At commit 854f4e6, paseo-omp/server/provider/connection.ts:269 returns only the fallback for every error that is not an OmpPublicError:
function errorDetails(error: unknown, fallback: string): { message: string } {
return { message: isOmpPublicError(error) ? error.message : fallback };
}
The session-open catch at line 800 calls this with OMP session failed to open, without retaining or logging the cause. We observed four failed opens on September 16 plus a fresh failed create after setting the supported per-call params.sessionDir. All surface the same message. Direct replication of the restricted spawn command reaches the RPC ready handshake, so the lost exception is the useful next diagnostic.
Please preserve a value-safe diagnostic at the collapse boundary (error class, sanitized message/stack and request correlation), while keeping the public response generic. Avoid logging environment, credential values or prompt content. A regression test should make an unexpected open exception visible in diagnostic output, without exposing those values.
Related local evidence: https://github.com/Trevor-Mengel/global-dev-governance/issues/377 . No credential data included.
At commit 854f4e6,
paseo-omp/server/provider/connection.ts:269returns only the fallback for every error that is not anOmpPublicError:The session-open catch at line 800 calls this with
OMP session failed to open, without retaining or logging the cause. We observed four failed opens on September 16 plus a fresh failed create after setting the supported per-callparams.sessionDir. All surface the same message. Direct replication of the restricted spawn command reaches the RPC ready handshake, so the lost exception is the useful next diagnostic.Please preserve a value-safe diagnostic at the collapse boundary (error class, sanitized message/stack and request correlation), while keeping the public response generic. Avoid logging environment, credential values or prompt content. A regression test should make an unexpected open exception visible in diagnostic output, without exposing those values.
Related local evidence: https://github.com/Trevor-Mengel/global-dev-governance/issues/377 . No credential data included.