Export captured traffic as HAR and Markdown

[Twinsen81]

Summary

A transaction left Lustro only as JSON from the API, as a cURL command, or as plain text, one at a time. This adds the two formats that matter for sharing: HAR, which browser devtools and most HTTP tools import and agents parse without instructions, and Markdown, for a bug report, a pull request, or a chat.

HAR on the server. GET /api/v1/network/transactions/_/export?format=har&ids=a,b returns a HAR 1.2 document of the transactions in ids, or of every one when ids is left out, oldest first. It is built from the store, so it has the redacted values of the detail.

Transaction	HAR
startedAt	startedDateTime, ISO 8601 in UTC with milliseconds
durationMs	time and timings.wait; send and receive are 0, the other phases -1
protocol	httpVersion (h2), empty when unknown
headers, URL query	headers, queryString (decoded)
request body	postData; bytes in base64 with _encoding: base64
response body	content.text; bytes in base64 with encoding: base64
requestBodyBytes, responseBodyBytes	bodySize, -1 when null; content.size is the larger of the byte count and the size kept, since a gzip body is kept inflated
Location	redirectURL
error	response._error, where Chrome's own export puts it
isMocked, categories, truncation flags, responseComplete, error, id	_lustro

_resourceType (fetch, or image) tells Chrome DevTools where to list the entry. Without it, DevTools listed every JSON response as a script.

CLI. lustro net export --har FILE [--ids ID ...] writes the document, or prints it for --har -. --ids takes short ids, and the command sends them in batches of 50 and joins the results, because the server rejects a request line and headers over 8 KB. It warns about ids that the app no longer has.

Console. Select adds a checkbox column, a header checkbox that selects every row the filters show (also those past the 200 rendered), and a bar with the count, Export HAR, and Copy Markdown. A filter change keeps only the selected rows that it still shows. Export HAR fetches the same batches and saves lustro-<date>-<time>.har. Copy Markdown fetches each detail, four at a time, and copies one document. The detail gets a Markdown button next to cURL and Copy.

A transaction in Markdown is the method and URL as a heading (one code span, so a URL can't become markup), a meta line, the headers in http blocks, and each body in a block tagged with a language for its content type. JSON is indented without changing a value. A fence is longer than any run of backticks in the body. A body that capture cut short says _Truncated: Lustro kept only the first part of this body._; a body kept as bytes, or in an encoding Lustro does not decode, says that instead of a block.

Contract. The OpenAPI document describes the route and the HAR shape (HarDocument, HarEntry). The golden fixture export-har.json has five entries: a POST with a JSON body, a mock, a GIF in base64, a truncated response, and a failure. A Kotlin test checks that the export writes exactly that fixture, the CLI tests validate it against the schema, and the CLI end-to-end run exports the sample's mocked request.

Decisions to check

1. The route goes into protocol 1.2, which is not released, as the body route and captureFilter did. _meta still says 1.2. If 1.2 counts as released, this needs a bump to 1.3.
2. An unknown id is left out, not a 404. A selection can outlive an eviction, and the export should still save what remains. The CLI warns, and the console's toast says how many were no longer captured.
3. ids= with no value is an empty selection, not every transaction, so a client that sends an empty selection never gets the whole store.
4. The export is built in memory. With the default 50 MB capture budget full, a single export of everything is a large string on the device. The CLI and the console send ids in batches of 50, which bounds each response; a plain curl without ids gets it all at once.
5. postData._encoding is our own field: HAR 1.2 has encoding on content only. Tools that ignore it see base64 text for a binary request body.
6. Markdown is oldest first and pretty-prints JSON, which makes a large body larger: the 250 KB echo is about 780 K characters. It copies as is, with no size cap.
7. Copy Markdown of a selection writes a ClipboardItem with a promise when the page is a secure context, because Safari allows a copy only during the click and the details arrive later. It falls back to the text copy elsewhere.

Verification

On a physical device (Android 17), with the debug sample's traffic: GET, POST, a 500, JSON, XML, a JPEG, a PNG request body, the 250 KB JSON echo, a br response, and a 2 s timeout. The console ran in headless Chrome, driven over CDP with real mouse events.

• Select all: 10 of 10 selected. Turning off the 5xx filter: 9 of 9 selected, and the header checkbox stays checked. Unchecking one row: 8 of 9, header indeterminate. Leaving Select mode clears it.
• Export HAR saved lustro-20261002-121122.har. lustro net export --ids with the same 9 ids wrote the same bytes (cmp).
• The DevTools frontend that Chrome serves imported the console's file, the CLI's export of all 10 transactions, and the golden fixture through the Network panel's own import path (onLoadFromFile), with no HAR load failure and no console message. It listed every request with its method, status, media type, protocol, and duration; JSON under Fetch, the JPEG under Img with its base64 bytes decoded, and the 250 KB echo with its 262,144 characters.
• The selection's Markdown (8 transactions, without the 780 K-character echo) rendered through GitHub's GFM render API with 8 headings, 17 sections, 7 rules, and json, xml, and http highlighting; the JPEG, PNG, and br notes rendered as italics.
• The detail's Markdown of the 250 KB echo starts with the request heading and marks the truncated response.
• lustro net export --har on the device: 10 transactions, 735 KB, with _lustro.responseBodyTruncated on the echo and error: "Canceled" with status 0 on the timeout.

Tests:

• ./gradlew :lustro:testDebugUnitTest :lustro:lintDebug detekt apiCheck checkFacadeParity passes, with NetworkHarExportTest (12 tests) and HarExportGoldenTest.
• pytest in lustro-cli: 238 pass, 13 of them new in test_cli_export.py, plus schema checks of the golden HAR.
• node --test lustro/src/test/js/*.test.js: 100 pass, 14 of them new in network-export.test.js for the Markdown and the batch merge.

Type of change

• Bug fix (non-breaking)
• New feature (non-breaking)
• Breaking change (public API or wire protocol)
• Documentation / tooling only

Checklist

• Tests added or updated for the change
• ./gradlew apiCheck passes (no unintended public API changes; dump updated if intended)
• ./gradlew detekt passes
• CHANGELOG.md updated under [Unreleased]
• Public API changes are documented with KDoc (none)
• Wire-protocol or schema changes are reflected in wire-protocol/ and tab OpenAPI fragments
• All commits are signed off (DCO): git commit -s

Compatibility notes

Additive: a new route, within the unreleased protocol 1.2. No Kotlin API change.