feat(container): publish findings service and distribute workflow runner

[kmbroai]

Summary

Distribute the findings service and a scanner-based workflow runner through the existing two-image container release process. The runner uses the scanner image and shared Dockerfile; the findings API remains a separately deployed service.

Stacked on #660. This update merges its consolidated SDK, workflow, checkpoint, dashboard, and canonical plugin changes into the container branch. The diff against #660 remains limited to container distribution, documentation, and integration tests. Neither PR is merged into its base.

Changes

• Share the release pipeline across scanner and findings images, preserving version checks, native architecture builds, anonymous pulls, SBOMs, provenance, and immutable stable tags.
• Findings consumer Compose pulls ghcr.io/openai/codex-security-findings; the existing image setting selects a version, digest, or local build.
• Add compose.runner.yaml using the scanner image, durable state/artifact mounts, and existing security and authentication settings. Document connectivity to a separate findings service, persistent source paths, AppArmor, and lifecycle management.
• Exercise CLI deduplication in disposable scanner containers across a Compose network. Verify workflow results and review checkpoints in the runner database, findings and groups in the service database, dashboard assets, and persistence across replacements. The service smoke uses the packaged default command.
• Read host fixtures from plugins/codex-security; generate the installed plugin payload during packaging. Keep the TypeScript SQLite/runner fixtures and existing workbench database initialization.
• No third image, duplicated workflow engine, new public CLI/API syntax, sandbox-policy change, or workflow/checkpoint implementation is added by this PR.

Testing

Local verification at a8be37a3:

• Focused integration tests: 153 passed, 12 skipped, zero failures.
• Full SDK suites passed with fixed seed 12345 and randomized seed 3716402526: each had 2,031 passed, 29 skipped, and zero failures.
• Canonical plugin generation/source ownership, generated-model consistency, TypeScript checks including MCP, formatting, and git diff --check passed.
• Docker smoke passed from a tracked source snapshot with no pre-generated host plugin: separate scanner runner, packaged service default command, independent SQLite state, dashboard assets and API, scoped candidates, screening/pair reviews, checkpoint reuse, missing-scan exit behavior, restart persistence, and clean shutdown.
• The Docker-built npm archive passed package and installed-consumer checks: 381 entries, including 121 plugin files. Verified public imports, NodeNext types, CLI, credential locking, MCP initialization, bundled Codex, dashboard assets, and nested-worker startup without global Codex. Test fixtures are excluded.
• Scoped lint passed for all three container workflows.
• Hosted CI is tracked by the checks attached to the current PR commit; these results describe local verification.

Risk and rollout

• Custom publication, group write-back, workflow resumption, review checkpoints, and the dashboard come from feat(typescript): persist findings and embeddings through the API #660. The SDK implementation, canonical plugin, SQL migrations, and unit tests match that base; durable mounts alone do not add those capabilities.
• GHCR publication requires public packages, repository Actions write access, and the existing container approval gate. This update does not publish images or merge PRs. Local testing covers Linux amd64; native arm64 release validation remains in CI.
• The service has no API authentication and its published host port remains loopback-only. Connect the runner over a private network or appropriately protected endpoint. Runner and service databases remain separate.
• Preserve and back up persistent state before migrations. Stable image versions cannot be overwritten. Restricted hosts require the documented AppArmor configuration for native sandbox execution.
• Tests use synthetic findings and provider responses; no live model reviews were run.

Public disclosure review

• No customer, partner, prospect, or user identities, data, or identifying details are included.
• No credentials, personal data, private source, scan findings, or nonpublic links or tickets are included.
• I reviewed the branch name, title, description, commits, changes, comments, logs, screenshots, attachments, and links for public disclosure.

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
🔒 Security Review	✅ Completed 2026-08-27T22:45:40.765080Z	a8be37a	New commits

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.