- None.
- Added interactive Command-Line template wizard:
ztxexp init-template- 7-question guided scaffold assembly for experiment scripts
- Added generated experiment skeleton protocol under
experiments/<name>/:main_experiment.py(withrun/analyze/cleansubcommands)configs/base.jsonartifacts/.gitkeep- optional
models/.gitkeep modules/<module>/__init__.pystubs
- Added soft prerequisite checks for vibe/skill initialization:
- missing
init-vibe/init-skillonly emits warnings and does not block template generation
- missing
- Added template management marker:
.ztxexp-managed-template.json- safe overwrite guard for unmanaged directories (
--forcerequired)
- Added automated tests for template CLI workflow:
tests/test_cli_template.py
- Improved non-interactive robustness for template initialization:
--no-interactivenow requires explicit--nameto avoid accidental anonymous scaffold creation.
- Added
init-templateusage to README CLI section. - Added Vibe Coding workflow section:
init-vibe -> init-skill -> init-template.
- Upgrade directly from
1.0.2:pip install -U ztxexp
- Quick scaffold generation:
ztxexp init-template --name my_experiment --no-interactive
- None.
- Added first-party skills integration for agent ecosystems:
- repository built-in skill directory:
skills/ztx-exp-manager/ - built-in skill files:
SKILL.md+agents/openai.yaml
- repository built-in skill directory:
- Added CLI subcommands for skill lifecycle management:
ztxexp init-skillztxexp show-skillztxexp remove-skill
- Added interactive target selection for
init-skillwhen--targetis omitted:1->skills/2->.codex/skills/3-> both
- Added managed marker protocol for safe skill updates/removals:
.ztxexp-managed-skill.json- default remove policy only deletes managed installs
- Added automated CLI and behavior tests for skill workflows:
tests/test_cli_skill.py
- Improved safety for skill deletion by skipping unmanaged directories unless
--forceis used. - Improved non-interactive behavior by defaulting
init-skilltoskills/when prompt is disabled.
- Added README sections for
init-skill/show-skill/remove-skillusage and options.
- Upgrade directly from
1.0.1:pip install -U ztxexp
- Initialize built-in skill in any target project:
ztxexp init-skill
- None.
- Added first-party CLI entrypoint
ztxexpwith subcommands:ztxexp init-vibeztxexp show-vibeztxexp remove-vibe
- Added agent integration persistence module
ztxexp.vibe:- managed block markers (
<!-- ztxexp:vibe:start --> ... <!-- ztxexp:vibe:end -->) - idempotent write/update/remove behaviors
- profile and language rendering (
webcoding/codex/cursor/cline/copilot,bilingual/zh/en)
- managed block markers (
- Added module entrypoint support for
python -m ztxexp.
- Added deterministic AGENTS file selection policy for cross-project initialization:
- reuse existing
AGENTS.md/agents.md/agents.MD - fallback create
AGENTS.md
- reuse existing
- Added CLI usage section in README for
init-vibe/show-vibe/remove-vibe.
- Upgrade directly from
1.0.0:pip install -U ztxexp
- After upgrade, initialize agent guidance in any target project:
ztxexp init-vibe
- None.
- None.
- Fixed release ordering confusion caused by historical
0.30.0and0.4.0coexistence. - Normalized package versioning baseline to
1.0.0to ensure unambiguous upgrade ordering on package index.
- None.
- Users on
0.30.0/0.4.0can upgrade directly to1.0.0with:pip install -U ztxexp
- None.
- Added reproducibility/governance dataclasses:
RunMetadataMetricEvent
- Added
RunContextenhancements:metalog_metric(step, metrics, split, phase)
- Added run artifact extensions (schema v2 compatible):
meta.jsonmetrics.jsonlevents.jsonlcheckpoints/
- Added
ExperimentPipelinegovernance APIs:name(...),group(...),tags(...),lineage(...)retry(...),track(...),random_search(...)
- Added tracker system:
Trackerprotocol- built-in
JsonlTracker - optional
MlflowTracker/WandbTrackeradapters
- Added CI workflow:
ruff check .pytestmkdocs build --strictpython -m build --no-isolationtwine check dist/*
- Added migration guide:
docs_src/migration-v04.zh.md
- Added template smoke test suite:
tests/test_templates_smoke.py
- Fixed docs build hard-failure by restoring
docs_src/directory. - Fixed mismatch between changelog quality claims and actual CI presence.
- Improved analyzer query capability with
experiment_name/group/tagsfilters. - Improved template docs generation with scenario copy matrix page.
- Refined dependency layering:
mlflow/wandbremain optional extras and are no longer installed viadevby default.
- README now includes v0.4 governance and reproducibility features.
- Template docs now include
场景复制矩阵for direct copy commands.
- Existing
0.3API usage remains valid; upgrades are additive. - Recommended migration:
- Add governance metadata (
name/group/tags/lineage). - Replace manual step logs with
ctx.log_metric(...). - Enable optional trackers via extras when needed.
- Add governance metadata (
- Bumped project version baseline from
0.2.0to0.3.0. - Updated package/runtime-facing version strings in metadata and examples.
- Added a copy-first template library under
examples/template_librarywith 27 runnable templates. - Added scenario coverage across basics, parallel scheduling, analysis/cleanup, ML, LLM, and ops workflows.
- Added automatic MkDocs template pages generation (
示例模板库) from template source files.
- README now highlights template-library-first workflow and direct copy paths.
- Template index table is generated and available in docs navigation.
- Upgraded runtime artifact schema to v2 (
config.json,run.json, optionalmetrics.json,artifacts/). - Experiment function contract changed to
exp_fn(ctx: RunContext) -> dict | None. - Success is now defined by
run.json.status == "succeeded". ResultAnalyzer.clean_resultsswitched from marker-based cleanup tostatuses/predicateAPI.- v0.2 no longer reads legacy run directory formats.
RunContextandRunSummarypublic dataclasses.ExperimentPipelinefacade for build + run workflows.- Experimental dynamic scheduler with bounded submission based on CPU threshold.
- Test suite for manager/runner/analyzer/pipeline.
- CI workflow for lint, tests, docs build, and package build.
- Completed-run filtering now checks v2 run status instead of only checking argument files.
- Config equality check now requires strict keyset equivalence.
- Analyzer CSV export now supports custom metrics filename pass-through.
- Top-level package exports now match documented usage.
- Packaging metadata now points to
README.md.
- Rewrote README around problem/solution/quickstart/common pitfalls/API map.
- Reorganized docs with Chinese mainline and English entry pages.
- Added v0.2 migration guide.
- Reworked examples for minimal run, parallel run, analysis, cleanup, and LLM template.