## Thinking Path
> - Paperclip is the control plane operators use to manage agent
execution environments, including plugin-declared sandbox providers.
> - The failing user path here was `Test draft` for an unsaved sandbox
environment using a schema field marked `format: "secret-ref"`.
> - Saved environments already resolve secret refs before provider use,
but the unsaved probe path was forwarding the selected secret UUID
directly to the provider, which made Novita draft probes fail.
> - Fixing that safely required a probe-only secret resolution path with
explicit actor authorization and audit context, because an unsaved draft
has no persisted environment binding to authorize against.
> - Once that was fixed, CI and review surfaced follow-up hardening
work: preserve actor source through the draft-probe path, prevent late
heartbeat finalization from overwriting already-terminal runs, avoid
duplicate successful-run handoff wakes for comment-driven runs, make SSH
git ref updates tolerate concurrent managed-runtime restores, and keep
the skills catalog build from failing on transient GitHub errors for
pinned references.
> - The result is that Novita draft probes now behave like saved
environments, the new secret access path is constrained and audited, and
the PR is green end-to-end with Greptile at 5/5.
## Linked Issues or Issue Description
No matching public GitHub issue was found after searching open and
closed Paperclip issues for `novita`. Related PR search found
[#8255](https://github.com/paperclipai/paperclip/pull/8255), but it
addresses Novita/dev-SDK linking rather than this draft probe bug.
Bug summary:
- What happened?
When a board user configured a sandbox environment backed by a
schema-driven plugin provider such as Novita, selecting an existing
company secret for `apiKey` and clicking `Test draft` failed because the
probe received the secret UUID instead of the resolved secret value.
- Expected behavior
`Test draft` should resolve secret-ref fields before calling the
provider probe, just like the saved runtime path does.
- Steps to reproduce
1. Open `Company Settings -> Environments`.
2. Create or edit a `Sandbox` environment using a provider with a
`format: "secret-ref"` field such as `Novita Agent Sandbox`.
3. Select an existing company secret for `apiKey`.
4. Click `Test draft`.
5. Observe the probe failure before this patch.
- Paperclip version or commit
Reproduced on a local `master` dev checkout; fixed and verified on
branch commit `ed982d0c0`.
- Deployment mode
Local dev (`pnpm dev`).
- Installation method
Built from source (`pnpm dev` / `pnpm build`).
- Agent adapter(s) involved
Not adapter-specific in the core bug path; affects schema-driven sandbox
provider plugins such as Novita.
- Database mode
Not database-related.
- Access context
Board (human operator).
- Node.js version
`v25.6.1`.
- Operating system
`macOS 15.7.4`.
- Relevant logs or output
The user-visible failure was `Novita sandbox probe failed` during `Test
draft`.
## What Changed
- Resolved schema-marked secret-ref fields during unsaved sandbox
environment probes by adding a dedicated probe-time secret resolution
path in `environment-config.ts`.
- Passed `companyId` plus the full authenticated actor context into the
draft probe normalization route so secret resolution stays
company-scoped, authorized, and auditable.
- Hardened ephemeral secret resolution so unsaved probes require
`secrets:read`, preserve the original actor source (`local_implicit`,
`agent_jwt`, etc.), and emit usable audit metadata.
- Added a conditional heartbeat run-status update so late adapter
completions cannot overwrite runs that were already cancelled or
otherwise terminal.
- Skipped successful-run handoff synthesis for comment-driven wakes,
which removes the extra wake/run that was breaking
`heartbeat-comment-wake-batching`.
- Retried managed-runtime SSH git ref updates on concurrent ref-lock
races instead of failing the restore path.
- Reused the previous skills-catalog manifest entry when a pinned GitHub
reference fails with a recoverable transient error during CI catalog
generation.
- Added focused regression coverage for the draft probe, ephemeral
secret access, heartbeat handoff behavior, SSH ref-lock races, and
catalog fallback behavior.
## Verification
- `pnpm vitest run server/src/__tests__/environment-routes.test.ts`
- `pnpm vitest run server/src/__tests__/secrets-service.test.ts`
- `pnpm vitest run
server/src/__tests__/heartbeat-comment-wake-batching.test.ts`
- `pnpm vitest run
server/src/services/recovery/successful-run-handoff.test.ts`
- `pnpm vitest run
server/src/__tests__/openclaw-gateway-adapter.test.ts`
- `pnpm exec vitest run packages/adapter-utils/src/ssh-fixture.test.ts
-t "merges concurrent remote commits through the managed runtime restore
path"`
- `pnpm exec vitest run
packages/skills-catalog/src/catalog-builder.test.ts`
- `pnpm --filter @paperclipai/server typecheck`
- `pnpm --filter @paperclipai/skills-catalog build`
- `pnpm --filter @paperclipai/adapter-utils build`
- `gh pr checks 8256`
- Manual/live validation: the same fix was cherry-picked into the
running local dev checkout and the user re-tested the Novita `Test
draft` flow successfully after the server restart.
## Risks
- Low risk: the Novita-specific user-facing fix is isolated to unsaved
sandbox draft probes for plugin schema fields marked `format:
"secret-ref"`.
- The new ephemeral secret resolution path is intentionally stricter
than the original broken behavior; regressions would most likely show up
as denied draft probes rather than accidental secret exposure.
- The heartbeat, SSH, and catalog changes are all defensive; if they
regress, they should affect test/CI orchestration paths rather than
persisted company data.
## Model Used
- OpenAI Codex Local (`codex_local` in Paperclip). The runtime does not
expose the exact backend model ID in agent metadata. GPT-5-class coding
model with shell/tool use, repository editing, test execution, GitHub
review handling, and issue-thread coordination.
## Checklist
- [x] I have included a thinking path that traces from project context
to this change
- [x] I have specified the model used (with version and capability
details)
- [x] I have checked ROADMAP.md and confirmed this PR does not duplicate
planned core work
- [x] I have searched GitHub for duplicate or related PRs and linked
them above
- [x] I have either (a) linked existing issues with `Fixes: #` / `Closes
#` / `Refs #` OR (b) described the issue in-PR following the relevant
issue template
- [x] I have run tests locally and they pass
- [x] I have added or updated tests where applicable
- [ ] If this change affects the UI, I have included before/after
screenshots
- [ ] I have updated relevant documentation to reflect my changes
- [x] I have considered and documented any risks above
- [x] All Paperclip CI gates are green
- [x] Greptile is 5/5 with no open P2s, recommendations, or follow-ups
- [x] I will address all Greptile and reviewer comments before
requesting merge
---------
Co-authored-by: Paperclip <noreply@paperclip.ing>
Devin Foley
04173b341d