scitex-dev 0.34.0__tar.gz → 0.36.0__tar.gz

{scitex_dev-0.34.0 → scitex_dev-0.36.0}/.github/workflows/cla.yml

@@ -61,11 +61,11 @@ jobs:
           PERSONAL_ACCESS_TOKEN: ${{ secrets.GH_PERSONAL_ACCESS_TOKEN }}
         with:
           path-to-signatures: "signatures/cla.json"
-          path-to-document: "https://github.com/ywatanabe1989/scitex-dev/blob/main/CLA.md"
+          path-to-document: "https://github.com/scitex-ai/scitex-dev/blob/main/CLA.md"
           branch: "cla-signatures"
           allowlist: bot*,ywatanabe1989,LLEmacs
           custom-allsigned-prcomment: |
             Thank you for signing the SciTeX CLA. Your contribution can now be reviewed.
           custom-notsigned-prcomment: |
-            Please sign the [SciTeX CLA](https://github.com/ywatanabe1989/scitex-dev/blob/main/CLA.md) before your contribution can be merged.
+            Please sign the [SciTeX CLA](https://github.com/scitex-ai/scitex-dev/blob/main/CLA.md) before your contribution can be merged.
             Comment `I have read and agree to the SciTeX CLA.` to sign.

scitex_dev-0.36.0/.scitex/dev/config.yaml

@@ -0,0 +1,49 @@
+# Per-repo overrides for scitex-dev auditors.
+# See _skills/general/01_ecosystem_06_local-state-directories.md and
+# _skills/general/02_package_06_project-structure-tests.md for the full
+# config schema and rule index.
+
+audit:
+  # PS-220 (no bare `print` — use scitex-logging) is reported but does NOT
+  # fail THIS repo's own audit yet. scitex-dev is the repository that DEFINES
+  # and enforces PS-220, so read that sentence twice: the rule's own author is
+  # currently exempt from it at warning level.
+  #
+  # REASON: scitex-dev's own source carries 132 flagged `print` sites (87
+  # prose-to-stdout, 33 stderr, 9 undecidable payload, 3 undecidable
+  # destination). Migrating them is NOT a mechanical rename: scitex-logging
+  # writes every console record to STDERR (scitex-logging/_handlers.py:52-101),
+  # so converting a `print` that currently writes to STDOUT MOVES THAT OUTPUT
+  # TO STDERR and can break any consumer parsing this CLI's stdout. That is a
+  # behavioural change to a parsed interface the whole fleet shells out to, and
+  # it needs its own reviewed PR with per-site judgement about what is a
+  # message and what is payload. A 132-entry `audit.exemptions` list was
+  # rejected as the alternative: every reason would read "not migrated yet",
+  # which is exactly the reasonless park the mandatory-reason design exists to
+  # catch.
+  #
+  # This is the declared migration state, not a suppression:
+  #   * the rule stays at ERROR for every other package — enforcement is not
+  #     weakened for anyone else;
+  #   * `warning`, never `off` — all 132 findings stay VISIBLE on every audit
+  #     run, so the debt cannot go quiet. See all 132 with (verified
+  #     2026-07-22 — prints 132 findings, exits 0):
+  #       scitex-dev ecosystem audit-project scitex-dev \
+  #           --severity warning --rule PS-220
+  #     (the flag is `--rule`, singular and repeatable — not `--rules`)
+  #
+  # REMOVAL CONDITION — this key is retired by scitex-todo card
+  #   ps220-selfmigrate-scitex-dev
+  # ("[PS-220] Migrate scitex-dev's own 132 print sites to scitex-logging,
+  # then delete audit.enforce-logging"), owner scitex-dev. That card's step 3
+  # IS the deletion of this key; its step 4 is scitex-dev's self-audit passing
+  # with PS-220 at E. Introduced by PR #406.
+  enforce-logging: warning
+
+  # Whitelisted top-level directories/files that PS-103 would otherwise
+  # flag as "junk at repo root". Empty for now; add an entry only when the
+  # artefact genuinely belongs in this repo's root (otherwise relocate to
+  # ./.dev/ or ./GITIGNORED/).
+  root-whitelist:
+    dirs: []
+    files: []

{scitex_dev-0.34.0 → scitex_dev-0.36.0}/CHANGELOG.md

@@ -5,6 +5,146 @@ All notable changes to `scitex-dev` are documented here.
 The format follows [Keep a Changelog](https://keepachangelog.com/en/1.1.0/);
 versions follow [Semantic Versioning](https://semver.org/).
 
