fix: pin supervisor image to match openshell CLI version

[waynesun09]

Summary

• Pin the OpenShell supervisor image to ghcr.io/nvidia/openshell/supervisor:${OPENSHELL_VERSION} via gateway.toml so the in-container supervisor always matches the installed gateway/CLI version.
• Applies to both action.yml (production agent runs) and functional-tests.yml (e2e tests).

Root Cause

OpenShell v0.0.73 was released at 15:31 UTC on June 30, re-pointing the supervisor:latest tag to a build containing NVIDIA/OpenShell#2001 (fix(supervisor): drop sandbox child capability bounding set). The cap_drop_bound() call fails with EINVAL in rootless Podman on GitHub Actions runners because CAP_SETPCAP is unavailable in user namespaces. The supervisor crashes immediately:

WARN openshell_supervisor_process::netns: Failed to delete network namespace
Error: × Invalid argument (os error 22)

The gateway defaults to supervisor:latest (not pinned to its own version), so every new runner after 15:31 UTC pulls the broken v0.0.73 supervisor regardless of what CLI version is installed. This is why reverting the CLI to 0.0.63 (PR #2787) did not fix sandbox creation.

Fix

Write $HOME/.config/openshell/gateway.toml in the gateway configuration step, setting supervisor_image to the version-tagged image matching OPENSHELL_VERSION from openshell-version.sh. This eliminates the dependency on the :latest tag and ensures version lock between CLI ↔ supervisor.

Test plan

• Verify sandbox creation succeeds in a fullsend-ai/.fullsend agent run after merge
• Verify functional-tests.yml e2e tests pass with the pinned supervisor

Fixes #2792

[qodo-code-review]

PR Summary by Qodo

Pin OpenShell supervisor image to installed CLI version via gateway.toml

🐞 Bug fix ⚙️ Configuration changes 🕐 10-20 Minutes

AI Description

• Pin OpenShell supervisor_image to ${OPENSHELL_VERSION} to avoid :latest regressions.
• Write ~/.config/openshell/gateway.toml during gateway configuration for prod and e2e.
• Ensure GitHub Actions runs use a supervisor matching the installed gateway/CLI.

Diagram

graph TD
  A["GitHub Action / Workflow"] --> B["Configure gateway step"] --> C["source openshell-version.sh"] --> D["Write gateway.toml"] --> E["OpenShell gateway"] --> F{{"Pull supervisor image"}} --> G["ghcr.io/nvidia/openshell/supervisor:${OPENSHELL_VERSION}"]

Loading

High-Level Assessment

The following are alternative approaches to this PR:

1. Pin supervisor by immutable digest

• ➕ Fully eliminates tag-mutation risk (even for version tags)
• ➕ Reproducible sandbox behavior across time and runners
• ➖ Requires a reliable source-of-truth mapping version→digest
• ➖ More operational overhead when releasing/rolling versions

2. Upgrade (or patch) to a fixed supervisor release

• ➕ Removes the root incompatibility for rootless Podman
• ➕ Avoids accumulating version pin logic across clients
• ➖ Depends on upstream release timing and adoption
• ➖ Does not protect against future :latest regressions if defaults remain unpinned

3. Set supervisor image via an action input/env var only

• ➕ Avoids writing persistent config files under $HOME
• ➕ Keeps configuration localized to the action interface
• ➖ Requires gateway to support/consume that mechanism consistently
• ➖ Harder to reuse across both action.yml and workflow consumers without duplication

Recommendation: The current approach (writing gateway.toml with a version-tagged supervisor_image) is the best immediate mitigation: it restores determinism, aligns CLI↔supervisor versions, and avoids relying on :latest. If the project wants maximum reproducibility, consider a follow-up to pin by digest or add an official action input that maps versions to digests.

Files changed (2) +16 / -0

Bug fix (2) +16 / -0

functional-tests.ymlPin supervisor image during e2e gateway configuration +8/-0

Pin supervisor image during e2e gateway configuration

• Sources 'openshell-version.sh' and writes '~/.config/openshell/gateway.toml' to set 'openshell.gateway.supervisor_image' to the version-tagged supervisor image. This makes functional tests immune to upstream 'supervisor:latest' retags and keeps supervisor aligned with the installed OpenShell version.

.github/workflows/functional-tests.yml

action.ymlPin supervisor image for production agent runs +8/-0

Pin supervisor image for production agent runs

• Sources the action’s 'openshell-version.sh' and writes '~/.config/openshell/gateway.toml' with 'supervisor_image = ghcr.io/nvidia/openshell/supervisor:${OPENSHELL_VERSION}'. This ensures the in-container supervisor matches the gateway/CLI version and avoids breakage from ':latest' changes.

action.yml

[github-actions]

Site preview

Preview: https://4f8247bf-site.fullsend-ai.workers.dev

Commit: facd02e71bcd31b052cb83725a977f742529e98a

[fullsend-ai-review]

🤖 Review · ❌ Terminated · Started 6:42 PM UTC · Ended 6:58 PM UTC
Commit: b95b879 · View workflow run →

[qodo-code-review]

Code Review by Qodo

🐞 Bugs (0) 📘 Rule violations (0) 📎 Requirement gaps (0)

Great, no issues found!

Qodo reviewed your code and found no material issues that require review

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

[fullsend-ai-review]

🤖 Finished Review · ❌ Failure · Started 6:42 PM UTC · Completed 6:58 PM UTC
Commit: facd02e · View workflow run →

[fullsend-ai-retro]

🤖 Finished Retro · ❌ Failure · Started 7:13 PM UTC · Completed 7:20 PM UTC
Commit: facd02e · View workflow run →