A pull's latest check reads as plain text at /<owner>/<repo>/pulls/<n>/check.txt for anyone who can read the pull #12226

GET /<owner>/<repo>/pulls/<n>/check.txt now answers a pull's latest check as plain text. I ran test/contribute-check-text.test.ts and test/pages-budget.test.ts (7 tests, all pass), and bun x tsc --noEmit is clean. I didn't run the full suite; the forge runs it on the push. scripts/gate.ts isn't in this clone, so I didn't run it.

• Response: it comes back as text/plain; charset=utf-8, no-store, with the security floor applied. Line 1 is green, red or pending, and line 2 is head: <oid>.
• States: queued, running and no check all read pending. Failed and cancelled both read red.
• Log tail: on red, a blank line and then the last 80 lines of the log follow. The tail goes through the same runLog door the log page uses. A reader that door hides the log from gets the verdict and head only.
• Access: it goes through loadPull, so an unknown pull, or one the reader cannot see, is 404.
• Files: the handler is the new src/http/check_text.ts, with its route registered next to the other pull routes in src/http/server.ts. The test covers green, red with its 80-line tail, pending in all three forms, and the 404. It also checks that an anonymous reader and a plain signed-in reader get no tail.
• Page budget: I raised MAX_UNFIXTURED from 147 to 148 in test/pages-budget.test.ts, with a comment. The new route has no page to fixture.
• Not done: the route doesn't read the pull page's checkWords summary, and it isn't linked from the pull page. The issue named the route only.
• Push hook: the pre-push hook wanted a commit saying how to try the change, so there is an empty Try: commit on the branch.
• Failing files: none outside this change that I know of; I ran only the two test files above.

To try it, open /<owner>/<repo>/pulls/<n>/check.txt on any pull.

I changed nothing for this one, because I don't think my code caused the failure.

• Preview is up: https://lab--12226--forge-trevato-apps.open-platform.sh/ returns 200.
• The message: "no tests declared, none run" says the deploy step ran no tests. forge.json has no test entry, and I didn't add or remove one. The log excerpt you pasted is only the forge's periodic SQL and process lines, with no error from my change.
• My tests: test/contribute-check-text.test.ts and test/pages-budget.test.ts passed in the sandbox (7 tests), and tsc is clean. The forge's check runs on the push, and I haven't seen a red from it.
• The route on the preview: /alice/notes/pulls/1/check.txt there gives a 404 with the forge's JSON not-found body. The preview has no alice/notes pull, so that 404 is what an unknown pull should return. It doesn't show that the route is registered there.

If the deploy is still marked failed, I'd need the part of its log that names the failing step. Adding a test entry to forge.json would change how every deploy of this repo runs, and I don't think that belongs in this pull.

I merged main into this branch and pushed it. The only conflict was in the import block of src/http/server.ts, and I kept both sides' imports: checkText from this pull and appSecretGrant/appSecretRevoke from main.

• Ran and passing: bun x tsc --noEmit, plus test/contribute-check-text.test.ts and test/pages-budget.test.ts (7 tests pass).
• Not run: the full suite. scripts/fast.ts reported the diff as too broad to name tests, so I picked the two files that cover check.txt. One earlier bun test invocation, run before I chose files, started the whole suite; I stopped it. I'm not sure bun scripts/gate.ts finished before I killed that run, so treat the gate as unconfirmed.
• Still to run: the forge's own check on this push, which runs the full suite.

The change itself is unchanged: a pull's latest check reads as plain text at /<owner>/<repo>/pulls/<n>/check.txt.

A pull's latest check now reads as plain text, and I've pushed a fix for the red check. Two of the failing tests I could not fix or confirm here.

What changed

• src/http/check_text.ts, wired in through src/http/server.ts, renders a pull's latest check as plain text. test/contribute-check-text.test.ts covers it, and test/pages-budget.test.ts was adjusted.
• The red check came from stale tests, not from this feature. I merged the latest main and fixed two of them:

- test/first-screen-line.test.ts used a client-row fixture without placement and where, which forgeRow now requires. I added both fields. The fixture was wrong, not the code, and all 5 tests in the file pass. - docs/MCP.md was stale for the forges tool. I regenerated it with bun scripts/mcp-table.ts.

What I could not resolve

• spawn-in-process-visible ("a request it answers is the stall watcher's child") still fails locally. The sandbox denied an outbound connection to homie--forge.test, so it looks environmental. I did not change it.
• Three other failures from the check log I did not touch:

- the deep-read record test; - the slow GET page test, 6.29 s against a 2 s budget; - the private-app credential test.

They are all unrelated to this change. I did not re-run the first two. I ran the file holding the third (boundary-tokens-follow) and it passed.

• The full suite was not run. After the fixes I ran only the first-screen, check-text, pages-budget, MCP-table and the other files named above. bun x tsc --noEmit and bun scripts/gate.ts printed no errors.

To try it: bun test test/first-screen-line.test.ts test/contribute-check-text.test.ts.