+## [Unreleased]
+
+## [0.36.0] - 2026-07-23
+
+### Changed
+- **PS-220 restaged: WARNING by default, error per-package opt-in.** 0.35.0
+  (#406) promoted PS-220 to ERROR ecosystem-wide. The measured blast radius
+  — 44 repos newly FAILING on 1856 findings, top-5 repos carrying 64 % of
+  them — led the operator to restage the rollout (Telegram 1691/1692):
+  「とりあえず warning で、移行できたものから red で」. The rule now
+  reports at `W` for every project type, and a package opts IN to an
+  error-level gate when its migration lands:
+
+  ```yaml
+  audit:
+    enforce-logging:
+      level: error
+      reason: "print migration complete (PR #412)"
+  ```
+
+  This is staging, not retreat: every finding stays visible on every audit
+  run, and nothing about what the rule DETECTS changed.
+
+- **`audit.enforce-logging` now demands a written reason.** `error` and
+  `off` deviate from the default and so carry a MANDATORY `reason`;
+  `warning` is accepted bare because it IS the default and changes
+  nothing. A missing or whitespace-only reason is a HARD CONFIG ERROR, not
+  a silent default: the declaration does NOT take effect (the project
+  falls back to `W`) AND the rejection is reported at `E`, so a package
+  can never believe it is gated — or silenced — when it is not. The old
+  bare shorthands `enforce-logging: error` / `enforce-logging: off` (and
+  their YAML 1.1 boolean spellings `on` / `off`) are rejected for exactly
+  this reason. Unrecognised values are now rejected loudly instead of
+  falling back silently. Parsing lives in
+  `_cli/audit/_config/_enforce_logging.py`.
+
+  Config errors are NOT staged: a rejected exemption entry and a rejected
+  enforce-logging declaration are both reported at `E` regardless of the
+  project's PS-220 severity. Staging covers migration debt; a malformed
+  override is not debt.
+
+### Removed
+- **The blanket `# noqa` hatch for PS-220, and its `PS-220-noqa-deprecated`
+  notice.** Deprecated in 0.35.0 for one release. A sweep of all 118 repos
+  under `~/proj` (8956 `src/**.py` files, 4448 flagged sites) found ZERO
+  sites using it, with a planted-user control confirming the sweep could
+  detect one. `audit.exemptions` — pinned to one rule at one file:line,
+  with a mandatory reason — is now the only per-site opt-out.
+
+### Fixed
+- **`--version` no longer answers for another package (#409).** `scitex-ui
+  --version` and `scitex-app --version` printed `scitex-dev <scitex-dev's
+  version>`; 36 downstream packages reported another package's identity.
+  `scitex_dev/_cli/__init__.py` fast-paths a bare `--version` to skip the
+  expensive `._root` import, but the predicate matched on `sys.argv[1:]`
+  alone and never checked WHO was asking. Because `scitex_dev._cli` is a
+  shared primitive that downstream CLIs import (`._completion`, `._root`),
+  the block fired during `import` in any process launched as `<other-cli>
+  --version`: it printed scitex-dev's name and version and raised
+  `SystemExit(0)` before the downstream CLI's own `main()` ran. Exit code 0,
+  so nothing looked broken — and the downstream packages' own `--version`
+  wiring was correct all along. The fast path now additionally requires
+  `sys.argv[0]` to identify scitex-dev's own console script (or `python -m
+  scitex_dev`); an unrecognised `argv[0]` declines the optimisation and
+  falls through to the real Click group — correct, just slower. No path
+  falls back to printing `scitex-dev`; that fallback WAS the defect.
+  A library must never read `sys.argv` or exit at import time.
+
+## [0.35.0] - 2026-07-22
+
+### Added
+- **PS-220 (no bare `print` in package source) promoted to ERROR (#406).**
+  The no-bare-print mandate is now enforced, not advisory, for SciTeX
+  ecosystem packages. Machine-readable stdout is spared STRUCTURALLY — a
+  stdout `print` whose sole argument is a serializer call or a rendered
+  payload variable does not fire, because scitex-logging writes to stderr
+  and would corrupt a `--json` payload or piped data. Everything else,
+  including any undecidable destination, fires and needs a per-site
+  `audit.exemptions` entry carrying a MANDATORY written reason; the
+  legacy blanket `# noqa` hatch is deprecated (`PS-220-noqa-deprecated`,
+  `W`) and honoured for one more release. Severity is project-type
+  scoped: an explicit `audit.enforce-logging` (`error` / `warning` /
+  `off`) always wins, otherwise a `research` project type resolves to `W`
+  rather than wedging a publish on a decision nobody has made.
+
+  Fixes a latent hole found while implementing it: `_SEVERITY_OVERRIDES`
+  was applied BEFORE the co-located rule registrations, making it a
+  SILENT no-op for 31 rules — adding `"PS-220": "E"` to the override
+  table did nothing at all, with no error and no warning. A severity
+  table that silently ignores entries is itself a gate that cannot fail.
+
+- **`audit.skip-rules` — sanctioned per-rule deferrals, honoured by
+  `audit-all` natively.** Declaring a deferral to a named migration
+  campaign in `<repo>/.scitex/dev/config.yaml` now affects BOTH gates.
+  Previously the per-repo pytest wrapper honoured its `skip_rules=`
+  kwarg while the org reusable workflow called `ecosystem audit-all`
+  directly and bypassed it, so `develop` could be green while unified CI
+  was red on identical code (measured in scitex-hub PR #433).
+
+  Honouring is never silent. `audit-all` always emits a **MASKED
+  INVENTORY** — the total, each rule id, its per-rule masked count and
+  its written rationale — in the normal output, not behind a verbosity
+  flag. The per-package summary now always states BOTH numbers (unmasked
+  errors AND masked count) for single- and multi-package runs alike; a
+  summary reporting only "0 errors" while 150 are masked is a lie of
+  omission. The exit code stays driven by unmasked findings.
+
+  **A skip entry with no written rationale is REJECTED** (exit 2, naming
+  the offending entry). Distinct from the legacy `audit.skip` (bare
+  codes, `audit-project`-only, silent) and from `audit.capabilities`.
+
+### Fixed
+- **The §6 MCP exemption is read from the AUDITED tree, not the runner's
+  registry checkout (#405).** `_audited_repo_root` preferred the ecosystem
+  registry's `local_path`, and a self-hosted CI runner's home often carries
+  a `~/proj/<pkg>` checkout at exactly that path. A PR declaring
+  `mcp_parity_exempt` / `mcp_tools_allowlist` in its own
+  `.scitex/dev/config.yaml` therefore could never turn its own
+  quality-audit green: the exemption was read from the STALE tree instead
+  of the editable-installed PR checkout under audit (scitex-orochi #460;
+  blocking scitex-hub #433, whose 52-entry `audit.mcp-tools-allowlist` was
+  correct and declared but had no effect in CI). Resolution order is now
+  import-derived root first (the tree actually being audited), with the
+  registry `local_path` deciding only for non-editable site-packages
+  installs where no audited tree exists on disk.
+- **Every auditor summary line now NAMES its category, pass or fail.**
+  The clean lines always did ("no project-structure violations"); the
+  FAILURE lines did not, so `audit-all` emitted three named SUCC lines
+  and then a bare `<pkg> (<path>): 3 error(s)` whose category was never
+  stated. sac PRs #813 and #814 both misread a real violation as a
+  broken gate because of it, wasting a CI cycle. Fixed across all five:
+  `CLI conventions`, `skills`, `Python API`, `project-structure`,
+  `Django-standard`.
+
+### Changed
+- `_cli/audit/_skills/_audit.py` (631 lines) split into `_registry.py` /
+  `_violation.py` / `_discovery.py` / `_checks.py` + a facade, mirroring
+  the sibling `_project/` package. Pure extraction; the facade
+  re-exports the original public surface.
+
 ## [0.34.0] - 2026-07-21
 
