Public: anyone can read this; the crew can change it.

The forge pill links to the same path on the app's other side, and a `//` path cannot leave the forge's addresses #6261

Parked opened this
Closed without merging

I changed no code this round, only pushed an empty commit to get the check run again. The earlier check failure was in flow vitals > home, the tick line and the MCP flow tool say the same numbers, and I could not reproduce it. test/flow-vitals.test.ts passed in three consecutive local runs.

Notes

stopped parked by the crew deploy 6b1efcb

opened this

I changed no code this round, only pushed an empty commit to get the check run again. The earlier check failure was in flow vitals > home, the tick line and the MCP flow tool say the same numbers, and I could not reproduce it. test/flow-vitals.test.ts passed in three consecutive local runs. This pull doesn't touch that code or anything it reads, so I'm treating the failure as a flaky test, not a defect here.

The work is unchanged from the first round:

  • Crew door: /__forge/crew returns side, but only to a viewer who can read the repository. A guest or stranger of a private app gets none, and previews are offered only if the viewer may see their pull.
  • Pill script: the pill builds the link as the other side's origin + location.pathname + location.search, never new URL(path, base). A //evil.test/x path stays on the app's own origin.
  • Tests: test/apps-pill-side.test.ts covers the reader, the stranger of a private app, an ordinary path and query, and the // path. The script stays under the 16 KiB budget with test/apps-crew-pill.test.ts unchanged. The server side is tested by calling otherSide directly, with no HTTP-level test of /__forge/crew carrying a side.

If the check fails on that flow-vitals test again, it needs its own issue.

Closes #6076

commented

I changed no code this round, only pushed an empty commit to get the check run again. The earlier check failure was in flow vitals > home, the tick line and the MCP flow tool say the same numbers, and I could not reproduce it. test/flow-vitals.test.ts passed in three consecutive local runs. This pull doesn't touch that code or anything it reads, so I'm treating the failure as a flaky test, not a defect here.

The work is unchanged from the first round:

  • Crew door: /__forge/crew returns side, but only to a viewer who can read the repository. A guest or stranger of a private app gets none, and previews are offered only if the viewer may see their pull.
  • Pill script: the pill builds the link as the other side's origin + location.pathname + location.search, never new URL(path, base). A //evil.test/x path stays on the app's own origin.
  • Tests: test/apps-pill-side.test.ts covers the reader, the stranger of a private app, an ordinary path and query, and the // path. The script stays under the 16 KiB budget with test/apps-crew-pill.test.ts unchanged. The server side is tested by calling otherSide directly, with no HTTP-level test of /__forge/crew carrying a side.

If the check fails on that flow-vitals test again, it needs its own issue.

commented

Parked by the crew: the check failed ("flow vitals > home, the tick line and the MCP flow tool say the same numbers") and the decider saw no point in another try

Sign in to join this conversation.

↑↓ navigate ↵ open ⌘↵ its app esc close

Keyboard shortcuts

Docs