- base.py: Storage ABC, FileMetadata, UploadInstructions, StorageCapabilities
- hookspecs.py: register_files_storage_types hook for plugin registration
- filesystem.py: Built-in FilesystemStorage with upload, read, list, delete
- __init__.py: Startup (schema + source instantiation), upload/serve/download
  routes, sources API, file info page, CSRF skip for uploads
- templates/file_info.html: HTML page showing file metadata
- pyproject.toml: Remove boto3, update to datasette>=1.0a24
- 16 passing tests covering base classes, filesystem storage, upload flow,
  file info/download/metadata endpoints, and sources API

https://claude.ai/code/session_016xxhVYZgWByonNJju2Zmuh

https://claude.ai/code/session_016xxhVYZgWByonNJju2Zmuh

Demonstrates the complete datasette-files filesystem workflow:
sources API, file upload via curl, metadata retrieval, download,
HTML file info pages (with image preview), error handling, and
on-disk storage layout.

https://claude.ai/code/session_016xxhVYZgWByonNJju2Zmuh

https://claude.ai/code/session_016xxhVYZgWByonNJju2Zmuh

- render_cell hook detects df-* file IDs in any table cell and emits
  <datasette-file> custom element with a fallback <a> link for no-JS
- Batch JSON endpoint (/-/files/batch.json?id=...) fetches metadata
  for all file IDs on a page in a single SQL query
- <datasette-file> web component collects all instances on page load,
  batch-fetches via single HTTP request, then hydrates each cell with
  filename links (non-images) or inline thumbnail previews (images)
- extra_js_urls hook injects the JS module on table/row/database views

https://claude.ai/code/session_016xxhVYZgWByonNJju2Zmuh

Showboat demo walking through the batch.json endpoint, render_cell hook,
web component hydration, and browser rendering with inline thumbnails.

https://claude.ai/code/session_016xxhVYZgWByonNJju2Zmuh

…search

- Define FileSourceResource as a top-level Resource for Datasette's SQL
  permissions system, with register_actions for files-browse, files-upload,
  and files-delete (all default deny)

- Add FTS5 virtual table (datasette_files_fts) with triggers to keep the
  index in sync on insert/update/delete, plus backfill at startup

- Implement permission_resources_sql hook to process files-browse config
  from datasette.yaml permissions block (global and per-source)

- Add /-/files/search endpoint (HTML + JSON) that uses
  allowed_resources_sql() to get browsable sources, then filters FTS and
  listing queries to only those sources

- Enforce files-browse permission on file_info, file_json, file_download,
  and batch_json endpoints

- 35 tests covering FTS indexing, search with/without query, source
  filtering, default deny, multi-source permission isolation, and
  actor-specific access control

https://claude.ai/code/session_016xxhVYZgWByonNJju2Zmuh

Showboat demo document showing:
- Per-source permission config (public-docs: everyone, team-docs: alice only)
- Anonymous vs authenticated search results (4 vs 6 files)
- FTS5 search by filename
- Permission enforcement on individual file pages (403 vs 200)
- Browser screenshots of the search UI for both actors

https://claude.ai/code/session_016xxhVYZgWByonNJju2Zmuh

- Add `search_text` TEXT column to datasette_files table, included in the
  FTS5 index so users can attach searchable descriptions to files

- Register `files-edit` action (abbr: fe) scoped to FileSourceResource,
  default deny like all other file actions

- File info page (/-/files/{id}) now handles POST to update search_text,
  gated by files-edit permission. Shows textarea form for actors with
  edit permission, read-only text or "No search text set" for others

- Startup migrates existing databases: adds search_text column if missing,
  drops and recreates FTS table + triggers to match new schema, backfills
  FTS index

- 43 tests: 8 new tests covering search_text column existence, FTS
  indexing of search_text, files-edit action registration, POST with/without
  permission, form visibility, and non-editor denial

https://claude.ai/code/session_016xxhVYZgWByonNJju2Zmuh

python-ulid imports typing_extensions but doesn't declare it as a
dependency. On Python 3.13 it isn't pulled in transitively, causing
an ImportError.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>