 ### Added

{scitex_dev-0.34.0 → scitex_dev-0.36.0}/CONTRIBUTING.md

@@ -16,7 +16,7 @@ See [CLA.md](CLA.md) for full details.
 
 ## Reporting Issues
 
-- Search [existing issues](https://github.com/ywatanabe1989/scitex-dev/issues)
+- Search [existing issues](https://github.com/scitex-ai/scitex-dev/issues)
   before opening a new one.
 - Include a minimal reproducible example when reporting bugs.
 - Specify your Python version, OS, and `scitex-dev` version.
@@ -24,7 +24,7 @@ See [CLA.md](CLA.md) for full details.
 ## Development Setup
 
 ```bash
-git clone git@github.com:ywatanabe1989/scitex-dev.git
+git clone git@github.com:scitex-ai/scitex-dev.git
 cd scitex-dev
 pip install -e ".[dev]"
 ```

{scitex_dev-0.34.0/src/scitex_dev.egg-info → scitex_dev-0.36.0}/PKG-INFO

@@ -1,11 +1,11 @@
 Metadata-Version: 2.4
 Name: scitex-dev
-Version: 0.34.0
+Version: 0.36.0
 Summary: Shared developer utilities for the SciTeX ecosystem
 Author-email: Yusuke Watanabe <ywatanabe@scitex.ai>
 License-Expression: AGPL-3.0-only
-Project-URL: Homepage, https://github.com/ywatanabe1989/scitex-dev
-Project-URL: Repository, https://github.com/ywatanabe1989/scitex-dev
+Project-URL: Homepage, https://github.com/scitex-ai/scitex-dev
+Project-URL: Repository, https://github.com/scitex-ai/scitex-dev
 Project-URL: Documentation, https://scitex-dev.readthedocs.io/
 Keywords: scitex,developer-tools,documentation,llm-friendly,mcp
 Requires-Python: >=3.9
@@ -75,14 +75,14 @@ Dynamic: license-file
 <p align="center">
   <a href="https://pypi.org/project/scitex-dev/"><img src="https://img.shields.io/pypi/v/scitex-dev?label=PyPI" alt="PyPI"></a>
   <a href="https://pypi.org/project/scitex-dev/"><img src="https://img.shields.io/pypi/pyversions/scitex-dev?label=Python" alt="Python"></a>
-  <a href="https://github.com/ywatanabe1989/scitex-dev/actions/workflows/rtd-sphinx-build-on-ubuntu-latest.yml"><img src="https://img.shields.io/github/actions/workflow/status/ywatanabe1989/scitex-dev/rtd-sphinx-build-on-ubuntu-latest.yml?branch=develop&label=RTD" alt="RTD"></a>
+  <a href="https://github.com/scitex-ai/scitex-dev/actions/workflows/rtd-sphinx-build-on-ubuntu-latest.yml"><img src="https://img.shields.io/github/actions/workflow/status/scitex-ai/scitex-dev/rtd-sphinx-build-on-ubuntu-latest.yml?branch=develop&label=RTD" alt="RTD"></a>
   <a href="https://www.gnu.org/licenses/agpl-3.0"><img src="https://img.shields.io/badge/License-AGPL--3.0-blue.svg?label=License" alt="License"></a>
 </p>
 <p align="center">
-  <a href="https://github.com/ywatanabe1989/scitex-dev/actions/workflows/pytest-matrix-on-ubuntu-py3-11-3-12-3-13.yml"><img src="https://img.shields.io/github/actions/workflow/status/ywatanabe1989/scitex-dev/pytest-matrix-on-ubuntu-py3-11-3-12-3-13.yml?branch=develop&label=Tests" alt="Tests"></a>
-  <a href="https://github.com/ywatanabe1989/scitex-dev/actions/workflows/import-smoke-on-ubuntu-py3-12.yml"><img src="https://img.shields.io/github/actions/workflow/status/ywatanabe1989/scitex-dev/import-smoke-on-ubuntu-py3-12.yml?branch=develop&label=Install-Check" alt="Install-Check"></a>
-  <a href="https://github.com/ywatanabe1989/scitex-dev/actions/workflows/scitex-dev-quality-audit-on-ubuntu-latest.yml"><img src="https://img.shields.io/github/actions/workflow/status/ywatanabe1989/scitex-dev/scitex-dev-quality-audit-on-ubuntu-latest.yml?branch=develop&label=Quality" alt="Quality"></a>
-  <a href="https://codecov.io/gh/ywatanabe1989/scitex-dev"><img src="https://img.shields.io/codecov/c/github/ywatanabe1989/scitex-dev/develop?label=CodeCov" alt="CodeCov"></a>
+  <a href="https://github.com/scitex-ai/scitex-dev/actions/workflows/pytest-matrix-on-ubuntu-py3-11-3-12-3-13.yml"><img src="https://img.shields.io/github/actions/workflow/status/scitex-ai/scitex-dev/pytest-matrix-on-ubuntu-py3-11-3-12-3-13.yml?branch=develop&label=Tests" alt="Tests"></a>
+  <a href="https://github.com/scitex-ai/scitex-dev/actions/workflows/import-smoke-on-ubuntu-py3-12.yml"><img src="https://img.shields.io/github/actions/workflow/status/scitex-ai/scitex-dev/import-smoke-on-ubuntu-py3-12.yml?branch=develop&label=Install-Check" alt="Install-Check"></a>
+  <a href="https://github.com/scitex-ai/scitex-dev/actions/workflows/scitex-dev-quality-audit-on-ubuntu-latest.yml"><img src="https://img.shields.io/github/actions/workflow/status/scitex-ai/scitex-dev/scitex-dev-quality-audit-on-ubuntu-latest.yml?branch=develop&label=Quality" alt="Quality"></a>
+  <a href="https://codecov.io/gh/scitex-ai/scitex-dev"><img src="https://img.shields.io/codecov/c/github/scitex-ai/scitex-dev/develop?label=CodeCov" alt="CodeCov"></a>
 </p>
 <!-- scitex-badges:end -->
 
