[Jammy2211]

Summary

Search-extensibility A5 (consumers): the assistant's search rosters are generated from the installed autofit release's search manifest instead of written by hand. Merge after PyAutoLabs/PyAutoHands#311 (generator, branch sha 5e4224e).

• skills/af_configure_search.md: the inline 15-name roster sentence becomes a style=list block (family, af.<Name>, experimental marks, worked-example citation). Adds the citation of measured per-task behaviour: autofit_inference:catalogue/search_catalogue.json at commit 3463aa19014b6bd83563018568f0fb4a2376f8eb (protocol gaussian_x3@1; wave 1 ranks nothing — coverage, not a leaderboard). The branches (nested / MCMC / MLE) stay hand-written.
• wiki/core/concepts/non_linear_search.md: the STALE 7-row roster table (no NSS, SMC, BFGS, MultiStart*) becomes a style=table block — all 15 searches with status, JAX use, gradient, evidence, posterior kind, example — plus the capability-matrix link. The one-line "character" notes stay hand-written below it. The "Choosing" pick guidance is untouched. Frontmatter re-pinned to PyAutoFit 3f4699b959e387b2eddca4508585c7812dc557be, last_updated: 2026-10-15, content_sha256 restamped with --write-provenance --page.
• .github/workflows/wiki-currency.yml: clones PyAutoHands into sources/ (so the existing PR-branch matching pairs it with #311) and runs check_search_roster.py --root . against the installed release; a changed roster fails the job, a stamp-only difference is a note.

Decisions taken (decide-and-flag)

1. Stamped by autofit 2026.10.12.1 (current PyPI release, clean py3.12 venv) — the worktree's source-checkout autofit reports a stale 2026.8.17.1; rosters are identical between them.
2. The generated table replaces the hand table's "character" column; those notes moved to a hand-written bullet list under it (prose outside markers is never generated).
3. The catalogue citation uses autofit_inference: notation; that repo is not in sources.yaml, so --check-citations does not resolve it (it is a pinned-commit pointer, as the ruling asks).
4. Literature pages: grep for UltraNest / PySwarms / Particle Swarm over wiki/ and skills/ finds nothing to flag in this assistant.

Heart RED override

Quoted from #310 ("Heart RED override"):

The live human answered the AskUserQuestion "Invoke the Heart RED development override for this batch?" with "Yes, override for development (Recommended)" on 2026-10-15, for tasks search-ext-b5b-score, search-ext-a5-consumers and search-ext-b4b-dev-retire (this issue: search-ext-a5-consumers).

Exact RED reasons at the door (Heart verdict, 2026-10-15): "Heart: RED — Release RED; monitoring RED · 10/100 · 415 unresolved"; "[red] release validation FAILED (stage integrate)"; "[red] PyAutoHands/workflows/Python Version Matrix: failure"; "[red] PyAutoHeart/workflows/Release Integrate: failure"; "[red] PyAutoMind/workflows/Lifecycle Drift: failure"; "[red] PyAutoArray/Tests: red"; "[red] autofit_workspace_test/Smoke Tests: red"; "[red] autogalaxy_workspace_test/Smoke Tests: red"; "[red] autolens_workspace_test/Smoke Tests: red"; "[red] Release validation: NOT release_ready — v2026.10.15.1.dev81701 profile=release (2026-10-15T07:04:34+00:00)"; "[red] Worktree drift: 5 orphan / 0 missing / 4 dirty"; "[red] autolens_inference: red"; "[red] autolens_profiling: red"; "[red] euclid_strong_lens_modeling_pipeline: red"

Permitted: commit, push, PR-open. Merge stays a separate human act; no release or rehearsal; this branch does not fix Heart.

Passed branch gates (tests, smoke, review) are recorded in the PR body at ship; a failed gate cannot be overridden.

Test plan

• Witness — --check regenerates byte-identically against the installed release (strict, stamp included):

Search roster check against <!-- search-manifest@1 autofit 2026.10.12.1 -->

(a) roster blocks: 2 file(s)
  all blocks regenerate byte-identically
rc=0

• wiki-currency checks locally against the installed release (autofit 2026.10.12.1 venv): --check-version clean, --scope all exit 0, --lint-idioms clean, --check-provenance 0 errors, --check-citations 0 missing.
• Tests / smoke: n/a — documentation + workflow only; no test dir change, no workspace script surface.

Ship gate

Leg	Result
1 Tests	PyAutoHands full suite 507 passed (+ test_search_roster.py 28 after the last fix); PyAutoBrain full suite 1265 passed (one test, test_grouped_organ_consumers, fails only when activate.sh exports PYAUTO_MIND, and passes in a clean env; unrelated to this diff); the doc repos have no test dir (n/a); PyAutoMemory make validate green
2 Smoke	n/a: no script in any smoke_tests.txt changed. al/ag guides/modeling/searches.py is a docstring-only edit (py_compile clean, not smoke-listed). The workspace smoke CI on these PRs is the record.
3 Review	Two independent fresh-context reviews over Hands + Brain. Round 1 was FINDINGS (1 major, 12 minor/nit, 1 status note); all fixed in Hands e720263 and Brain db0916b. Round 2 confirmed 12/13 resolved, with one partial: the installed-mode manifest could still read a source checkout via the cwd. That is fixed in Hands 5e4224e with a test. No third round was run (decide-and-flag).
4 Heart	RED. The human development override is recorded on PyAutoLabs/PyAutoHands#310 (quoted above).
Witness	6/7. ✓ --check regenerates byte-identically against the installed autofit 2026.10.12.1 in every consumer (output shown in each consumer PR). ✓ navigator --roster writes the curated llms.txt block. ✓ af/al/ag assistant blocks come from the same manifest. ✓ Brain promoted tier and gaps are keyed by manifest name and integration_test. ✓ The workspace CI search_roster job checks anchors against the installed release, not main; it goes live once Hands merges. ✓ al/ag guides are domain advice plus a matrix link. ✗ Capability matrix → B3 catalogue card link: this is in PyAutoFit docs/_generate_searches.py, which is not in this task's repos.
Firewall	PyAutoBrain needs the Mind FIREWALL_ALLOWLIST to add "autofit_inference" for agents/faculties/samplers/_samplers.py and samplers.sh. Until then the Brain PR's firewall leg is red; its pytest leg is green.

🤖 Generated with Claude Code