Metadata-Version: 2.4
Name: fable-dev-shell
Version: 0.1.0
Summary: Fable: a production multi-agent developer shell.
Requires-Python: >=3.10
Description-Content-Type: text/markdown
Requires-Dist: prompt_toolkit
Requires-Dist: rich
Requires-Dist: huggingface_hub
Requires-Dist: kaggle
Requires-Dist: python-dotenv
Requires-Dist: anthropic
Requires-Dist: requests

# FABLE

`fable` is an interactive Python REPL shell for routing development work across
Codex, Claude Code, Antigravity, Hugging Face dataset search, Kaggle training,
custom skills, and goal-based orchestration.

Install locally:

```bash
pip install -e .
```

Install from PyPI after release:

```bash
pipx install fable-dev-shell
fable
```

or:

```bash
pip install fable-dev-shell
fable
```

Run:

```bash
fable
```

Technical documentation:

```text
docs/TECHNICAL_DOCUMENTATION.md
```

Agent runs show Codex-style generation metadata, live provider output, provider
diagnostics, and a workspace change table. Transcripts are saved under
`~/.fable/runs`. Inside the shell, use `/runs` and `/runs show <id>` to inspect
past generations, including streamed output and files created or modified.

Provider prompts are passed through as the user's text only. `fable` does not
print internal lane instructions as chat content. On Windows, Codex defaults to
`--sandbox danger-full-access` to avoid the missing
`codex-windows-sandbox-setup.exe` failure; set `FABLE_CODEX_SANDBOX` to override.

Production control commands:

```text
/settings show
/settings set codex_sandbox_windows danger-full-access
/mcp add filesystem npx -y @modelcontextprotocol/server-filesystem C:\PROJECTS
/mcp list
/mcp doctor
/mcp export codex
/mcp export claude
```

Deployment commands:

```text
/login github
/login vercel
/login supabase
/login openclaw
deploy status
deploy github
deploy vercel --prod
deploy supabase status
github repo view
vercel ls
supabase status
openclaw status
marketing audit my landing page and write launch copy
openclaw configure
openclaw channels add
openclaw channels status
```

Ponytail commands:

```text
ponytail build the login backend with the least code that is still secure
ponytail-review src/auth
ponytail-audit .
ponytail-debt app
ponytail-gain api
mode ponytail
ponytail install codex
ponytail install agy
ponytail install openclaw
```

`ponytail` runs a Codex lane with a minimum-necessary engineering discipline:
reuse existing code first, prefer stdlib/native/dependencies before new code,
and keep validation, security, accessibility, tests, and required behavior.
The install subcommands call the official Ponytail integrations for Codex,
Antigravity, and OpenClaw.

If an agent run fails, or exits with no output and no workspace changes, `fable`
can automatically run one Codex repair pass. Control it with:

```text
/settings set auto_repair true
/settings set auto_repair_on_noop true
```

For full app builds, use `/goal ... --parallel` to run independent backend,
frontend, dataset search, and training lanes at the same time while dependent
polish/integration steps wait for them.

If Anthropic planning or Claude frontend execution is unavailable, `fable`
falls back to separate Codex lanes for backend and frontend work.


