AI CSV / Spreadsheet Insights Sidebar

[LukasHirt]

AI-generated · OSPO-54 · Gate: ✅ 1.00

Problem

Data workers store CSV exports and data dumps in oCIS for team sharing, but the platform gives no quick insight layer — every analysis requires downloading the file and opening a spreadsheet app. Teams lose time cycling between tools just to answer "what's in this file?"

Solution

Selecting a CSV file shows an "Insights" tab in the right sidebar. The extension parses the file client-side (headers + first ~200 rows), then sends a structured preview to the configured LLM endpoint. The panel renders column types, value ranges, detected patterns (e.g. date trends, outliers), and 2–3 natural-language observations. Tier 1 (text completion): prose analysis split on section headers. Tier 2 (no LLM configured): panel is hidden.

Extension points

global.files.sidebar (sidebarPanel) · global.files.context-actions (action)

Why ship this now

oCIS is increasingly used as a team data hub; the CSV use-case is high-frequency and zero existing extension touches tabular files. Gives analysts an instant "first read" without leaving the browser.

What was built

web-app-ai-data-insights-sidebar registers two extension points: a sidebarPanel that appears as an "Insights" tab when a .csv or .tsv file is selected, and a context-action entry in the file list. Both are gated at registration time — the extension checks that an LLM endpoint is configured and that the selected file is a supported type, so neither entry point appears when neither condition is met. There is no Tier 2 fallback UI; when no LLM is configured the panel is simply absent.

The core data path lives in two composables. useInsights fetches the raw file text over WebDAV, runs it through a client-side RFC-4180 state-machine parser (csv-parse.ts) that handles quoted fields, embedded newlines, and CRLF, then derives column types (boolean / number / date / string), value ranges, and up to five representative samples per column. It caps output at 30 columns before sending the structured preview to the LLM endpoint as a json_object request. A same-origin check is enforced before the POST so credentials are never sent cross-origin. useLlm is a thin wrapper around the configured endpoint, following the pattern established in web-app-ai-doc-summary; the async keyword was removed from ensureReady to satisfy the require-await lint rule without changing behavior.

InsightsPanel.vue renders four distinct states driven by composable-exposed reactive refs: an "Analyzing…" placeholder while the request is in flight, a red alert banner on error, a column-types table with Range column plus an observation list on success, and an idle state with a primary "Analyze" button. A "Re-analyze" button is available once a result has been returned. All ODS utility classes and oc-button usage follow the patterns from SummaryPanel.vue and AltTextPanel.vue.

Unit test coverage spans the parser (including RFC-4180 edge cases), file-type detection, both composables (24 tests for useInsights covering the full HTTP error matrix and non-JSON graceful handling), and all four panel render states. The Playwright acceptance suite uploads a minimal sample.csv fixture, asserts that the "Insights" context-menu entry is visible for CSV and invisible for a JPEG, then confirms the sidebar panel mounts when the action is clicked. Analysing files larger than 200 rows or wider than 30 columns is explicitly out of scope; the parser truncates silently at those limits before the LLM call.

Gate

Check	Result
Hygiene	✅ ok
Build	✅ ok
Lint	✅ ok
Unit tests	✅ ok
E2E tests	✅ ok
Score	1.00

---

_{Effort: S · 🤖 Generated by extctl}

[kw-security]

✅ Snyk checks have passed. No issues have been found so far.

Status	Scan Engine	Critical	High	Medium	Low	Total (0)
✅	Open Source Security	0	0	0	0	0 issues
✅	Licenses	0	0	0	0	0 issues
✅	Code Security	0	0	0	0	0 issues

💻 Catch issues earlier using the plugins for VS Code, JetBrains IDEs, Visual Studio, and Eclipse.