@@ -324,7 +324,7 @@ scitex-dev skills export --package scitex-dev  # Export to Claude Code
 # Private skills (~/.scitex/*/skills/*-private/) are symlinked automatically
 ```
 
-> **[Full skills directory](https://github.com/ywatanabe1989/scitex-dev/tree/develop/src/scitex_dev/_skills)**
+> **[Full skills directory](https://github.com/scitex-ai/scitex-dev/tree/develop/src/scitex_dev/_skills)**
 
 | Skill | Content |
 |-------|---------|

{scitex_dev-0.34.0 → scitex_dev-0.36.0}/README.md

@@ -16,14 +16,14 @@
 <p align="center">
   <a href="https://pypi.org/project/scitex-dev/"><img src="https://img.shields.io/pypi/v/scitex-dev?label=PyPI" alt="PyPI"></a>
   <a href="https://pypi.org/project/scitex-dev/"><img src="https://img.shields.io/pypi/pyversions/scitex-dev?label=Python" alt="Python"></a>
-  <a href="https://github.com/ywatanabe1989/scitex-dev/actions/workflows/rtd-sphinx-build-on-ubuntu-latest.yml"><img src="https://img.shields.io/github/actions/workflow/status/ywatanabe1989/scitex-dev/rtd-sphinx-build-on-ubuntu-latest.yml?branch=develop&label=RTD" alt="RTD"></a>
+  <a href="https://github.com/scitex-ai/scitex-dev/actions/workflows/rtd-sphinx-build-on-ubuntu-latest.yml"><img src="https://img.shields.io/github/actions/workflow/status/scitex-ai/scitex-dev/rtd-sphinx-build-on-ubuntu-latest.yml?branch=develop&label=RTD" alt="RTD"></a>
   <a href="https://www.gnu.org/licenses/agpl-3.0"><img src="https://img.shields.io/badge/License-AGPL--3.0-blue.svg?label=License" alt="License"></a>
 </p>
 <p align="center">
-  <a href="https://github.com/ywatanabe1989/scitex-dev/actions/workflows/pytest-matrix-on-ubuntu-py3-11-3-12-3-13.yml"><img src="https://img.shields.io/github/actions/workflow/status/ywatanabe1989/scitex-dev/pytest-matrix-on-ubuntu-py3-11-3-12-3-13.yml?branch=develop&label=Tests" alt="Tests"></a>
-  <a href="https://github.com/ywatanabe1989/scitex-dev/actions/workflows/import-smoke-on-ubuntu-py3-12.yml"><img src="https://img.shields.io/github/actions/workflow/status/ywatanabe1989/scitex-dev/import-smoke-on-ubuntu-py3-12.yml?branch=develop&label=Install-Check" alt="Install-Check"></a>
-  <a href="https://github.com/ywatanabe1989/scitex-dev/actions/workflows/scitex-dev-quality-audit-on-ubuntu-latest.yml"><img src="https://img.shields.io/github/actions/workflow/status/ywatanabe1989/scitex-dev/scitex-dev-quality-audit-on-ubuntu-latest.yml?branch=develop&label=Quality" alt="Quality"></a>
-  <a href="https://codecov.io/gh/ywatanabe1989/scitex-dev"><img src="https://img.shields.io/codecov/c/github/ywatanabe1989/scitex-dev/develop?label=CodeCov" alt="CodeCov"></a>
+  <a href="https://github.com/scitex-ai/scitex-dev/actions/workflows/pytest-matrix-on-ubuntu-py3-11-3-12-3-13.yml"><img src="https://img.shields.io/github/actions/workflow/status/scitex-ai/scitex-dev/pytest-matrix-on-ubuntu-py3-11-3-12-3-13.yml?branch=develop&label=Tests" alt="Tests"></a>
+  <a href="https://github.com/scitex-ai/scitex-dev/actions/workflows/import-smoke-on-ubuntu-py3-12.yml"><img src="https://img.shields.io/github/actions/workflow/status/scitex-ai/scitex-dev/import-smoke-on-ubuntu-py3-12.yml?branch=develop&label=Install-Check" alt="Install-Check"></a>
+  <a href="https://github.com/scitex-ai/scitex-dev/actions/workflows/scitex-dev-quality-audit-on-ubuntu-latest.yml"><img src="https://img.shields.io/github/actions/workflow/status/scitex-ai/scitex-dev/scitex-dev-quality-audit-on-ubuntu-latest.yml?branch=develop&label=Quality" alt="Quality"></a>
+  <a href="https://codecov.io/gh/scitex-ai/scitex-dev"><img src="https://img.shields.io/codecov/c/github/scitex-ai/scitex-dev/develop?label=CodeCov" alt="CodeCov"></a>
 </p>
 <!-- scitex-badges:end -->
 
@@ -265,7 +265,7 @@ scitex-dev skills export --package scitex-dev  # Export to Claude Code
 # Private skills (~/.scitex/*/skills/*-private/) are symlinked automatically
 ```
 
