Conversation
|
create-cloudflare
@cloudflare/kv-asset-handler
miniflare
@cloudflare/pages-shared
@cloudflare/unenv-preset
@cloudflare/vite-plugin
@cloudflare/vitest-pool-workers
@cloudflare/workers-editor-shared
wrangler
commit: |
|
Codeowners approval required for this PR:
Show detailed file reviewers
|
|
|
||
| Recommend **CLOSE** (state_reason: `not_planned`) if: | ||
|
|
||
| - The requested change would be a breaking change in the current major version |
There was a problem hiding this comment.
i think this we would likely want to keep open and label as breaking change, rather than close
There was a problem hiding this comment.
I think that should be covered by "The team has explicitly decided not to implement this feature", but I guess we can see how this behaves in practice?
| Write a single tab-separated line to `./data/<issue_number>/summary.md` with these 8 fields: | ||
|
|
||
| ``` | ||
| [<issue_number>](https://github.com/<owner>/<repo>/issues/<issue_number>) <title> <CLOSE|KEEP OPEN|NEEDS MORE INFO|NEEDS VERIFICATION> <easy|medium|hard|n/a> <brief reasoning> <brief suggested action> <Yes|No|N/A> |
There was a problem hiding this comment.
it looks like this is removing the difficulty assessment. i would argue we should still keep it, though looking at this now i am slightly unclear how it has been working up to this point, since i don't see anything in this skill that seems to evaluate how difficult the issue would be to address. am i missing something?
|
/bonk can you fix the formatting on this pr? |
There was a problem hiding this comment.
Good improvements overall. The expanded closure taxonomy with specific templates and the richer component mapping table are both solid upgrades over the previous flat list. A few things I noticed in the current state:
1. Runtime issues template claims the agent "moved" the issue (line ~108)
The template for runtime issues in 2e says:
"...so I've moved the issue there."
But the skill preamble explicitly states it "runs without bash or network access." The agent can't move issues — it can only recommend that a maintainer move it. This template will produce a comment that claims an action the agent didn't (and can't) take. Consider rewording to something like:
"The right place to track this is cloudflare/workerd — could a maintainer transfer this issue there?"
Or, if the intention is that a human posts these comments (not the agent directly), that should be clarified somewhere in the skill.
2. Breaking changes in 2i — close vs keep open
lrapoport-cf raised this and penalosa responded, but I think the concern is still valid for the first bullet: "The requested change would be a breaking change in the current major version." This doesn't mean the team decided not to do it — it means it's deferred. The template itself says "will reconsider ahead of a future major version," which implies the issue should stay open as a tracking item with a breaking change label rather than being closed. Closing it risks losing the signal.
The other three bullets in 2i (intentional/documented, explicitly decided against, cost outweighs benefit) are genuine "won't fix" cases and closing makes sense for those. It might be worth splitting the breaking-change case into its own category that recommends KEEP OPEN + label.
3. Duplicate detection (2d) and changelog checking (2c) assume external knowledge
The skill runs without network access and only has context.json. But 2d asks the agent to identify duplicates ("the same problem as an existing open or recently closed issue") and 2c asks it to match "a known fix in the changelog." The agent has no way to search other issues or read changelogs. In practice it can only catch duplicates if they're mentioned in the issue comments. It might be worth scoping these sections to clarify: "Based on information in the issue comments or linked references..." so the agent doesn't hallucinate duplicate issue numbers.
4. Minor: state_reason guidance was removed from report template
Earlier iterations had explicit **State Reason:** <completed | not_planned | N/A> in the report output template. The latest version removed it (per penalosa's suggestions), which simplifies things. But the triage steps still reference state_reason values (e.g., "Recommend CLOSE (state_reason: completed)"). It's slightly unclear where the agent should record this. If the state_reason in the step descriptions is just guidance for the type of closure (informational), that's fine — but if it's meant to be captured in output, it needs a field.
|
Review posted successfully. Here's a summary of what I flagged:
|
Co-authored-by: devin-ai-integration[bot] <158243242+devin-ai-integration[bot]@users.noreply.github.com> Co-authored-by: lrapoport-cf <107272160+lrapoport-cf@users.noreply.github.com>
Co-authored-by: Somhairle MacLeòid <samuel@macleod.space>
…ing change from won't-fix, clarify state_reason, fix 2e action guidance
5579c42 to
220231e
Compare
|
Codeowners approval required for this PR:
Show detailed file reviewers
|
Uh oh!
There was an error while loading. Please reload this page.