("ExtractHelpCommand", EXTRACT_EXTRA_HELP),
+)
+def extract(
+    provider: ExtractProvider = typer.Argument(
+        ...,
+        metavar="PROVIDER",
+        help="Provider to extract: claude, codex, pi, or hermes.",
+    ),
     output: Path = _extract_output_option(),
     sessions_dir: list[Path] | None = _extract_sources_option(),
     model: str | None = _extract_model_option(),
     no_anon: bool = _extract_no_anonymize_option(),
     private: bool = typer.Option(False, "--private", help="Create the Hugging Face dataset as private if upload is confirmed."),
 ) -> None:
-    """Extract local Hermes Agent state sessions."""
-    _run_extract_command("hermes", output, sessions_dir, model_filter=model, skip_anonymize=no_anon, private=private)
+    """Extract local sessions for one provider."""
+    _run_extract_command(provider, output, sessions_dir, model_filter=model, skip_anonymize=no_anon, private=private)
 
 
-@app.command()
+@app.command(
+    help="Copy trace files and scrub common secrets/user identifiers.",
+    short_help="Anonymize trace files.",
+    cls=_help_command("AnonymizeHelpCommand", ANONYMIZE_EXTRA_HELP),
+)
 def anonymize(
     input_path: Path = typer.Argument(Path("output"), help="Trace file or directory to anonymize"),
     output: Path = typer.Option(
@@ -452,7 +554,11 @@ def anonymize(
         console.print("[yellow]No emails, usernames, or API keys were detected.[/yellow]")
 
 
-@app.command()
+@app.command(
+    help="Convert raw or extracted traces into normalized Teich JSONL rows.",
+    short_help="Convert traces to Teich JSONL.",
+    cls=_help_command("ConvertHelpCommand", CONVERT_EXTRA_HELP),
+)
 def convert(
     input_path: Path = typer.Argument(..., help="Raw trace JSONL file or folder to convert"),
     output: Path = typer.Option(
@@ -485,7 +591,11 @@ def convert(
     console.print("[cyan]Output format:[/cyan] Teich JSONL with prompt, messages, tools, and metadata.")
 
 
-@pool_app.command("upload")
+@pool_app.command(
+    "upload",
+    help="Reserved placeholder for future Teich pool uploads.",
+    short_help="Reserved pool upload placeholder.",
+)
 def pool_upload(
     path: Path = typer.Argument(Path("output_anonymized"), help="Anonymized trace directory to upload later"),
 ) -> None:
@@ -496,7 +606,11 @@ def pool_upload(
     raise typer.Exit(1)
 
 
-@app.command()
+@app.command(
+    help="Run a configured prompt batch and write trace/data artifacts.",
+    short_help="Generate traces from config.",
+    cls=_help_command("GenerateHelpCommand", GENERATE_EXTRA_HELP),
+)
 def generate(
     config: Path = typer.Option(
         Path("config.yaml"),
@@ -835,7 +949,11 @@ class BatchProgressReporter:
             return Group(Panel.fit(summary, title="Batch Summary"), table)
 
 
-@app.command()
+@app.command(
+    help="Create a starter config.yaml and prompts.jsonl project.",
+    short_help="Initialize a Teich project.",
+    cls=_help_command("InitHelpCommand", INIT_EXTRA_HELP),
+)
 def init(
     path: Path = typer.Argument(Path("."), help="Directory to initialize"),
 ) -> None:
@@ -1016,7 +1134,11 @@ def _configure_studio_event_loop_policy(
     return True
 
 
-@app.command()
+@app.command(
+    help="Launch the local Teich Studio browser UI.",
+    short_help="Launch Teich Studio.",
+    cls=_help_command("StudioHelpCommand", STUDIO_EXTRA_HELP),
+)
 def studio(
     path: Path = typer.Argument(Path("."), help="Project directory (created/initialized if needed)"),
     host: str = typer.Option("127.0.0.1", "--host", help="Host to bind the studio server to"),

{teich-0.2.2 → teich-0.2.3}/tests/test_extract_anonymize_cli.py

@@ -37,16 +37,40 @@ def _write_minimal_hermes_state_db(state_db: Path, *, session_id: str = "session
         connection.close()
 
 
-def test_extract_subcommands_exist():
+def test_extract_help_covers_providers_options_defaults_and_examples():
+    root_help = runner.invoke(app, ["--help"])
+
+    assert root_help.exit_code == 0
+    root_output = _plain_cli_output(root_help.output)
+    assert "Extract existing local agent sessions" in root_output
+    assert "teich extract claude --model fable-5 --out data" in root_output
+
     group_help = runner.invoke(app, ["extract", "--help"])
 
     assert group_help.exit_code == 0
     group_output = _plain_cli_output(group_help.output)
-    assert "--sessions-dir PATH" in group_output
+    assert "PROVIDER" in group_output
+    for provider in ("claude", "codex", "pi", "hermes"):
+        assert provider in group_output
+    assert "--sessions-dir" in group_output
+    assert "--output,--out" in group_output
+    assert "--model" in group_output
+    assert "--no-anon" in group_output
+    assert "--no-anonymize" in group_output
     assert ".claude" in group_output
+    assert ".claude/projects" in group_output
     assert ".codex" in group_output
+    assert ".codex/sessions" in group_output
     assert ".pi" in group_output
+    assert ".pi/agent/sessions" in group_output
+    assert ".pi/sessions" in group_output
     assert ".hermes" in group_output
+    assert ".hermes/state.db" in group_output
+    assert "CLAUDE_CONFIG_DIR/projects" in group_output
+    assert "CODEX_HOME/sessions" in group_output
+    assert "HERMES_STATE_DB" in group_output
+    assert "fable-5" in group_output
+    assert "teich convert data --out teich-training.jsonl" in group_output
 
     for provider in ("claude", "codex", "hermes", "pi"):
         result = runner.invoke(app, ["extract", provider, "--help"])

{teich-0.2.2 → teich-0.2.3}/uv.lock

@@ -1734,7 +1734,7 @@ wheels = [
 
 [[package]]
 name = "teich"
-version = "0.2.2"
+version = "0.2.3"
 source = { editable = "." }
 dependencies = [
     { name = "datasets" },