-> **[Full skills directory](https://github.com/ywatanabe1989/scitex-dev/tree/develop/src/scitex_dev/_skills)**
+> **[Full skills directory](https://github.com/scitex-ai/scitex-dev/tree/develop/src/scitex_dev/_skills)**
 
 | Skill | Content |
 |-------|---------|

scitex_dev-0.36.0/docs/adr/0004-execution-fabric-unified-jobspec.md

@@ -0,0 +1,295 @@
+# ADR-0004: Execution fabric — one declarative JobSpec, intent/mechanism orthogonality, best-effort allocation onto the shared pool, phased adoption
+
+## Status
+
+Accepted (2026-07-22). Operator-approved direction (2026-07-22): "declare jobs
+uniformly; separate INTENT (daemon/periodic/one-shot) from MECHANISM
+(local/Slurm/runner); allocate best-effort onto idle resources; phased
+adoption: tests first, then CI, then agent jobs." This ADR records the
+unifying architecture; it does not re-litigate the direction.
+
+## Context
+
+The SciTeX ecosystem runs (at least) four families of compute jobs today, each
+with its own declaration surface, its own executor wiring, and no shared
+resource model:
+
+1. **Test suites** — the test-execution recipe shipped in #385
+   (`src/scitex_dev/_core/test_execution.py`): a per-package
+   `TestExecutionConfig` (`local` | `remote-required`) with a free-form
+   `submit_template`, a `test_execution` knob, and an auto-loaded `pytest11`
+   guard (`_test_execution_plugin.py`) that blocks local pytest when a package
+   mandates remote.
+2. **CI** — GitHub Actions across ~70 repos, mid-consolidation onto the ONE
+   canonical thin org-reusable caller
+   (`src/scitex_dev/_ecosystem/ci_template/templates/ci.yml.tmpl`, #400): job
+   bodies live once in `scitex-ai/.github@main`; runner selection is the
+   Actions Variable `vars.CI_RUNS_ON`. The 2026-07-21 CI-drift inventory
+   found **five** self-hosted runner-label vocabularies plus 651
+   `ubuntu-latest` violations spread over six workflow generations — the cost
+   of never having had one declaration model.
+3. **Experiments / research pipelines** — ad hoc `srun`/`sbatch`/ssh onto
+   Spartan, no uniform receipt, no uniform resource declaration.
+4. **Agent/fleet jobs** — periodic and daemon work (sac restart passes,
+   drift timers, sweeps), partially federated through the
+   `scitex_dev.jobs.JobSpec` entry-point contract
+   (`scitex_dev/jobs/__init__.py`, group `scitex_dev.jobs`), partially still
+   hand-rolled crons on hosts.
+
+Two prior design artifacts cover slices of this space and must be reconciled,
+not duplicated:
+
+- **The phase-1 execution-fabric design** (`docs/design/execution-fabric-phase1.md`,
+  merged as PR #388) locks four interfaces — `ContainerContract`,
+  `ResourceSpec`, `JobSpec`/`JobHandle` with the
+  `submit/status/logs/cancel/collect` `ExecutionBackend` Protocol, and the
+  anti-false-claim `JobReceipt` — and scopes phase-1 to exactly two backends
+  (`local`, `slurm`). Its strategic rule stands: **do not reinvent the
+  executor** (Snakemake/Parsl/Slurm/Batch are prior art); SciTeX adds the
+  policy + can't-misuse + agent-DX layer.
+- **The JobSpec federation contract** (fleet card + GITIGNORED contract doc,
+  2026-07-20) establishes that a leaf never runs its own cron/systemd: it
+  *declares* `JobSpec`s via the `scitex_dev.jobs` entry-point group and
+  scitex-dev discovers, merges, lowers, and installs them
+  (`ecosystem {systemd,cron,up,run}`). Its open thread — a `kind` taxonomy
+  note separating INTENT (daemon/periodic) from MECHANISM
+  (service/timer/cron) — is exactly the axis split this ADR decides.
+
+The problem: these two `JobSpec`s (the fabric's batch-job spec and the
+federation's scheduled-job spec) grew independently, CI declares nothing in
+either, and "where may this run, with what resources" is answered by four
+different mechanisms. Meanwhile the operator directed the runner fleet toward
+**one shared pool** (org runners + Spartan, best-effort onto idle capacity),
+which no per-family mechanism can express.
+
+## Decision
+
+Adopt **one ecosystem-wide execution-fabric model** with four load-bearing
+decisions.
+
+### 1. One declarative JobSpec; INTENT and MECHANISM are orthogonal axes
+
+Every job — test run, CI job, experiment, agent/fleet job — is declared by a
+single conceptual JobSpec with two independent axes:
+
+- **INTENT** (what lifecycle the work wants):
+  `one-shot` | `periodic` | `daemon` | `reactive`.
+  - `one-shot` — run once to completion (a test suite, an experiment trial).
+  - `periodic` — run on a schedule (drift timers, sweeps, nightly CI).
+  - `daemon` — long-running, supervised, restart-on-death (sac agents,
+    listeners).
+  - `reactive` — run when an external event fires (PR push → CI, card event
+    → handler).
+- **MECHANISM** (what engine executes it):
+  `local-exec` | `slurm` | `gh-runner` | `systemd/cron`.
+
+**No declaration may name a mechanism where an intent belongs, and vice
+versa.** A leaf says "this is periodic, every 2 min, needs 1 cpu"; the fabric
+(scitex-dev) chooses systemd-timer vs cron vs a Slurm scrontab. A test recipe
+says "remote-required, 8 cpu"; the fabric chooses the slurm backend and the
+partition. This generalizes what the federation contract already enforces
+("`kind` is INTENT … Do NOT pick the mechanism") and resolves its open
+taxonomy thread: the current `service`/`timer`/`cron` kinds are re-read as
+`daemon` / `periodic` / `periodic` intents, with `timer`-vs-`cron` demoted to
+a mechanism choice the lowering step makes (and the #369 rule stands: lowering
+that would silently drop a guarantee like `timeout_sec`/`venv` must refuse
+loudly instead).
+
+**Resources are declared, never implied.** The backend-neutral `ResourceSpec`
+(`cpu` / `memory_gb` / `gpu` / `walltime_min`) from the phase-1 design is the
+one resource vocabulary. Declarations never contain `partition:`,
+`instance_type:`, or runner hostnames; each mechanism adapter translates
+(`cpu → --cpus-per-task`, `cpu → pytest -n <allocated>`, labels → runner
+selection). The allocated-cpus rule from phase-1 §3c is part of this: a job
+sees its *allocated* budget, never the literal node core count.
+
+The phase-1 batch `JobSpec` (container + resources + command) and the
+federation `JobSpec` (name + schedule + command + guarantees) are hereby two
+**profiles of the same model** — a one-shot profile and a periodic/daemon
