Conversation
Matches what the dashboard repository now has, so a ticket in either place starts the same way. One workflow turns an `agent`-labelled issue into a pull request; the other reviews every pull request. Neither may merge -- required checks decide that, which is the only reason letting a model write code here is reasonable. Both prompts point at this repository's own rules and at the two failures that hurt most from here: a version change that leaves compatibility.json behind, which ships a mismatched engine to the dashboard and the desktop application, and a protocol or published-surface change nothing downstream was updated for. Dependabot is grouped per ecosystem, with torch on its own because it decides what the engine can run and how large every artifact is. The workflow files parse. Neither has run: both need secrets that are not set on this repository yet.
Every job here sets up Python with `cache: pip`, and on this fleet that step fails outright: "No file ... matched to [**/requirements.txt or **/pyproject.toml], make sure you have checked out the target repository". All seven pyproject.toml files are present in `packages/` at that exact path -- checked on the runner's own workspace while a job sat failed. The runners keep `_work` on another drive behind a symlink, and the action's dependency-file glob does not see through it. It has taken out `blocked-deps`, `client`, `docs` and all five `extras` jobs, and it took out the dashboard's release workflow earlier today for the same reason. It is not reliably reproducible -- an earlier run on main globbed fine -- which makes it worse, not better: a pipeline that merges on green will see spurious red it cannot distinguish from a real failure. Removing it costs nothing. A self-hosted runner keeps `~/.cache/pip` between jobs by itself, so the action was caching a cache.
The two workflows called anthropics/claude-code-action and wanted an ANTHROPIC_API_KEY. This repository is not going to have one, so both now speak plain chat-completions: the review posts through scripts/llm_review.mjs, and the issue worker drives aider with OPENAI_API_BASE pointed wherever LLM_BASE_URL says. OpenRouter and DeepInfra both serve that protocol, so the provider is two variables. LLM_MODEL has no default on purpose. A guessed slug fails at the provider with an opaque error on somebody else's pull request, which is worse than refusing to start. The issue body reaches the agent through a file rather than a shell argument. It is arbitrary text from whoever opened the issue, and interpolating it into a command would let its contents decide what runs. Also raises the subprocess timeout in test_rebuild_is_bit_exact_in_a_ subprocess from 180s to 600s. It timed out today on a runner sharing this workstation with two other jobs, and reported it as a bit-exactness failure that had not happened -- red that looks exactly like the defect the test is named after. Run here: the 9 llm_review tests, and both workflow files parse. The workflows themselves have not run; they need secrets that are not set.
This branch has not been deployed
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
1 participant
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.
Matches what
spikeforge-dashboardnow has, so a ticket in either placestarts the same way.
agent-labelled issue into a pull request.Neither may merge. Required checks decide that, which is the only reason
letting a model write code here is reasonable.
Both prompts point at this repository's own rules and at the two failures
that hurt most from here: a version change that leaves
compatibility.jsonbehind — which ships a mismatched engine to the dashboard and the desktop
application — and a protocol or published-surface change that nothing
downstream was updated for.
Dependabot is grouped per ecosystem, with torch on its own because it decides
what the engine can run and how large every artifact is.
Secrets this needs
ANTHROPIC_API_KEY— both workflows. The review workflow skips itselfwithout it rather than failing every pull request.
AUTOMATION_TOKEN— the issue workflow. A pull request opened with thebuilt-in token never gets checks, so it could never auto-merge.
What was run
The workflow files parse. Neither has run: both need secrets that are not set
on this repository yet.