fix(ui): unblock .pptx upload in Agent UI

[itomek]

The Agent UI hard-blocked PowerPoint uploads even though .pptx RAG shipped end-to-end in #1224 — the RAG backend and the FastAPI ALLOWED_EXTENSIONS already accept .pptx, but the React UnsupportedFeature gate still classified it as an unsupported "Microsoft Office" type and rejected it before upload. A headline feature was unreachable in the UI, and users were filing feature requests (#1072, #1291) for something that already worked. This drops .pptx from the frontend blocklist, adds it to the supported set (now matching the server allowlist exactly), and fixes the now-false rejection copy. .doc/.docx/.ppt/.xls stay blocked. No server change — the backend already accepts .pptx.

Closes #1363

Test plan

• cd src/gaia/apps/webui && npm test — Vitest 18/18 pass locally (new assertions verify .pptx is accepted and .doc/.ppt/.xls stay blocked). Note: the test-webui-vitest CI job doesn't currently trigger on diffs confined to src/gaia/apps/webui (pre-existing path-filter gap in test_electron.yml), so this was verified locally, not in CI.
• cd src/gaia/apps/webui && npm run build — tsc && vite build clean
• code-reviewer pass — no Critical/Advisory findings; independently confirmed the frontend SUPPORTED_EXTENSIONS now matches the server-side ALLOWED_EXTENSIONS and no consumer (DocumentLibrary, FileBrowser) breaks
• Real-world (Agent UI on an AMD GPU box, Gemma-4-E4B): uploaded a .pptx through the Document Library → accepted (no "Microsoft Office not supported" toast) → indexed (1 chunk via nomic-embed) → Doc Agent answered questions grounded in slide body text (92.7% success rate), a table cell (14.5 Nm torque), and speaker notes (Bluejay-44 codename)

[github-actions]

Summary

Correct targeted fix: .pptx was silently rejected by the React upload gate even though the backend (ALLOWED_EXTENSIONS in src/gaia/ui/utils.py) and RAG pipeline have accepted it since #1224. The frontend SUPPORTED_EXTENSIONS set and UNSUPPORTED_FILE_CATEGORIES blocklist now match the server allowlist exactly. Two files changed, tests pin the behavior in both directions. Three minor/nit findings; none are blocking.

---

Issues Found

🟢 Minor — featureTitle omits .doc and .ppt from the parenthetical (UnsupportedFeature.tsx:97)

The blocked set after this PR is ['.doc', '.docx', '.ppt', '.xls'], but the feature-request title reads 'Support Microsoft Office (docx, xls) indexing'. A user filing that feature request via the link won't see .ppt or .doc in the title.

        featureTitle: 'Support Microsoft Office (doc, docx, ppt, xls) indexing',

🟢 Nit — Issue reference in test description will rot (UnsupportedFeature.test.tsx:61)

The test name 'returns null for .pptx — PowerPoint indexing shipped in #1224' embeds a GitHub issue number. Per CLAUDE.md conventions, inline issue references rot — the assertion should stand on its own without tracking history.

    it('returns null for .pptx — PowerPoint indexing is supported', () => {

🟢 Nit — First alternative shows "Save as .pptx" to Word users (UnsupportedFeature.tsx:93)

All three alternatives render for every extension in the Microsoft Office category. A .docx upload now shows 'Save modern PowerPoint as .pptx — GAIA indexes .pptx directly' — PowerPoint advice for a Word user. The message body (line 89) already mentions .pptx support, so this alternative adds noise rather than signal for Word/Excel cases.

The cleanest fix here depends on how downstream consumers render alternatives — no one-click suggestion, but worth a follow-up since .pptx now being supported makes the .ppt-specific advice the primary reason this entry exists.

---

Strengths

• Exactly minimal: only the two files that needed updating are touched — no collateral reformatting or unrelated edits.
• Bidirectional test coverage: new tests explicitly assert .pptx is unblocked AND that .ppt/.doc/.xls remain blocked — the risk of regressing in either direction is very low.
• Code comment update at UnsupportedFeature.tsx:120–121 is a high-value invariant note: it explains why the sets are maintained separately and calls out the python-pptx dependency explicitly, which will save the next contributor from re-checking the backend.

---

Verdict

Approve with suggestions. The core fix is correct and well-tested. The featureTitle nit (#97) is a one-click apply; the test-name nit is low-stakes but easy to apply. Neither blocks merge.

[itomek]

Addressed in c955d17:

• featureTitle — now Support Microsoft Office (doc, docx, ppt, xls) indexing, matching the blocklist (.doc/.docx/.ppt/.xls) exactly so the auto-filed feature-request title is accurate.
• Test name — dropped the #1224 reference; the assertion now reads returns null for .pptx — PowerPoint indexing is supported and stands on its own.

Deferring the third nit (per-extension alternatives): the alternatives list is rendered category-wide for every Office extension by design — the pre-fix code did the same — so making the .ppt → .pptx tip conditional on the uploaded extension is a rendering change beyond this targeted gate fix. Worth doing as a follow-up rather than widening scope here.

Vitest remains 18/18; neither change touches asserted behavior.

[github-actions]

A focused, correct fix — removes a frontend gate that was silently blocking a feature the backend already supported. Code is clean and all affected paths are consistently updated.

Summary

.pptx support shipped in the RAG backend (#1224) but the React UnsupportedFeature guard never got the memo, leaving users staring at a rejection toast for a capability that already worked end-to-end. This PR makes the frontend tell the truth: drops .pptx from the blocked set, adds it to SUPPORTED_EXTENSIONS, updates the error copy and the alternatives list, and fixes the featureTitle to name only the formats that are still actually missing. No backend changes needed; all three mutation sites in UnsupportedFeature.tsx are updated consistently.

Issues Found

None. No blocking issues identified.

Strengths

• Surgical scope — exactly three call sites in one file touched, zero unrelated changes. The diff is easy to reason about in its entirety.
• Tests prove the contract — new assertions verify both directions: .pptx is accepted and the legacy formats (.ppt, .doc, .xls) stay blocked. Covering both sides of the boundary is the right move.
• User-facing copy is meaningfully improved — the updated alternatives list now gives a clear ranked path (use .pptx directly → convert Word to PDF → re-save legacy Excel), which is more actionable than the previous one-size-fits-all "Save as PDF" suggestion.

Verdict

Approve — ready to merge. The fix is correct, fully tested, and the PR description is a model of how to explain user-observable impact concisely.