Repository navigation
Merge dev into main: structured review session tags and no retry on usage limits - #143
Merged
Merged
Conversation
`@octokit/graphql` resolves `response.data.data` and only throws when the body carries an `errors` array, so an HTTP 200 whose body is not a GraphQL envelope resolves `undefined`. `fetchPRBranchData` dereferenced that value directly, turning a transient GitHub answer into `TypeError: undefined is not an object (evaluating 'prResult.repository')` that killed the `Prepare validator` step of Droid Auto Review and skipped the review pass that posts the comments. Route both fetchers in this module through one helper that checks the response shape before reading it, retries the request via `utils/retry.ts` per the repo convention for GitHub API calls, and reports the HTTP status, content type and body preview captured through a request hook, since `@octokit/graphql` otherwise discards the body that would explain the failure. A present-but-null `repository` stays a stable "not found" answer and is not retried. Co-authored-by: factory-droid[bot] <138933559+factory-droid[bot]@users.noreply.github.com>
Add a dedicated hidden marker to PR review tracking comments and preserve it across sanitized summary updates. Co-authored-by: factory-droid[bot] <138933559+factory-droid[bot]@users.noreply.github.com>
Replace the review-specific marker with a generic PR validation marker carrying an explicit source value for future validation comment types. Co-authored-by: factory-droid[bot] <138933559+factory-droid[bot]@users.noreply.github.com>
Resolve a typed run identity before tracking comments are created, pass it through each mode and MCP server, and derive PR validation metadata independently of comment presentation. Co-authored-by: factory-droid[bot] <138933559+factory-droid[bot]@users.noreply.github.com>
Check automatic security run-once state before creating the tracking comment. Skip comment creation for security-only no-op runs and downgrade dual-review comments when security will not run. Co-authored-by: factory-droid[bot] <138933559+factory-droid[bot]@users.noreply.github.com>
Replace the single-value validation source with exact run-type metadata. Mark code review, security review, and full security scan comments while keeping fill and CI Steward outside the PR validation marker. Co-authored-by: factory-droid[bot] <138933559+factory-droid[bot]@users.noreply.github.com>
Add a dedicated droid-default run type for bare @droid commands and return null when no command or automatic mode identifies a run. Propagate unresolved runs without exporting or marking a false review identity. Co-authored-by: factory-droid[bot] <138933559+factory-droid[bot]@users.noreply.github.com>
Distinguish PR issue and inline comment markers so review consumers can identify both the comment surface and originating run type. Co-authored-by: factory-droid[bot] <138933559+factory-droid[bot]@users.noreply.github.com>
Keep shared code/security classification across sticky reuse, progress updates, and final summaries instead of replacing or losing the marker. Co-authored-by: factory-droid[bot] <138933559+factory-droid[bot]@users.noreply.github.com>
feat(review): tag tracking comments for classification
Co-authored-by: factory-droid[bot] <138933559+factory-droid[bot]@users.noreply.github.com>
…-the/2a92c804-520e-4c49-9184-ce275ab392d9 fix(github): survive GraphQL responses that carry no envelope
Every review `droid exec` carried a bare `--tag code-review`, so the warehouse had to identify review sessions by matching prompt text and pair the two passes by parsing "PR #N in owner/repo" out of the prompt. That breaks whenever the prompt is reworded, never matched GitLab's "MR !N", and truncates repo names at the first dot. The tag now carries metadata: pass (candidates|validator), reviewType (code|security), platform (github|gitlab), repo, pr, runId and runAttempt. The name stays `code-review` so existing filters keep working. - GitHub: `githubReviewSessionTagArg` builds the fragment from the parsed context; quoted with shell-quote so `droid_args` splits it as one token. - GitLab: gitlab-prepare writes the per-pass JSON into the resolved-env shim; the template passes it as its own quoted `--tag` argument with a bare-name fallback for shims from older prepare versions. Co-authored-by: factory-droid[bot] <138933559+factory-droid[bot]@users.noreply.github.com>
runDroid retried every non-zero exit three times with backoff. A 402 from
the Factory API (5-hour/weekly/monthly usage limit, exhausted Extra Usage,
credit limit) or a BYOK provider's empty balance cannot clear in the
seconds between attempts, so each failing review opened three failed
sessions and the workflow reruns multiplied that further. Sep 10-17 review
telemetry showed an average of 4.7 pass-1 sessions per PR-day that ended on
a usage limit versus 1.5 otherwise.
In stream-json mode the CLI reports the fatal error as
{type:"error",source:"agent_loop",message:"402 {...}"}; there is no result
event. Capture that message per attempt, stop after the first attempt when
it is a usage limit, and surface the body's human-readable detail as the
step's error_message so the PR comment says which limit was hit. The same
captured message now also feeds the existing 403 model-policy fallback and
the generic error_message, which previously only looked at result events.
Test hygiene: run-droid-mcp.test.ts installs a process-wide
mock.module("child_process") that leaked into every later test file; it now
restores the real module in afterAll.
Co-authored-by: factory-droid[bot] <138933559+factory-droid[bot]@users.noreply.github.com>
fix(base-action): do not retry droid exec after a usage-limit failure
Co-authored-by: factory-droid[bot] <138933559+factory-droid[bot]@users.noreply.github.com>
feat(review): send structured session tag on every review pass
Contributor
|
Droid finished @factory-nizar's task —— View job Validated two actionable classification bugs: sticky-comment fallback can reuse a combined marker on a new comment, and validator preparation converts an intentionally unclassified custom-trigger run into a code review. 2 inline comments posted |
![factory-droid[bot]](https://avatars.githubusercontent.com/in/358017?s=60&v=4){kind=small}
| repo, | ||
| issue_number: context.entityNumber, | ||
| body: initialBody, | ||
| body: issueCommentBody, |
Contributor
There was a problem hiding this comment.
[P2] Rebuild the marker when sticky-comment fallback creates a new comment
After a sticky comment is found, issueCommentBody is mutated to combine the previous and current run types. If updating that comment then fails, the fallback creates a separate comment from this mutated body, so a comment belonging only to the current run is incorrectly classified as a combined review-and-security comment. Build the fallback body from createInitialBody("issue-comment") instead.
| throw new Error("prepare-validator requires a pull request context"); | ||
| } | ||
| const runType = | ||
| parseDroidRunType(process.env.DROID_EXEC_RUN_TYPE) ?? DroidRunType.Review; |
Contributor
There was a problem hiding this comment.
[P2] Preserve an intentionally unclassified run type
A review triggered through a custom phrase can resolve to null, and the candidate pass intentionally leaves DROID_EXEC_RUN_TYPE unset, but this fallback converts that absence to droid-review for the validator and final posting steps. The same run therefore starts unclassified and ends classified as a code review, contrary to the new null-run-type path; carry an explicit null sentinel between passes instead of defaulting every missing value to Review.
This was referenced Sep 28, 2026
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
2 participants
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.
Promotes
devtomainfor thev8release. Clean merge (git merge-treereports no conflicts; every commit onmainis already indev).Included
Every review pass now sends
--tagwith{name:"code-review", metadata:{pass, reviewType, platform, repo, pr, runId, runAttempt}}so the warehouse can identify and pair review sessions without parsing prompt text. Name stayscode-review.A 402 (usage limit, exhausted Extra Usage, credit limit, BYOK empty balance) stops after the first attempt instead of three, and the PR comment shows which limit was hit.
Tracking and inline comments carry the run type; centralizes droid run types; skips comments for skipped security runs.
After merge
Create the
v8GitHub release onmain(v7is at #127;mainalso carries the unreleased #129, #131, #136, #140 dev merges).