Skip to content

Troubleshooting

  • Confirm the source is supported and the session file is complete.
  • Claude Code and Codex use fixed local discovery roots.
  • Letta Trajectory files require an explicit --trajectory path.
  • Remove overly narrow --since, --project, or --source filters.
  • Use the real project directory; project matching uses the recorded working directory.

A refusal is a safety result, not an upload failure. Inspect the safe reason. Common causes include a residual secret after redaction, privacy-filter canary failure, malformed source data, or an oversized envelope. Do not bypass the guard; remove the sensitive source or exclude the session.

  • Check system time; short-lived signatures depend on it.
  • For InviteMalformed, copy the complete invite link, including the #CODE fragment.
  • InviteNotValid intentionally covers an unknown, expired, or revoked invite; ask the operator to check it rather than guessing.
  • For InviteAlreadyConsumed, reuse the saved enrollment or request a new invite. Do not keep retrying with newly generated device keys.
  • Treat InviteRegistryNotConfigured and InviteRegistryStale as service-side failures and contact the operator.
  • Confirm issuer and ingest hosts match the persisted allowlist.
  • Verify the device key still matches the registered device_key_id.
  • Do not reuse an upload claim after expiry or against another audience.

See Invites and enrollment for what redemption changes and what remains local.

  • Start or restart trace-commons-contributor daemon using the platform’s supported service path.
  • Verify the contributor state directory belongs to the current user and has restrictive permissions.
  • On Windows, confirm the named-pipe ACL check succeeded for the installed build.
  • Do not weaken socket/pipe permissions as a workaround.
Terminal window
ironclaw traces status --json
ironclaw traces queue-status --json
ironclaw traces ingest-health \
--endpoint https://ingest.tracecommons.ai

Look for missing policy, expired credentials, retry holds, offline/DNS/timeout classes, privacy holds, or malformed quarantined queue files. The queue worker continues past an isolated malformed entry; repair the cause instead of deleting the whole queue.

The server is currently moving through a coordinated merge/deploy. Compare the deployment version with the source marker on the relevant page. If they differ, prefer the deployed OpenAPI/schema and current protocol crate, then open a docs issue with both commit identifiers.

Include the page URL, client and server versions, command or endpoint name, and a sanitized error code. Never attach a bearer token, device private key, raw trace, or unredacted server body.