[devin-ai-integration]

Part of DEV-1673: application Git identity for ordinary execution and simulations. This reports the customer agent/application revision, not the SDK revision, and does not choose simulation candidates.

Changes

• Add optional WithAppGit(AppGitOptions) and WithAppGitDisabled; infer SHA from the main application binary or reliable configured/runtime sources. Explicit selected source directories are exclusive; never fall back to the observer's checkout if a selected source is unavailable.
• Populate existing raindrop.app.commit_sha, raindrop.app.commit_dirty, and raindrop.app.branch meanings without overwriting caller properties or SDK/release identity. New automatic branch collection is opt-in.
• Freeze operation identity across partials, flushes, retries, and associated spans; retain explicit null/empty overrides. Bound discovery and child-process output/cleanup; unavailable or pending optional metadata is omitted without blocking hot calls.
• Add public driver config, maintained regression tests, README guidance, and the approved non-release CI harness pin.

E2E Proof

Current head 9d123a842f3046efd7d65ae1609ba5ab93883fee adds only the normal SDK version bump from 0.1.5 to 0.1.6 to the previously qualified 49920bf29cd54419055789ed1e87e2bba05b4cb9. The version-only delta passed go test ./... again; no Git collection or transport code changed. Product candidate e3319159138041706da55cf3fc5d08ba7410213f and its later maintained-test/conformance updates received the independent fresh-context verification below:

go test ./...       PASS
go test -race ./... PASS
go vet ./...        PASS
go build ./...      PASS
scratch public consumer: 8 top-level tests PASS, including 3 fake-Git cases
harness f5d8 fault corpus: 19 pass, 34 expected_fail, 12 unsupported, 4 not_applicable
final ten-case application Git matrix: 9 pass, 1 expected_fail; complete gate PASS
real compiled AUT build-info test: 3 repeats PASS

The scratch consumer exercised real loopback event/trace traffic, main-binary VCS build info, two conflicting repositories, inherited Git selector isolation, unavailable selected source, malformed/slow/noisy/incomplete Git, checkout races, operation overrides/absence, branch opt-in and cross-client separation. CI YAML/expression validation passed. Only the conformance driver is compiled with -buildvcs=false to isolate deployment fixtures; the separate real-compiled-app regression proves actual AUT build info still wins over ambient provider/CI metadata.

Published dedicated-test-org evidence is historical, at dcee3de68fcfec9c145ab682fbc13a33a5b5c0d6 / harness f5d8: both Git scenarios PASS, verifying exact default SHA and explicit SHA plus branch through ingestion and Query API. CI run and artifacts. Its prod report is 11 pass, 2 unsupported, 2 not_applicable, no failures; its fault report is 19 pass, 34 expected-fail, 12 unsupported, 4 not-applicable. An earlier a345 run had a non-Git signal-readback failure; the dcee3de report passes that case, without any transport implementation change or assertion weakening.

The final local ten-case matrix used the same Go runtime/driver with harness d43a55edb8a66ef826f650ebff75e3659c0bff9d: 9 pass / 1 expected-fail, with only the existing DEV-1149 plain-track routing gap expected to fail. Final harness caa57a1661f3176cd59957ef0576628d96530214 changes only the runner test's scenario count and fixture expectation from d43a55e; runtime code and scenarios are identical, and its final harness qualification passed 14/14 checker and 89/89 runner tests. The Go commits after the evaluated driver are CI pin changes only.

Pre-version-bump dedicated-test-org qualification at 49920bf29cd54419055789ed1e87e2bba05b4cb9 / final harness caa57: run34924946442. Fault artifact:20pass/34expected-fail/12unsupported/4not-applicable; Git9pass/1declaredDEV-1149 with the complete gate passing. Prod artifact:11pass/2unsupported/2not-applicable, zero failures. Both Git Query readbacks pass with observed records. Fresh CI is running for the version-only head; no tag or package release has been published yet.

Safety Verification / Risk Assessment

• Additive public API; no intentional routing, auth, billing, event-count, application-result/error, or release-field changes.
• No optional discovery on hot calls. Known SHA is retained if dirty state is unknown; a bad Git process cannot fabricate clean state or crash the agent.
• Existing transport gaps remain visible. The stricter new plain-track Git scenario exposes the same pre-existing events/track route gap (DEV-1149) and has one exact new-case ratchet. Its endpoint assertion is not weakened; separate public-path capture still verifies raw null/empty Git values on the SDK's existing partial route.
• Local independent verification and dedicated-test-org Git readback passed as detailed above. No customer credentials/data used, and no packages published or production changes made.
• Rollback is a code revert. Confidence high for tested Go behavior, limited to the evidence above.

Dependencies and rollout

CI pins the immutable DEV-1673 harness candidate caa57a1661f3176cd59957ef0576628d96530214: https://github.com/invisible-tools/raindrop-sdk-harness/pull/98. Backend normalization prerequisite: https://github.com/invisible-tools/dawn/pull/7081. Regular event-property collection needs no database migration. Pavel has authorized normal merging/publication in dependency order after required reviews and checks; the intended Go release is v0.1.6, whose tag must match version.go. No new credentials or release-workflow edits are required. The review sweep found no outstanding Go threads and confirmed no-destination clients already skip optional Git work; that finding is being propagated to other SDKs.

deliver v0.2.2 · independent Go evaluation PASS · Devin PR publisher.

[devin-ai-integration]

🤖 Devin AI Engineer

I'll be helping with this pull request! Here's what you should know:

✅ I will automatically:

• Address comments on this PR that start with 'DevinAI' or '@devin'.
• Look at CI failures and help fix them

Note: I can only respond to comments from users who have write access to this repository.

⚙️ Control Options:

• Disable automatic comment, CI, and merge conflict monitoring

[pavel-y-ivanov]

Final review: unchanged candidate 9d123a8 passed the maintained Git fault gate and exact-candidate production readback rerun 35013417833 attempt 2. Actual report: 11 pass, 2 unsupported, 2 not-applicable, zero failures; both Git records observed without no-show. No unresolved substantive review threads. Approved under Pavel explicit delegation.