maintenance: finish provider-neutral agent setup

AGENTS.md

@@ -28,7 +28,7 @@ The repo has exactly **two content homes**, enforced by
 `make validate-structure` (CI runs it on every push/PR):
 
 - **`wiki/<domain>/`** — every sub-wiki, following the shared schema in
-  [`wiki/CLAUDE.md`](wiki/CLAUDE.md). New sub-wikis are added beside the
+  [`wiki/AGENTS.md`](wiki/AGENTS.md). New sub-wikis are added beside the
   existing ones, never at the repo root.
 - **`bibliography/`** — the *only* place BibTeX lives. One canonical file
   (`pyautomemory.bib`); never add loose `.bib` files anywhere else.

README.md

@@ -79,7 +79,7 @@ and the claim support together, then passes `make validate` (CI-enforced on
 every push).
 
 The wiki schema is defined in
-[`wiki/CLAUDE.md`](wiki/CLAUDE.md) and inherited by every
+[`wiki/AGENTS.md`](wiki/AGENTS.md) and inherited by every
 sub-wiki. How agents should read this repo: [AGENTS.md](AGENTS.md). The
 organism this repo is the Memory of:
 [PyAutoBrain/ORGANISM.md](https://github.com/PyAutoLabs/PyAutoBrain/blob/main/ORGANISM.md),

bibliography/README.md

@@ -15,7 +15,7 @@ personal library in 2026-07 — the merge audit is in git history.)
 2. Search `pyautomemory.bib` by DOI, arXiv ID, and title. Reuse the existing canonical key;
    otherwise add verified metadata under a unique, stable author-year key.
 3. Add a compact section to the relevant `../wiki/<domain>/sources/*.md` using the inherited
-   schema in [`../wiki/CLAUDE.md`](../wiki/CLAUDE.md).
+   schema in [`../wiki/AGENTS.md`](../wiki/AGENTS.md).
 4. Add concept/entity links only where the paper materially supports existing text.
 5. Run `make validate-literature-citations`.
 

index.md

@@ -3,7 +3,7 @@
 The PyAuto organism's long-term knowledge: a multi-domain knowledge base of
 accumulated scientific and project knowledge, organised as several
 Karpathy-style LLM wikis over a shared canonical bibliography. Each sub-wiki uses
-the schema in [`wiki/CLAUDE.md`](wiki/CLAUDE.md).
+the schema in [`wiki/AGENTS.md`](wiki/AGENTS.md).
 
 ## Sub-wikis
 
@@ -62,8 +62,8 @@ the schema in [`wiki/CLAUDE.md`](wiki/CLAUDE.md).
 ## Schema
 
 The schema is defined once in
-[`wiki/CLAUDE.md`](wiki/CLAUDE.md) and inherited by
-all sub-wikis. Each sub-wiki has its own `CLAUDE.md` that diverges
+[`wiki/AGENTS.md`](wiki/AGENTS.md) and inherited by
+all sub-wikis. Each sub-wiki has its own `AGENTS.md` that diverges
 only on the `## Scope` section. Per-page frontmatter, naming
 conventions, `[[wiki-links]]`, status flags (`stub | drafted |
 reviewed`), and source-page layout are uniform across sub-wikis.

reading-queue.md

@@ -28,7 +28,7 @@ Papers waiting to be read and filed. Format:
   - A line that is *only* an arXiv link is still a paper — the title was just
     never written, so the board labels it `arXiv:<id>`.
 - Filing a read paper: add its canonical entry to `bibliography/`, stub it in
-  the matching `wiki/<domain>/sources/*.md` page per `wiki/CLAUDE.md`, then
+  the matching `wiki/<domain>/sources/*.md` page per `wiki/AGENTS.md`, then
   mark the line DONE and run `make validate`.
 
