# Taac Competition Environment

> Use before running TAAC repository Python commands, pytest, console scripts, packaging/docs CLIs, or run.sh, and when working on local setup, uv/CUDA dependency flow, online training or inference bundles, package contents, platform Python execution, or environment/debug docs. Use when this capability is needed.

- Skill: `tomevault-io/taac-competition-environment` (Agent Skill, multi-file: 2 files)
- Install (CLI): `npx skillmds@latest add tomevault-io/taac-competition-environment`
- Raw SKILL.md: https://api.skillmd.com/api/skills/tomevault-io/taac-competition-environment/raw
- Safety review: pending (external: skill-scanner PASS, skillspector PASS)
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: AI & ML
- Author: tomevault-io (https://skillmd.com/u/tomevault-io)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/tomevault-io/taac-competition-environment

---


# TAAC Competition Environment

Use this skill to find the live environment contract quickly. Do not copy behavior from this file when the source has changed.

## Command Contract

- In the local repository, assume the system `python`/`python3` may be absent. Use `uv run python ...`, `uv run pytest ...`, and `uv run <console-script> ...` on the first attempt.
- Use bare `python` only for generated online bundle simulation or platform docs where `TAAC_RUNNER=python`/`TAAC_PYTHON` is the behavior under test.
- `bash run.sh train|val|eval|infer` is valid locally because `run.sh` dispatches through `uv` outside bundle mode.

## Read First

For local vs online runner behavior:

- `run.sh`
- `src/taac2026/application/bootstrap/run_sh.py`
- `src/taac2026/infrastructure/platform/{deps,env,imports}.py`
- `docs/guide/competition-online-server.md`
- `docs/guide/online-training-bundle.md`
- `docs/getting-started.md`

For package layout:

- `src/taac2026/application/packaging/{cli,service}.py`
- `src/taac2026/infrastructure/bundles/`
- `src/taac2026/application/bootstrap/inference_bundle*.py`
- `tests/unit/application/packaging/`
- `tests/unit/application/bootstrap/`

For dependencies:

- `pyproject.toml`
- `uv.lock`
- docs that mention setup or bundle execution

## Non-Obvious Context

- Local repository commands are expected to use `uv`; generated online bundles must not require `uv`.
- Bundle validation must inspect what gets zipped, not just whether local imports work from the repository root.
- The online platform may provide its own Python/CUDA stack. Avoid changes that make bundle mode depend on dev extras, repo-local files, or `uv.lock`.
- `docs/archive/files/...` are historical competition references, not active runtime code.
- Archived schema fixtures live under `docs/archive/files/schema/`, including `sample_1000_raw.schema.json` and `online_schema.schema.json`. If a local smoke command fails because `data/sample_1000_raw/schema.json` is absent and the command supports it, pass `--schema-path docs/archive/files/schema/sample_1000_raw.schema.json` explicitly.

## Validation

For environment/bootstrap changes:

```bash
uv run pytest tests/unit/application/bootstrap -q
uv run pytest tests/unit/application/packaging -q
```

For bundle-impacting changes, build and inspect at least one tiny bundle:

```bash
uv run taac-package-train --experiment experiments/baseline --output-dir /tmp/taac-training-bundle --json
uv run taac-package-infer --experiment experiments/baseline --output-dir /tmp/taac-inference-bundle --json
uv run python -m zipfile -l /tmp/taac-training-bundle/code_package.zip | sed -n '1,80p'
```

For docs-only environment edits, validate docs with the docs skill instead of running the full unit suite by default.

---
> Source: [Puiching-Memory/TAAC_2026](https://github.com/Puiching-Memory/TAAC_2026) — distributed by [TomeVault](https://tomevault.io).
<!-- tomevault:4.0:skill_md:2026-06-29 -->