fix: classify the shared AI policy in validators

[Jammy2211]

Overview

The shared root AI_POLICY.md is currently unclassified by the assistant clone boundary, so every autolens assistant PR fails the boundary check and future clone births are blocked. Classify the policy as generic clone infrastructure and align the maintainer-facing boundary documentation across the three assistant cells. The analogous PyAutoMemory root allowlist regression remains tracked and fixed under PyAutoMemory #30.

Plan

• Add AI_POLICY.md to the Clone Agent's shared generic file set.
• Document that generic classification in each assistant cell's maintainer boundary guide.
• Keep the current supported reference profiles; do not invent an AutoCTI clone profile.
• Run focused clone-conductor tests and boundary checks only.

Detailed implementation plan

Affected Repositories

• PyAutoBrain (primary)
• autolens_assistant
• autofit_assistant
• autocti_assistant

Branch Survey

Repository	Current Branch	Dirty?
./PyAutoBrain	main	skills/WORKFLOW.md modified (unrelated; isolated worktree required)
./autolens_assistant	main	clean, two commits behind origin/main
./autofit_assistant	main	clean
./autocti_assistant	main	clean

Suggested branch: feature/ai-policy-validator-followups

Implementation Steps

1. Add AI_POLICY.md to _SHARED_GENERIC in PyAutoBrain/agents/conductors/clone/_clone.py.
2. Add a focused regression assertion in PyAutoBrain/tests/test_clone_conductor.py if existing coverage does not already pin the shared generic classification.
3. Update modes/maintainer.md in autolens_assistant, autofit_assistant, and autocti_assistant to identify the root policy as generic template infrastructure.
4. Run the focused Clone Agent tests and boundary checks for the existing autolens and autofit profiles. Confirm AutoCTI remains intentionally outside REFERENCE_PROFILES.
5. Keep the PyAutoMemory structure allowlist correction in its existing Python 3.12 workflow task and issue feat: intake census + dashboard modes — Mind backlog page #30.

Key Files

• PyAutoBrain/agents/conductors/clone/_clone.py — shared generic clone classification.
• PyAutoBrain/tests/test_clone_conductor.py — focused classification regression coverage.
• autolens_assistant/modes/maintainer.md — lens assistant boundary documentation.
• autofit_assistant/modes/maintainer.md — fit assistant boundary documentation.
• autocti_assistant/modes/maintainer.md — CTI assistant boundary documentation.

Original Prompt

Click to expand starting prompt

clone-boundary is red on every autolens_assistant branch — AI_POLICY.md unclassified

Difficulty: small

The clone-boundary CI check in @autolens_assistant has failed on every branch
since feature/ai-policy merged on 2026-07-27. It is not branch-specific work that
breaks it — the check fails on the merge commit of any PR:

check_boundary: 1 file(s) in autolens_assistant fall on neither side of the template
boundary:

  ✗ AI_POLICY.md

AI_POLICY.md was added without being classified, and check_boundary.py refuses any
unclassified tracked file — "the Clone Agent refuses to birth a newborn while any is
unclassified", so this also blocks every future assistant birth, not just CI.

Fix

Classify AI_POLICY.md as generic, domain or mixed and record it in both
places, which the check requires to agree:

1. @autolens_assistant modes/maintainer.md, ## Assistant-as-template — the prose that
   owns the boundary.
2. @PyAutoBrain agents/conductors/clone/_clone.py —
   REFERENCE_PROFILES['autolens_assistant'].

Almost certainly generic: the AI-use policy is framework-level, not lensing science,
and a newborn assistant should inherit it near-verbatim.

Grounding already done (2026-07-30):

• AI_POLICY.md is a root-level file absent from _SHARED_GENERIC
  (_clone.py:52-72), which is why it lands unclassified. Adding "AI_POLICY.md" to
  that shared list covers both existing profiles at once.
• All three cells (autolens_assistant, autofit_assistant, autocti_assistant) carry
  an AI_POLICY.md, and none of their modes/maintainer.md files mention it — so the
  prose half is missing everywhere, not just in the lens cell.
• REFERENCE_PROFILES has entries for autolens_assistant and autofit_assistant only;
  autocti_assistant has no profile, and no boundary check ran on autofit_assistant's
  PRs — which is why only the lens cell goes red. Decide whether the other two should
  gain the check as part of this fix or in a separate task.

Related

• A separate AI_POLICY.md structure-test failure on origin/main is already
  recorded as the ship-blocker for python-312-memory-validation-ci in
  active.md (@PyAutoMemory). Check whether that is the same root cause before
  fixing either in isolation.
• Found while shipping assistant-output-folder-pointer
  (docs: assistants point novices at the output folder when a fit starts autolens_assistant#96 / feat(hygiene): crlf / config / artifacts file-hygiene modes (phase 5) #97), whose own content is unrelated to it.

[Jammy2211]

Cleanup PRs are open and labeled pending-release:

• PyAutoBrain#182 — shared generic classification + focused regression test
• autolens_assistant#98 — maintainer boundary prose
• autofit_assistant#27 — maintainer boundary prose
• autocti_assistant#17 — maintainer boundary prose; no new clone profile

Validation: 28 Clone Agent tests pass; autolens and autofit boundaries are complete locally; AutoCTI correctly takes the no-profile skip path. Heart YELLOW 70 was explicitly acknowledged with reasons workspace validation not passing (0 failed, cloud#30516167217) and release validation stale: source moved since rehearsal (PyAutoFit, PyAutoGalaxy, PyAutoLens).

Merge order: PyAutoBrain#182 first, then the assistant PRs. Autolens CI reads PyAutoBrain main, so its current AI_POLICY.md boundary failure is expected until #182 lands.

[Jammy2211]

CI settled:

• PyAutoBrain#182: no configured PR checks; local focused suite 28/28 green
• autofit_assistant#27: GREEN
• autocti_assistant#17: GREEN
• autolens_assistant#98: two confirmed base dependencies only — AI_POLICY.md remains unclassified until Brain#182 is on main, and the stale example_cpu.py citation remains until autolens_assistant#97 is on main

Safe merge order: Brain#182 → rerun/merge autolens#97 → refresh and merge autolens#98. The fit and CTI documentation PRs are independent. No merges performed yet.

[Jammy2211]

Shipped: PyAutoBrain#182, autolens_assistant#98, autofit_assistant#27, and autocti_assistant#17 are merged. The focused Clone Agent suite passed (28 tests), both supported reference boundaries are complete, and AutoCTI remains intentionally outside the reference profiles. This issue is ready for human close.

[Jammy2211]

Shipped via PyAutoBrain#182 (merged 2026-07-30) plus assistant PRs #98/#27/#17; record complete/2026/07/ai-policy-validator-followups.md (ready for human close). Closing on 2026-07-30 human cleanup authorization.