 (Moved from `admin_jammy/papers.md` on 2026-05-22, when this repo — then

scripts/board.py

@@ -5,7 +5,7 @@
 how mature each sub-wiki is — each work queue carrying a one-tap 📋 copy
 block holding a paste-ready Claude Code prompt that executes this repo's own
 documented workflow (``bibliography/README.md`` "Adding a paper",
-``wiki/CLAUDE.md``'s schema) — plus contents cards with ``/memory <domain>``
+``wiki/AGENTS.md``'s schema) — plus contents cards with ``/memory <domain>``
 recall chips. Reading-queue sections expand to the individual papers: each
 title links out to the arXiv abstract page (or, when the line carries no ref,
 a title search — see ``scripts/arxiv_refs.py``), carries a 📄 button onto the
@@ -508,7 +508,7 @@ def _queue_issue_url(snapshot: dict, section: str, paper: dict,
                 f"{where}\n\n"
                 f"{notes}\n\n"
                 "Workflow (bibliography/README.md \"Adding a paper\", "
-                "wiki/CLAUDE.md): verify the paper against an authoritative "
+                "wiki/AGENTS.md): verify the paper against an authoritative "
                 "record, add its canonical entry to the bibliography/ BibTeX "
                 "file, add a minimal section to the matching "
                 "wiki/<domain>/sources/ page — canonical key plus the notes "
@@ -522,7 +522,7 @@ def _queue_issue_url(snapshot: dict, section: str, paper: dict,
                 f"{where}\n\n"
                 f"{notes}\n\n"
                 "Workflow (bibliography/README.md \"Adding a paper\", "
-                "wiki/CLAUDE.md): verify the paper against an authoritative "
+                "wiki/AGENTS.md): verify the paper against an authoritative "
                 "record, add its canonical entry to the bibliography/ BibTeX "
                 "file, stub it in the matching wiki/<domain>/sources/ page — "
                 "incorporating the notes above — mark its queue line "
@@ -661,7 +661,7 @@ def _read_prompt(snapshot: dict, section: str) -> str:
             f"{repo}/reading-queue.md: verify it against an authoritative "
             f"record, add its canonical entry to the bibliography/ BibTeX "
             f"file, stub it in the matching wiki/<domain>/sources/ page per "
-            f"wiki/CLAUDE.md, mark its queue line 'DONE <date> — <title>', "
+            f"wiki/AGENTS.md, mark its queue line 'DONE <date> — <title>', "
             f"and run make validate.")
 
 
@@ -678,7 +678,7 @@ def _stub_prompt(snapshot: dict, wiki: str) -> str:
     repo = snapshot.get("repo") or "the memory repo"
     return (f"Upgrade one stub page in {repo}/wiki/{wiki}/ to drafted: verify "
             f"its claims against the cited sources, expand it per the schema "
-            f"in wiki/CLAUDE.md, set status: drafted, and run make validate.")
+            f"in wiki/AGENTS.md, set status: drafted, and run make validate.")
 
 
 def _inbox_freshness(snapshot: dict, key: str = "inbox_last_digest") -> dict:

scripts/validate_wikilinks.py

@@ -1,8 +1,8 @@
 """Resolve every ``[[wikilink]]`` in the wiki against the pages that exist.
 
-``[[slug]]`` is the wiki's cross-reference mechanism (``wiki/CLAUDE.md``,
+``[[slug]]`` is the wiki's cross-reference mechanism (``wiki/AGENTS.md``,
 "Cross-references"): a slug is a page's filename without ``.md``, slugs are
-unique across sub-wikis so a link needs no path (``wiki/smbh/CLAUDE.md``), and
+unique across sub-wikis so a link needs no path (``wiki/smbh/AGENTS.md``), and
 a link with no target yet is *legal* — it marks a page somebody meant to write.
 