+profile — and converge field-by-field as phase 3 lands (shared `ResourceSpec`,
+shared receipt/log conventions, one discovery surface). Neither is rewritten
+today; new fields land in the unified vocabulary only.
+
+Reproducibility-bearing jobs (tests, experiments) additionally carry the
+phase-1 `ContainerContract` (commit + image + recipe, no dirty-tree state) and
+must yield a `JobReceipt` from `collect()` — the anti-false-claim artifact
+(real commit, backend, allocated cpu, pass/fail). Ephemeral fleet plumbing
+(a restart pass) is exempt from the container contract but not from logging
+(`~/.scitex/<pkg>/runtime/logs/<name>.log`, scitex-dev-owned).
+
+### 2. Allocation: best-effort onto the shared pool
+
+There is **one shared compute pool**: the org self-hosted runners plus
+Spartan idle/backfill capacity. Allocation is **best-effort** — jobs state
+resources and constraints; the fabric places them onto whatever pool capacity
+is free, with no per-repo or per-team runner ownership.
+
+- The **CI-facing slice** of this pool is the runner-label unification
+  already in flight: `runs-on: ${{ fromJSON(vars.CI_RUNS_ON || '["self-hosted","Linux","X64","scitex-ci"]') }}`
+  in the org-side reusable bodies, with `vars.CI_RUNS_ON` (repo → org
+  precedence) as the single selection knob. The five current label
+  vocabularies collapse to ONE (the CI-drift inventory's target); GitHub-
+  hosted runners remain forbidden (PS-169, ERROR).
+- The **batch-facing slice** is the phase-1 `slurm` backend: `srun --overlap`
+  into an existing lease where one is held (no queue wait), `sbatch`
+  otherwise, compute nodes only — never the login node. Spartan is ONE
+  resource; internal placement is delegated to Spartan's native Slurm via
+  the `submit_template`. The fabric never nests a second scheduler
+  controller inside Slurm.
+- Best-effort means **no reservations and no fairness engine in SciTeX**:
+  contention is resolved by the underlying engines (Slurm priorities, the
+  Actions queue). The fabric's job is to route to *idle* capacity and to
+  refuse *forbidden* placements (laptop-heavy-test, login-node pytest,
+  GPU-test-on-CPU), not to arbitrate between permitted ones.
+- The routing selector (`--target fastest|cheapest|local-only|secure`) stays
+  deferred exactly as phase-1 left it; until it lands, backend choice is
+  explicit per declaration.
+
+### 3. Ownership: ports-and-adapters boundaries (per ADR-0003)
+
+The seams are extension points, not hard imports — the ecosystem's default
+seam rule (ADR-0003) applied to compute:
+
+| Owner | Owns | Does NOT own |
+| --- | --- | --- |
+| **scitex-dev** | The fabric **primitives** (JobSpec model, `ResourceSpec`, receipt schema, intent taxonomy), the **policy/guard layer** (recipes, knobs, pytest11 guard, auditor rules), and **aggregation**: the single job aggregator for all repetitive jobs — plugin federation via the `scitex_dev.jobs` entry-point group; leaves declare, scitex-dev discovers/merges/lowers/installs; **no hardcoded or leaf-local crons**. | Slurm mechanics, container lifecycle. |
+| **scitex-hpc** | Spartan **capacity mechanics**: the `ExecutionBackend` implementations (`local`, `slurm`), `ResourceSpec` → backend-vocabulary translation (partition/account/`--cpus-per-task`), the resource registry + liveness probes, runner provisioning on Spartan. | Policy (what is allowed to run where), job declaration schemas. |
+| **sac** | **Container/daemon lifecycle**: agent containers, supervised daemon processes, restart/heal passes. sac keeps its detection + restart logic; it registers only the *scheduling* of its passes as federation JobSpecs (intent `periodic`), per the federation contract. | Scheduling installation (cron/systemd rendering — scitex-dev's), Spartan submission. |
+
+Leaf packages interact with the fabric only through declaration surfaces
+(recipe YAML, knob, entry-point provider) — kind-c edges in the ADR-0003
+taxonomy — never by importing backend internals. The interface-home question
+from phase-1 §6 (shared spec module vs a thin `scitex-execspec` package)
+remains open with the same recommendation: a thin shared spec so neither side
+co-locks the other's release (standalone-independence).
+
+### 4. Failure doctrine: fail-safe on the invocation path, fail-loud in gates
+
+The house pattern, already shipped twice, is now fabric-wide law:
+
+- **Invocation paths fail safe.** Anything auto-loaded into every
+  environment must never brick the host process. Precedents: the #385 recipe
+  loader (malformed YAML / unreadable file / discovery error → warn +
+  inert `mode=local`, pytest never crashes) and the currency gate's
+  FRESHNESS half (#396: never a blocking network call; offline → PASS).
+  Fabric corollaries: a missing/unparseable JobSpec declaration disables
+  that job with a loud warning, it does not take down the aggregator run;
+  an unreachable backend fails *that submission* with an actionable error,
+  not the fabric.
+- **Gates fail loud.** Anything whose job is to *verify* must never report
+  green for what it could not check. Precedents: the currency gate's
+  INTEGRITY half raising `StalenessError` with the exact remedy, the
+  deterministic audit target-tree resolution (#397 — an audit that might
+  grade the wrong checkout now names which tree it resolved and how), and
+  the PS-221 `[all]`-closure ERROR rule. Fabric corollaries: the pytest11
+  remote-required guard aborts with `UsageError`; a `collect()` that cannot
+  fill the mandatory receipt fields (commit, backend, allocated cpu,
+  pass/fail) is non-conformant and errors; the timer→cron lowering refuses
+  to drop guarantees (#369); the CI auditor ERRORs on `ubuntu-latest` and
+  non-canonical workflow files.
+
+The dividing line: **is this code path *running* work or *vouching* for
+work?** Running → degrade gracefully, warn loudly. Vouching → refuse loudly;
+skipped must never count as clean.
+
+### Phased adoption — with exit criteria
+
+| Phase | Scope | Status | Exit criteria |
+| --- | --- | --- | --- |
+| **P1** | **Tests** | Shipped (#385, v0.33.0; design locked by #388) | (a) recipe + `test_execution` knob + auto-loaded `pytest11` guard live (done); (b) literal `-n auto` replaced by allocated-cpus per backend (`$SLURM_CPUS_PER_TASK` / `ResourceSpec.cpu`); (c) at least one `remote-required` package runs end-to-end through the scitex-hpc `slurm` backend and every run emits a conformant `JobReceipt`. |
