Repository navigation
Conversation
Shared Workflows CIShared workflows CI passed. View run output |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Replace the owner-rejected custom integration with the explicitly selected official OpenCode V1 GitHub action.
anomalyco/opencode/github, pinned to the verifiedv1.19.4commit.planagent selection.Owner explicitly accepted the limited V1 scope after evaluation. No linked issue.
Type of change
Validation
Current head:
9717f7ad44794abb28f99f3f95541ad2b3a29e6d. Lint/actionlint passed on the preceding cleanup head in run 40419511947. The former 17 passing mock tests covered the rejected implementation and are not evidence for this replacement. No local tests, inference, or workflow dispatch were run. No bespoke replacement harness was added.Independent completed reviews: adversarial ADVANCE, infrastructure security ADVANCE for disabled authoring only, DevOps ADVANCE after inspecting the complete replacement contract, QA GAPPED with documented runtime gaps and approval of obsolete-test removal. No unresolved Critical/High disabled-draft finding or waiver.
Runtime-contract validation found that the upstream GitHub handler ignores the action
agentinput. The workflow now uses the documented native inlinedefault_agent: plansetting instead. Checkout credentials are retained, matching upstream token-mode requirements for authenticated private-repository fetches. These corrections were independently re-reviewed (adversarial/security/DevOps ADVANCE for disabled authoring only). No end-to-end provider/comment run has occurred; a caller, approved model/key, scoped enablement, and explicit pilot authorization remain required.Static lint does not prove event gates, runtime dependencies, review quality, comment behavior, or credential isolation. A separately authorized security pilot remains mandatory before enabling.
Impact and rollout
shared-workflows/ Make IT Work Cloud maintainers own the wrapper;tfroot-githubowns future generated.github/workflows/opencode-review.ymlcallers, which must select a reviewed immutable shared commit. No caller, model choice, credential provisioning, repository setting, apply, adoption, or activation is delivered here. Existing ARC/OpenTofu workflows and lint remain unchanged. No chart release, image publication, or cluster rollout.Accepted upstream limits: new comment per run rather than sticky; actor permission checks can reject bots/Dependabot; forks excluded; no exact-SHA stale-publication guard, chart reviewer reuse, completion parser, guaranteed failure comment, or output sanitizer. Cancellation/re-execution can leave partial or duplicate comments. Green/skipped checks are not review-completion gates.
The action source is pinned, but its installer downloads the latest CLI and its nested cache uses a tag. Repository configuration/plugins and the composite installer can encounter provider credentials. The prompt and
planagent are not a security sandbox. Security review advances only the disabled draft; trusted execution and credential boundaries must be established in an approved pilot before activation. No extra scripts are proposed to emulate missing upstream features.Stages: authored and statically reviewed; current cleanup-head lint/actionlint pending; downstream caller selection/adoption remains unperformed; merge, apply, credential provisioning, pilot, and explicit enablement remain separately confirmation-gated. Rollback is leaving both enablement gates off; no runtime adoption occurs from this PR.
Safety and secrets
AI-assisted implementation and independent AI review. No merge or activation is authorized.
Authorized pilot result
The owner separately authorized two PR-49-only attempts with the existing OPENAI_API_KEY secret and openai/gpt-6.1-sol, reporting the key/gate ready and then the gate corrected. Both attempts finished skipped before the review job ran any steps:
0128fd1fbc0f2e73001d057fe23a61b58d563806.3982bad03c90ff916042050fc0f76b87d492abd9.No provider call, authenticated runtime Git fetch, selected-agent execution, or review comment is verified. Inference: the remaining repository-variable enablement condition is false/unavailable; variable metadata cannot be read through the available GitHub MCP tools, so that root cause is not independently confirmed. No gate bypass or automatic retry occurred.
The temporary caller was removed after each terminal attempt; final cleanup is
10b605ec1b08cad216bf8fd88555b43322bf732e. The final net diff is again only the 50-line reusable workflow, README row, and guide. Both default-off controls remain; there is no installed caller. Further credentialed execution needs a verified gate and fresh owner authorization. This implementation is statically CI-validated, not end-to-end validated.Third separately authorized pilot
After the owner explicitly confirmed the gate verified and authorized one more attempt, run 40420031353 at
3f0083dbff54490d2a25a99581ebe90578c3daefalso completed skipped, with no steps inpilot / review. The temporary caller was removed at9717f7ad44794abb28f99f3f95541ad2b3a29e6d; the source tree matches the preceding cleanup tree. No inference or review comment was verified.The gate-variable explanation above remains only a hypothesis, not a confirmed root cause; the owner reports it configured. Repeating the identical credentialed pilot is not justified. A credential-free diagnostic of the evaluated conditions is the next proposed investigation, not yet implemented or authorized here. No gate bypass, automatic retry, merge, or distributed adoption occurred.