Affected Repositories

• PyAutoHands (primary), PyAutoNerves, PyAutoBrain, PyAutoHeart, PyAutoMind, autocti_workspace

Branch Survey

Suggested branch: feature/cti-release-train-wiring

Implementation Steps

1. Nerves autonerves/setup_colab.py: add _AUTOCTI_STACK = _AUTOFIT_STACK + ["autoarray", "autocti"] and a _PROJECTS["autocti"] entry with pre_install="arcticpy". _colab_setup runs that hook before the pip install: it fetches raw.githubusercontent.com/PyAutoLabs/PyAutoHeart/main/.github/actions/install-arcticpy/install_arcticpy.sh and runs it with ARCTICPY_SUDO=false PYTHON=sys.executable. Add for_autocti() and tests.
2. Hands:
   • autohands/tag_and_merge.sh: add PyAutoCTI to LIB_PROJECTS.
   • pre_build.sh: add the rows autocti_workspace autocti true false and autocti_workspace_test autocti false false.
   • autohands/build_util.py: add autocti to COLAB_PROJECTS.
   • autohands/config/workspaces.yaml: add autocti / autocti_test to run_all, PyAutoCTI to libraries, and autocti to colab_setup and colab_gemini (assistant: autocti_assistant).
   • .github/workflows/release.yml: add PyAutoCTI to release_test_pypi and release (with if-gated install-arcticpy and autocti install steps), autocti_workspace to the run_smoke_tests matrix (arcticpy-gated), clear_pending_release_labels repositories, publish_release_notes, the release_workspaces rows, and a wiki_currency_check_autocti / wiki_drift_issue_autocti pair. The pyautocti-release visualization dispatch is included if autocti_visualization has a receiver.
   • python_matrix.yml: CTI unit and smoke legs on 3.12 and 3.13.
   • generate_release_notes.py, slack_release_notes.py, aggregate_results.py, generate_markdown.py, navigator.py, bin/autohands: add the repo maps.
3. Brain: add PyAutoCTI to LIBRARIES in agents/conductors/release/{rehearse,validate}.sh, the nightly library list, and bin/clean_slate.sh PACKAGING_REPOS.
4. Heart: set config/repos.yaml PyAutoCTI release_gate: true, and add CTI rows to the Release Integrate / workspace-validation.yml / docs-build.yml maps.
5. Mind: add FIREWALL_ALLOWLIST tokens.
6. autocti_workspace: control-test generate.py on an unchanged script, then run generate.py autocti and check that no notebook has a Finish. cell or a literal # %% or '''. Add config/build/profile_release.yaml and update the Notebook regeneration line in AGENTS.md.
7. Validation: generate.py <unknown> still refuses with notebooks/ intact; generate.py howtolens gives a zero diff; Hands, Nerves, Brain and Heart test suites pass; then the TestPyPI rehearsal and Heart ingest.

Key Files

• PyAutoHands/.github/workflows/release.yml, python_matrix.yml, pre_build.sh, autohands/{tag_and_merge.sh,build_util.py,config/workspaces.yaml}
• PyAutoNerves/autonerves/setup_colab.py
• PyAutoHeart/config/repos.yaml
• PyAutoBrain/agents/conductors/release/{rehearse,validate}.sh

Risks

• The release job uploads to PyPI with fail-fast on; the human confirmed on 2026-11-04 that the PYPI token covers autocti.
• After release_gate: true, Heart's readiness depends on CTI's CI.

CTI release-train wiring — first modern autocti release

Type: release
Target: PyAutoHands
Repos:

• @PyAutoHands
• @PyAutoNerves
• @autocti_workspace
  Themes:
• release
• cti
  Difficulty: medium
  Autonomy: human-required
  Priority: normal
  Status: formalised
  Consequence: judge
  Review-minutes: 20
  Unattended: never
  Epic: cti-upkeep
  Phase: 1
  Filed: 2026-07-17 (backfilled from git)

Follow-up to the CTI resurrection epic (Phases 0-5; workspace_test + Heart
polling shipped 2026-07-17). Wire autocti into the release train and produce
its first modern PyPI release. Deliberately split from the epic's Phase 5:
this touches the nightly's most dangerous machinery and must be done with a
TestPyPI rehearsal, in a fresh session.

Scope

1. PyAutoHands: tag_and_merge.sh LIB_PROJECTS += PyAutoCTI;
   pre_build.sh run_workspace lines for autocti_workspace (+ _test,
   generate=false); release.yml / python_matrix.yml matrices;
   generate_release_notes.py + slack_release_notes.py +
   aggregate_results.py repo maps; build_util.py COLAB_PROJECTS +=
   autocti; FIREWALL_ALLOWLIST tokens in Mind for every file gaining CTI names.
2. PyAutoNerves autonerves/setup_colab.py: an autocti _PROJECTS entry
   whose package list handles arcticpy correctly on Colab (apt
   libgsl-dev + pip install numpy cython + arcticpy==2.6 --no-build-isolation --no-deps — a naive pip install downgrades numpy
   below 2.0). May need a per-project pre_install hook in the setup
   machinery.
3. Notebook generation: generate.py autocti (blocked today by the
   COLAB_PROJECTS registry check) → commit regenerated notebooks to
   autocti_workspace. This is a fork, not a formality: either register
   autocti (items 1-2 above, if its notebooks are meant to carry Colab setup)
   or give generate.py a documented no-Colab mode for workspaces that do
   not. Whichever path is taken, update autocti_workspace/AGENTS.md's
   Notebook regeneration line to match it — that line currently implies the
   standard path works, which it does not.
4. Rehearsal then release: TestPyPI rehearsal of the extended train
   (release rehearse / release validate through the Release Agent), fix
   fallout, then the first modern autocti release rides the next nightly or
   a human-authorized release. Never hand-dispatch the nightly.