+| **P2** | **CI** | Wave 1 in flight (#400 canonical caller; drift inventory 2026-07-21) | (a) every scitex repo carries the thin org-reusable `ci.yml` caller and nothing else (six generations → one); (b) runner labels unified to ONE vocabulary selected via `vars.CI_RUNS_ON`; (c) zero `ubuntu-latest` / GitHub-hosted jobs in scitex repos (PS-169 at ERROR); (d) auditor rule ERRORs on non-canonical workflow files, so drift cannot silently return. |
+| **P3** | **Agent/fleet jobs** | Design open (federation contract 2026-07-20; this ADR fixes the taxonomy) | (a) `scitex_dev.jobs.JobSpec` carries the intent/mechanism split (intents accepted; mechanism chosen at lowering; back-compat mapping for `service`/`timer`/`cron`); (b) zero leaf-local or hand-rolled host crons — every repetitive job discoverable via `scitex-dev ecosystem` from an entry-point provider; (c) sac's daemon/restart passes declared through the federation (sac keeps the logic, scitex-dev installs the schedule); (d) periodic fleet jobs eligible for pool placement (Spartan backfill) where they don't need host-locality. |
+
+A phase is *entered* when its predecessor's exit criteria are met or
+explicitly waived by the operator; phases may overlap in build but not in
+"declare victory" order. Experiments/research pipelines ride P1's interfaces
+(same backends, same receipts) and need no separate phase.
+
+### Non-goals
+
+- **Not a new scheduler kernel.** Slurm, systemd, cron, and the GitHub
+  Actions queue remain the engines; the fabric declares, routes, guards, and
+  audits. No DAG scheduler, retry engine, distributed executor, or
+  storage-staging layer is built here (delegate to Snakemake/Nextflow if
+  ever needed — phase-1 §5/§7 stand).
+- **No paid cloud.** No AWS/GCP/Azure backends, no cost-bearing capacity.
+  The interfaces keep the door open (`cost_usd` stays `None`; a cloud
+  backend would be "+1 adapter"), but no cloud code, budget, or data-
+  governance machinery ships under this ADR.
+- **No fairness/quota arbitration** between permitted jobs (see Decision 2).
+
+## Consequences
+
+**Positive**
+
+- One vocabulary for "what runs where with what": four job families stop
+  answering the question four ways. New job types (a benchmark sweep, a docs
+  build) get declaration + placement + receipts for free instead of a new
+  bespoke runner script.
+- The intent/mechanism split makes mechanism migration invisible to leaves:
+  moving a periodic job from host cron to a systemd timer to Slurm scrontab
+  is a lowering change in scitex-dev, zero leaf PRs — the same property the
+  thin CI caller already gives workflows (job bodies change org-side, repos
+  don't churn).
+- The shared pool ends runner-label balkanization: capacity added to the
+  pool (a new Spartan lease, a new runner box) becomes available to CI,
+  tests, and fleet jobs simultaneously via one label vocabulary and one
+  `ResourceSpec` translation point.
+- The receipt + fail-loud-gate doctrine extends the anti-false-claim
+  property from tests to every job family: "it ran" is always backed by an
+  artifact naming commit/backend/resources, which is what lets agents be
+  trusted to self-report.
+- Ownership boundaries are enforceable with the existing ADR-0003 lint
+  machinery (entry-point federation is kind-c; a leaf importing backend
+  internals is the a2 smell).
+
+**Negative / cost**
+
+- Convergence debt is real: two JobSpec shapes exist today and must be
+  field-reconciled during P3 without breaking registered providers
+  (back-compat mapping for `service`/`timer`/`cron` is mandatory).
+- Best-effort allocation means no latency guarantees: a saturated pool
+  queues CI and fleet jobs alike. Accepted — the alternative (reserved
+  capacity per family) recreates the balkanization this ADR removes.
+- scitex-dev becomes a scheduling single-point-of-policy: an aggregator bug
+  can mis-install fleet-wide schedules. Mitigated by the fail-safe
+  invocation doctrine (one bad declaration never kills the run) and by the
+  engines staying authoritative (a wrong lowering is visible in
+  systemd/cron state, not hidden in a bespoke daemon).
+- P2 completion requires a ~70-repo sweep (the drift inventory's wave plan);
+  until (d) lands, drift can re-enter behind the sweep.
+
+**Avoided cost (vs. status quo)**
+
+- Without this ADR, the two JobSpecs keep diverging, CI label vocabularies
+  keep multiplying (five and counting), and every new job family invents a
+  fifth declaration mechanism — each later unification strictly more
+  expensive than this one.
+
+## Notes
+
+- Operator approvals: direction approved 2026-07-22 (intent/mechanism split,
+  best-effort pool, tests→CI→agent-jobs phasing); shared-pool + label-
+  unification directive TG 1580/1583/1592 (2026-07-21); executor-reuse rule
+  TG1524 (via phase-1 doc).
+- Source documents synthesized (not duplicated) here:
+  - `docs/design/execution-fabric-phase1.md` (#388) — the four locked
+    interfaces, the two phase-1 backends, the scitex-dev/scitex-hpc split,
+    the deferred list. Remains the authoritative interface spec; this ADR
+    is the ecosystem-wide architecture above it.
+  - `src/scitex_dev/_core/test_execution.py`,
+    `src/scitex_dev/_core/_test_execution_plugin.py`,
+    `src/scitex_dev/_core/_knobs.py` (#385) — the shipped P1 policy layer.
+  - `src/scitex_dev/_ecosystem/ci_template/templates/ci.yml.tmpl` (#400) —
+    the canonical thin caller; CI-drift inventory 2026-07-21 (GITIGNORED)
+    — the P2 ground truth and wave plan.
+  - `scitex_dev/jobs/__init__.py` + the JobSpec federation contract
+    (2026-07-20) — the P3 substrate; its open `kind`-taxonomy thread is
+    closed by Decision 1.
+- Open threads this ADR leaves open (tracked on the fleet board, not here):
+  the interface-home question (shared module vs `scitex-execspec`), the
+  routing selector, sharding, cost/budget fields, data governance — all
+  deferred exactly as phase-1 §5 lists them.
+- Related ADRs: ADR-0003 (ports-and-producers — supplies the seam rule
+  Decision 3 applies); ADR-0001/0002 (package-absorption precedent for the
+  "reverse via a new ADR" convention this file follows).