 That last rule is why this lint is a **ratchet, not a gate**. The wiki carries
@@ -64,7 +64,7 @@
 
 #: The schema page itself is documentation *of* the syntax, not a page of
 #: references: every example slug in it is illustrative.
-SKIP_FILES = ("wiki/CLAUDE.md",)
+SKIP_FILES = ("wiki/AGENTS.md", "wiki/CLAUDE.md")
 
 
 def strip_code(text: str) -> str:

tests/test_board.py

@@ -175,7 +175,7 @@ def test_work_queue_prompts_reference_the_documented_workflow(tmp_path):
     assert "reading-queue.md" in html          # file-the-next-paper prompt
     assert "make validate" in html             # every prompt ends at the gate
     assert "/memory demo" in html              # the recall chip
-    assert "wiki/CLAUDE.md" in html            # the schema anchor
+    assert "wiki/AGENTS.md" in html            # the schema anchor
 
 
 def test_paper_links_and_issue_actions(tmp_path):

wiki/AGENTS.md

@@ -0,0 +1,167 @@
+# PyAutoMemory wikis — shared schema + usage rules
+
+Every sub-wiki under `wiki/` gives an AI assistant broad scientific context
+for one domain. They follow Karpathy's "LLM Wiki" pattern: concise,
+cross-linked pages are read at query time, while canonical citation metadata
+lives separately in [`../bibliography/`](../bibliography/README.md). This
+file defines the schema once; each sub-wiki's own `AGENTS.md` diverges only
+on scope.
+
+## Layout
+
+```
+PyAutoMemory/                  # repo root
+├── wiki/                     # all sub-wikis (in git)
+│   ├── AGENTS.md             # this file — schema + usage rules
+│   └── <domain>/             # one domain per sub-wiki, e.g. lensing/
+│       ├── AGENTS.md         # the sub-wiki's scope notes
+│       ├── index.md          # the sub-wiki's own navigation
+│       ├── log.md            # append-only compilation log
+│       ├── concepts/         # one topic per page — the science
+│       ├── entities/         # named things: surveys, instruments, software
+│       ├── sources/          # compact claim support (one paper = one section)
+│       └── seed/             # unverified 2026-05 import, not citeable yet
+└── bibliography/             # canonical BibTeX and citation tooling
+```
+
+Papers are the ground truth; wiki pages are syntheses. If they disagree, update
+the wiki and note the change in the sub-wiki's `log.md`.
+
+## References and citation metadata
+
+- `sources/*.md` records compact guidance about what claims a paper supports.
+- `seed/*.md` is the same shape, unverified: the 2026-05 import's stubs, held
+  apart so nothing cites them by accident.
+- `../bibliography/pyautomemory.bib` records canonical metadata and keys.
+
+Never record local PDF paths or fabricate metadata — source PDFs live
+off-repo. One declared exception: a page's frontmatter may carry an
+`archive:` list — the **read-only archival filenames** of the off-repo PDFs
+the 2026-05 import seeded it from (e.g. `Strong_Lens/Suyu2016Holicow.pdf`).
+These are provenance history, never live paths; nothing may resolve or grow
+them — new papers cite the bibliography, not an archive path. A canonical
+key is local to PyAutoMemory: resolve it against a target project's `.bib`
+before patching LaTeX. See
+[`../bibliography/README.md`](../bibliography/README.md) for the workflow.
+
+## Page types
+
+| Type        | Folder       | Scope                                                 |
+|-------------|--------------|-------------------------------------------------------|
+| Concept     | `concepts/`  | One scientific concept (e.g. mass-sheet degeneracy)   |
+| Entity      | `entities/`  | One named thing (survey, lens, code, collaboration)   |
+| Sources     | `sources/`   | Claim support for one topic, one section per paper    |
+| Seed        | `seed/`      | Source pages from the 2026-05 import whose every entry is still a filename-inferred stub (`Canonical BibTeX key: TODO`) — never cite one until the paper is verified and the entry promoted into `sources/` |
