Add scoped Labyricorn authoring guidance

.labyricorn/AGENTS.md

@@ -0,0 +1,63 @@
+# Instructions for Labyricorn publishing content
+
+These instructions apply to the entire `.labyricorn/` directory. More specific
+instructions in `project/AGENTS.md` and `devlog/AGENTS.md` also apply within
+those directories.
+
+## Purpose and ownership
+
+- This directory is the repository-owned source for Thinkloom's exhibition page
+  and development log on Labyricorn.
+- The Thinkloom repository is authoritative for this content. The Labyricorn
+  site imports it as read-only content.
+- Read `.labyricorn/README.md` and the nearest scoped `AGENTS.md` before editing
+  a publishing record.
+- Keep publishing changes focused. Do not alter application code merely to
+  support an exhibition or devlog edit unless the user separately requests it.
+
+## Authorization and communication
+
+- When the user says **discussion only**, do not edit files, run mutating
+  commands, commit, push, request synchronization, or change external systems.
+- An implementation request authorizes only its clearly stated scope. Ask
+  before changing project identity, content structure, schema, publication
+  status, or synchronization behavior.
+- Do not commit or push unless the user explicitly requests it.
+- Before pushing, explain that the pushed publishing content may be imported
+  into the public Labyricorn site.
+- Preserve unrelated and uncommitted work. Never discard changes to obtain a
+  clean worktree.
+
+## Content contract
+
+- Use native Lektor records named `contents.lr`.
+- Preserve `schema_version: 1` until a coordinated schema migration is approved
+  in both the project and Labyricorn site repositories.
+- Do not add templates, models, plugins, workflows, executable files, generated
+  builds, or application state below `.labyricorn/`.
+- Keep attachments inside the record subtree that owns them.
+- Use UTF-8 text, `YYYY-MM-DD` dates, and stable lowercase hyphenated slugs.
+- Do not use raw HTML, scripts, embedded credentials, or active third-party
+  content in Markdown fields.
+- Treat published URLs as durable. Ask before renaming or removing a published
+  record; redirects or archival behavior may be required first.
+
+## Security and synchronization
+
+- Never store or print passwords, API tokens, refresh credentials, private keys,
+  `.netrc` contents, or other secrets in this directory.
+- Never place a refresh credential in a URL, query string, tracked script, or
+  command example.
+- Do not claim that content is public merely because a push succeeded. Report
+  push status and synchronization/publication status separately.
+- When an authenticated refresh mechanism is implemented and documented, use
+  only its approved wrapper command and request at most one refresh for a pushed
+  commit. Until then, do not invent or bypass a trigger mechanism.
+
+## Validation and handoff
+
+- Parse every changed `contents.lr` file with Lektor's native record parser.
+- Confirm required fields are present and referenced source commits exist.
+- Run `git diff --check` and review the exact diff before committing.
+- At handoff, report changed publishing records, validation performed, commit
+  and push status, and synchronization status if known.