… path traversal

Add regression coverage for the two behaviors wired up in the preceding fix:

- list --available/--all: queries the catalog, filters installed IDs,
  marks discovery-only entries, reports an empty catalog, and exits 1 on
  catalog failure.
- add --from <url>: a label containing path separators is sanitized so the
  download cannot escape the downloads cache dir.

Both suites were verified red against the pre-fix behavior and green after.

…ction

- Escape untrusted catalog fields (name, version, id, description,
  catalog name) and the catalog-query error before embedding them in
  Rich markup, preventing markup injection in `extension list
  --available/--all` output
- Always print the "Installed Extensions:" header when the installed
  section is shown, with a "No extensions installed." note when empty,
  so `--all` output is no longer missing the section on an empty project
- Update the command docstring to reflect --available/--all catalog
  listing so `--help` is accurate

…ries

Catalog entries are untrusted (remote/community catalogs) and only
guaranteed to carry an injected `id`; every other field comes straight
from `**ext_data`. Hard subscripts and unguarded `.get()` chains on that
data could crash `search`, `info`, `add`, and `update`, and a couple of
catalog-controlled values were printed without Rich-markup escaping.

- Use `.get()` fallbacks for name/version/description across
  search/info/add resolution and the download message
- Guard `requires`/`provides` with isinstance(dict) before `.get()`, and
  skip non-dict tool entries
- Catch KeyError alongside InvalidVersion in `update` version parsing
- Escape catalog-controlled `stars` before printing
- Correct the add --from path-traversal test docstring to describe the
  real mitigation (generated tempfile in the downloads dir, not label
  sanitization)
- Add regression tests for malformed catalog entries in list/search/info

Remote extension catalogs are untrusted, and catalog entry IDs feed into the cached ZIP filename used by extension installation. The download path now validates IDs against the manifest ID rule before catalog lookup and verifies the final ZIP path stays under the downloads directory before writing. The issue-template agent list is also synchronized with runtime integrations so consistency checks pass for newly registered agents.

Constraint: Remote catalog metadata controls extension IDs and versions before local manifest validation can run

Rejected: Sanitize path separators in the filename | silently rewriting remote IDs could install an extension under an unexpected identity

Confidence: high

Scope-risk: narrow

Directive: Keep catalog-sourced identifiers validated before using them in filesystem paths

Tested: uv run pytest tests/test_extension_catalog_robustness.py tests/test_extension_add_path_traversal.py tests/test_extension_update_hardening.py tests/test_extension_list_available.py -q

Tested: uv run pytest tests/test_agent_config_consistency.py -q

Tested: uv run python -m compileall -q src/specify_cli/extensions tests/test_extension_catalog_robustness.py

Tested: git diff --check

Not-tested: Full uv run pytest

Assisted-by: Codex (model: GPT-5, autonomous)

Co-authored-by: OmX <[email protected]>

…ersion

Address remaining Copilot review on PR github#3051:

- list --available/--all and search now skip catalog entries without a
  valid id and coalesce JSON null name/version/description to placeholders
  instead of rendering the literal "None".
- extension info gets the same null coalescing for name/version/
  description/author/license via a shared _catalog_str helper.
- download_extension sanitizes the catalog-provided version into a
  filename-safe token so a version with separators cannot produce a
  nested path or write failure, rather than relying on the post-hoc
  resolve() containment check alone.
- Fix a stale mock in test_add_from_url_sanitizes_traversal_label
  (missing extra_headers kwarg, non-ZIP body) so the suite is green.

New tests cover null-field rendering, id-skipping, the version token
sanitizer, and the download filename path.

…lbacks

- list --available: filter with _catalog_id() before the empty check so an
  all-invalid-id catalog shows the "no additional extensions" fallback
- search: count/iterate only valid-id entries and render _catalog_name via
  _catalog_str() so an explicit JSON null no longer prints "Catalog: None"
- add: render download status name/version via _catalog_str() so null fields
  fall back to the resolved id / "unknown" instead of the literal "None"
- add regression tests for each path

Remote/community catalog entries may contain null values or malformed tag data. Search filtering runs before CLI rendering fallbacks, so it must normalize unsafe field shapes before matching query, author, and tag filters.

Coerce non-string search fields to empty text, ignore non-list tags, and drop non-string tag entries during matching. Add a regression test that exercises the real ExtensionCatalog.search path by stubbing _get_merged_extensions instead of replacing search itself.