+| Index/log   | root         | Navigation and provenance                             |
+
+## Naming
+
+- File names are lowercase kebab-case: `mass-sheet-degeneracy.md`,
+  `h0licow.md`, `time-delay-cosmography.md`.
+- One concept per concept page. If a page tries to cover two ideas, split it.
+- Source-collection pages are named by topic: `sources/dark-matter-substructure.md`.
+
+## Cross-references
+
+Use `[[page-slug]]` for wiki-internal links — for example
+`[[mass-sheet-degeneracy]]` or `[[h0licow]]`. Slugs match the filename without
+`.md`; a `sources/` page also answers to `[[sources-<topic>]]` and a `seed/`
+page to `[[seed-<topic>]]`. A `[[link]]` that has no target file yet is fine —
+it marks a future page to write, and `scripts/validate_wikilinks.py` ratchets
+that backlog (it fails only on a *new* unresolved target, against
+`scripts/wikilink_baseline.txt`).
+
+External references use verified DOI, arXiv, journal, or author/year/title
+metadata, never a local path.
+
+## Frontmatter
+
+Every wiki file starts with YAML frontmatter:
+
+```yaml
+---
+title: Mass-sheet degeneracy
+type: concept            # concept | entity | sources | meta
+topics: [degeneracies, cosmography]
+sources:                 # optional — papers most relevant to this page
+  - Suyu et al. 2017 — H0LiCOW overview
+  - Birrer et al. 2020 — TDCOSMO IV
+status: stub             # stub | drafted | reviewed
+---
+```
+
+Sources may be `[]` for pages that synthesize general field knowledge.
+
+## Concept page structure
+
+```
+# Title
+
+## TL;DR
+One paragraph an assistant can quote back to a user.
+
+## What it is
+The physics / definition.
+
+## Why it matters
+How this concept shows up in the decisions the domain's software users make.
+
+## Key results from the literature
+Bullet list. Each bullet ends with `([[author-year-stub]])` so the LLM can
+follow the link to the per-paper section.
+
+## See also
+- [[related-concept-1]]
+- [[related-concept-2]]
+```
+
+## Entity page structure
+
+Same idea but the headings are "What it is / Key facts / Papers / See also".
+Use entity pages for: surveys (BELLS, H0liCOW), specific objects
+(Abell 1201, the Cosmic Horseshoe), software (PyAutoLens, lenstronomy),
+collaborations (TDCOSMO, Space Warps).
+
+## Source-collection page structure
+
+```
+# Sources: <topic>
+
+Papers covering this topic. Each paper has its own H2 section; cross-link with
+`[[sources-<topic>#author-year-slug]]`.
+
+## Author Year — short tag
+
+**Canonical BibTeX key:** `KeyYYYY`
+**Reference:** DOI/arXiv/journal reference if known
+**Concepts:** [[concept-1]], [[concept-2]]
+
+**Supports:**
+- Claim this paper directly supports.
+- Another claim this paper directly supports.
+
+**Use when:**
+- Situation where the citation is appropriate.
+
+**Do not use for:**
+- Similar but unsupported claim.
+```
+
+Keep entries short: normally 2–5 support bullets and no long prose. Do not copy
+abstracts or infer claims from filenames. Add a TODO when support is unverified.
+
+## How the assistant should use a wiki
+
+1. On a user question, first open the sub-wiki's `index.md`.
+2. Follow the relevant `concepts/` or `entities/` page.
+3. Follow the source entry for claim scope and its canonical key for metadata.
+4. Resolve that key against downstream `.bib` files before changing LaTeX.
+5. If support or metadata is unclear, read the public paper and add a TODO
+   rather than guessing. Log verified upgrades.
+
+## Adding a sub-wiki
+
+Create `wiki/<domain>/` with its own `AGENTS.md` (scope notes only — this
+schema is inherited, not copied), `index.md`, `log.md`, and the page folders it
+needs (`seed/` only if it is taking an unverified import). Give it a row in the
+repo-root `index.md` and `README.md` tables.