Context

• Item 3's blocker is now a clean refusal, not a destructive one (fix: generate.py deletes notebooks/ before validating the project name #215
  / PR fix: validate the project before generate.py deletes notebooks/ #216, 2026-07-30). generate.py used to rmtree the whole notebooks/
  tree before reaching the COLAB_PROJECTS check, so probing generate.py autocti deleted 113 tracked notebooks and then aborted. It now validates up
  front and exits with "Nothing was modified", leaving the tree untouched — so
  this task can probe the registry check freely. The registration itself
  (items 1-2) is untouched and still belongs here. Pinned by
  PyAutoHands/tests/test_generate_validates_project.py; the ordering is
  invisible in a passing run, so do not remove that test while reworking
  generate.py for item 3.
• Why this stayed invisible. autocti_workspace is absent from
  PyAutoHands/pre_build.sh's run_workspace matrix entirely (the
  generate=true set is autofit_workspace, autogalaxy_workspace,
  autolens_workspace, HowToGalaxy, HowToLens, HowToFit), so no release path has
  ever exercised generate.py autocti. Corroborating: 0 of the repo's 79
  notebooks carry a Colab setup cell, dating them to before
  inject_colab_setup became strict. They do still track scripts/ 1:1 (79
  scripts, 79 notebooks, no orphans either way), so they were maintained by
  some path that no longer works. Item 1's pre_build.sh line is what closes
  this hole for good.
• Backlog item 3 will clear when it runs: autocti_workspace/notebooks/
  currently carries 34 Finish. markdown cells (from the crutch sweep,
  docs: remove the Finished./Finish. notebook-generation crutch autocti_workspace#16) and 4 mangled code cells containing a literal
  # %% and ''' — a SyntaxError if run — in
  notebooks/imaging_ci/modeling/features/{cosmic_rays,non_uniform,serial_cti,visualize_full}.ipynb,
  plus the 5 script-reference fixes from autocti_workspace#15. All are already
  correct in scripts/; one successful generate.py autocti clears all of them.
  So item 3 is the single gate on three separate merged sweeps reaching the
  notebooks.
• Do not hand-roll the regeneration. build_util.py_to_notebook alone is not
  equivalent to generate.py: control-tested against an unchanged script
  (scripts/dataset_1d/extract.py vs the committed
  notebooks/dataset_1d/extract.ipynb), the committed notebook carries a
  trailing empty code cell that py_to_notebook does not emit — 343 vs 336
  lines, otherwise identical. That cell is the signature of the pre-2026-07-24
  generator (before PyAutoHands 6916814 a closing docstring always emitted a
  # %%), so these 79 notebooks predate it — expect the regeneration to drop
  those empty cells across the board, which is correct. Control-test against an
  unchanged script before trusting any regeneration path here; that is what
  caught this.
• Epic records: PyAutoMind/complete/2026/07/cti-resurrection-phase{0..5}.md.
• PyAutoCTI pyproject floors are release-ready (setuptools-scm, Phase 0);
  CI green via Heart lib-tests (Phase 3); Heart polls the CTI repos (Phase 5).
• arcticpy traps + CI install recipe: PyAutoCTI/AGENTS.md and
  autocti_workspace_test/.github/scripts/smoke_install.sh.

Validation

• generate.py autocti exits 0 from autocti_workspace/; all 79 notebooks
  regenerate; no code cell contains a literal # %% or '''; no markdown cell
  is Finish..
• generate.py <unknown-project> still exits non-zero with notebooks/
  intact and no stray .ipynb beside any script
  (PyAutoHands/tests/test_generate_validates_project.py stays green).
• generate.py howtolens from HowToLens/ still gives a zero diff — the
  registry change must not perturb the projects already on the train.
• autocti_workspace/AGENTS.md's Notebook regeneration line matches whatever
  path item 3 actually chose.
• pre_build.sh's run_workspace matrix names autocti_workspace, so the next
  nightly exercises the path rather than skipping it.

Provenance

Folded 2026-08-26: draft/bug/pyautohands/generate_rejects_autocti_after_deleting_notebooks.md
was retired into this file. That prompt tracked two defects — the destructive
ordering in generate.py (fixed by #215 / PR #216,
regression-tested) and the missing autocti registration. Only the second was
ever this file's items 1-3, so the bug prompt had no startable work of its own;
its residual requirements (the AGENTS.md doc line, the pre_build-matrix
evidence, the control-test detail) are absorbed above. It had itself
consolidated three independently-filed prompts describing the same defect.