{scitex_dev-0.34.0 → scitex_dev-0.36.0}/docs/to_claude/skills/SKILL.md

@@ -39,7 +39,7 @@ scitex-ui                 github.com/ywatanabe1989/scitex-ui
 scitex-app                github.com/ywatanabe1989/scitex-app
 scitex-audio              github.com/ywatanabe1989/scitex-audio
 scitex-scholar            github.com/ywatanabe1989/scitex-scholar
-scitex-dev                github.com/ywatanabe1989/scitex-dev
+scitex-dev                github.com/scitex-ai/scitex-dev
 singularity-template      github.com/ywatanabe1989/singularity-template
 ```
 

{scitex_dev-0.34.0 → scitex_dev-0.36.0}/pyproject.toml

@@ -174,8 +174,8 @@ scitex-dev = "scitex_dev._system_deps:provide"
 #   scitex-dataset = "scitex_dataset._gate_plugin:provide"
 
 [project.urls]
-Homepage = "https://github.com/ywatanabe1989/scitex-dev"
-Repository = "https://github.com/ywatanabe1989/scitex-dev"
+Homepage = "https://github.com/scitex-ai/scitex-dev"
+Repository = "https://github.com/scitex-ai/scitex-dev"
 Documentation = "https://scitex-dev.readthedocs.io/"
 
 [tool.setuptools_scm]