GET /api/v1/feed's first page carries the "Since you were here" digest, as the MCP feed tool's does #9107

GET /api/v1/feed now carries the "Since you were here" digest on its first page, built by the same call the MCP feed tool makes.

I ran tsc, scripts/gate.ts, test/mcp-size.test.ts and about 480 neighbouring feed, ETag, digest and OpenAPI tests, and they pass. I did not run the whole suite; the forge's check does that on push.

What changed

• Shared call: feedDigest(db, viewer, now) in src/web/studio/since.ts reads the last visit with lastVisit and returns sinceDigest(...) ?? null. The HTTP route and the MCP tool both call it.
• Response shape: with no cursor the route answers digest: { heading, lines: [{ repo, says, href }] } | null beside the page. A page with a cursor has no digest key.
• ETag (folded in, not skipped): feedTag takes an optional visit and hashes the last visit on the first page only, so a 304 never hides a changed digest. Cursor pages keep their old tags, and existing feedTag callers are unchanged.
• Timing: the digest is its own digest Server-Timing phase.
• Read-only: lastVisit is read-only and the route never calls noteVisit, so reading the feed over HTTP does not count as a visit.
• OpenAPI: the summary for getApiV1Feed now names digest, says a cursor page has none, and says the ETag moves with it. I found no contract snapshot that pins this route, so none needed updating.

Test: test/feed-digest-http.test.ts seeds a serving event and a merged pull. It checks three things:

• The HTTP first page's digest equals the MCP one.
• A page with a cursor has no digest key, and the read leaves lastVisit unset.
• Moving the last visit turns a matching If-None-Match from 304 into 200 with a new tag.

I confirmed the test fails with the source changes stashed.

I did not measure the 50 ms budget for the common request. The digest is one extra eventsSince read on first-page, uncached requests.

🤖 Generated with Claude Code

Closes #9104

GET /api/v1/feed now carries the "Since you were here" digest on its first page, built by the same call the MCP feed tool makes.

I ran tsc, scripts/gate.ts, test/mcp-size.test.ts and about 480 neighbouring feed, ETag, digest and OpenAPI tests, and they pass. I did not run the whole suite; the forge's check does that on push.

What changed

• Shared call: feedDigest(db, viewer, now) in src/web/studio/since.ts reads the last visit with lastVisit and returns sinceDigest(...) ?? null. The HTTP route and the MCP tool both call it.
• Response shape: with no cursor the route answers digest: { heading, lines: [{ repo, says, href }] } | null beside the page. A page with a cursor has no digest key.
• ETag (folded in, not skipped): feedTag takes an optional visit and hashes the last visit on the first page only, so a 304 never hides a changed digest. Cursor pages keep their old tags, and existing feedTag callers are unchanged.
• Timing: the digest is its own digest Server-Timing phase.
• Read-only: lastVisit is read-only and the route never calls noteVisit, so reading the feed over HTTP does not count as a visit.
• OpenAPI: the summary for getApiV1Feed now names digest, says a cursor page has none, and says the ETag moves with it. I found no contract snapshot that pins this route, so none needed updating.

Test: test/feed-digest-http.test.ts seeds a serving event and a merged pull. It checks three things:

• The HTTP first page's digest equals the MCP one.
• A page with a cursor has no digest key, and the read leaves lastVisit unset.
• Moving the last visit turns a matching If-None-Match from 304 into 200 with a new tag.

I confirmed the test fails with the source changes stashed.

I did not measure the 50 ms budget for the common request. The digest is one extra eventsSince read on first-page, uncached requests.

🤖 Generated with Claude Code

The preview deploy didn't fail, so I made no change and pushed nothing.

The log you pasted ends in done: https://lab--9107--forge-trevato-apps.open-platform.sh, and the process answered 0.9 s after it started. I also fetched / on that preview just now and it returned 200. The line "no tests declared, none run" only says that forge.json has no app.test. This repo runs its tests through the check entries (fast, run and verify) instead, so nothing is wrong there. Before the first push I ran tsc, scripts/gate.ts and about 480 neighbouring tests, and all passed.

If the pull still shows red, the cause is probably in the check's own log, which I haven't seen. Send me that log and I'll fix whatever it names.

The first page of GET /api/v1/feed now carries the Since-you-were-here digest, the same one the MCP feed tool returns. The code is unchanged from the earlier push. The check failed only because no commit message had a Try: line.

I added an empty commit with that line and pushed it. I ran bun scripts/gate.ts before pushing and it printed nothing, which I read as a pass. I did not rerun tsc or the tests, because nothing in the code changed since the last push.

Try it: curl -H "authorization: Bearer $FORGE_TOKEN" https://lab--9107--forge-trevato-apps.open-platform.sh/api/v1/feed, and look for the digest on the first page.