The sentinel files no fix issue in trevato/forge for another repository's red main #3898

The sentinel now files a red:<repoId> span only when repoId is the sentinel's own repository, so a red main in another repo (like trevato/arm) no longer becomes a builder issue here.

What changed

• In budgetBreaches (src/crew/sentinel.ts), a red: span is skipped unless its id matches the repo sentinelRepoPath() resolves to. A skipped span stays on the degraded list.
• queue:check, stuck:, budget: and shared: are filed exactly as before, and so is a red main of the sentinel's own repo.
• I did not touch degradedCauses, the span, or red_fix.ts.

Tests

• A new case in test/sentinel.test.ts covers both sides: a red: span for another repo yields no finding, and one for the sentinel's own repo still yields it with its "Fix the failing test on main" line. I didn't run the first half without the change, but without the guard every red: span is filed, so that assertion would fail.
• bun x tsc --noEmit is clean and bun test test/sentinel.test.ts passes (33 tests).

Not done

• No surface says "main is red in another repo" beyond the existing degraded list.

Try it

• Run bun test test/sentinel.test.ts.

Closes #3895

The sentinel now files a red:<repoId> span only when repoId is the sentinel's own repository, so a red main in another repo (like trevato/arm) no longer becomes a builder issue here.

What changed

• In budgetBreaches (src/crew/sentinel.ts), a red: span is skipped unless its id matches the repo sentinelRepoPath() resolves to. A skipped span stays on the degraded list.
• queue:check, stuck:, budget: and shared: are filed exactly as before, and so is a red main of the sentinel's own repo.
• I did not touch degradedCauses, the span, or red_fix.ts.

Tests

• A new case in test/sentinel.test.ts covers both sides: a red: span for another repo yields no finding, and one for the sentinel's own repo still yields it with its "Fix the failing test on main" line. I didn't run the first half without the change, but without the guard every red: span is filed, so that assertion would fail.
• bun x tsc --noEmit is clean and bun test test/sentinel.test.ts passes (33 tests).

Not done

• No surface says "main is red in another repo" beyond the existing degraded list.

Try it

• Run bun test test/sentinel.test.ts.

The full gate on land was red and named no test; the pull is open again.

Parked by the crew: the check failed in the place it ran, not in the change, and failed the same way when it was run again

I merged main into the branch and pushed. The type check is clean, and the sentinel tests (test/sentinel.test.ts and test/tui-sentinel.test.ts) pass, 36 of 36. I did not wait for the full bun test run, so the forge's check on this push is the first full run.

• The conflict: main and this branch had both made the sentinel file a red main only for its own repository, in different ways. In src/crew/sentinel.ts I kept main's ownCause(cause, forgeRepoId) check. I extended it to cover shared: causes as well as red:, so another repository's red or shared-test failure stays on the degraded list instead of becoming an issue. I dropped this branch's own repository lookup in budgetBreaches.
• One behaviour changed: with no sentinel repository found, nothing is filtered, as on main. Before the merge, every red: and shared: span was skipped in that case.
• One test changed: "a red main is filed only for the sentinel's own repository" now passes the repository id to budgetBreaches. It relied on the old skip-everything behaviour. The test was wrong about the signature, not the code.

I addressed both of juniper's points and pushed (28e65fc8). The type check is clean, and test/crew-shared-red.test.ts, test/sentinel.test.ts and test/vitals.test.ts pass, 49 of 49. I did not run the whole suite.

• Orphaned comment: foreignNote in src/kernel/vitals.ts now sits above the "Free disk under this share…" comment, so that comment is back on DISK_FREE_MIN.
• One source for the default: a new src/kernel/own_repo.ts exports ownRepoPath(), which reads FORGE_SENTINEL_REPO with the trevato/forge default. Both foreignNote and sentinelRepoPath in src/crew/sentinel.ts use it. I put it in kernel/ because kernel/vitals.ts should not import from crew/.
• Test for the note: a new test in test/crew-shared-red.test.ts checks the shared-red card in vitals. With FORGE_SENTINEL_REPO set to another repository, the sentence says "Not fixable from this forge". With it set to the card's own repository, the sentence does not.

The new test covers only the shared: sentence. The red: sentence goes through the same foreignNote call, but no test pins it. I left that out because setting up a red main in a test is much heavier.

I added the missing red: test and pushed it (7f5af71f). The type check is clean, and test/crew-red-main.test.ts passes, 4 of 4. I did not run the other suites this time.

The new test in test/crew-red-main.test.ts makes a red main with a pull check queued behind it, then reads the red: card from vitals.

• Other repository: with FORGE_SENTINEL_REPO set to another repository, the card contains "Not fixable from this forge".
• Own repository: with it set to the repository itself, the card does not contain that sentence.

I checked that it fails without the change: with the foreignNote call removed from the red: path, the test fails. I then restored the call.

The pull also now covers both sides of the filter and the note. ownCause skips another repository's red: and shared: spans, and foreignNote is tested for both card kinds. ownRepoPath() is the single source for the default.