bug-provider-slash — Workgraph live mirror

Metadata

Status	done
Assigned	`agent-2404`
Agent identity	`02e879681e52e0a384106169be043416c4d946e850ab26b2269c57681b52a6e7`
Created	2026-05-04T19:07:17.948306219+00:00
Started	2026-05-04T19:19:46.231998339+00:00
Completed	2026-05-04T19:37:41.335265079+00:00
Tags	`eval-scheduled`
Eval score	0.86
└ blocking impact	0.90
└ completeness	0.88
└ coordination overhead	0.85
└ correctness	0.94
└ downstream usability	0.86
└ efficiency	0.82
└ intent fidelity	0.79
└ style adherence	0.92

Description

cargo test test_create_provider_prefixed_is_openai --test integration_provider currently fails independently of the agency hash migration work. The test config sets native/openai endpoints, calls create_provider(tmp.path(), "deepseek/deepseek-chat"), and expects provider name oai-compat; current code returns openai.

Validation

Reproducer: env -u WG_LLM_PROVIDER -u WG_MODEL cargo test test_create_provider_prefixed_is_openai --test integration_provider fails before the fix
Implementation makes the provider expectation and routing behavior consistent
cargo build + cargo test pass with no regressions

## Description
`cargo test test_create_provider_prefixed_is_openai --test integration_provider` currently fails independently of the agency hash migration work. The test config sets native/openai endpoints, calls `create_provider(tmp.path(), "deepseek/deepseek-chat")`, and expects provider name `oai-compat`; current code returns `openai`.

## Validation
- [ ] Reproducer: `env -u WG_LLM_PROVIDER -u WG_MODEL cargo test test_create_provider_prefixed_is_openai --test integration_provider` fails before the fix
- [ ] Implementation makes the provider expectation and routing behavior consistent
- [ ] cargo build + cargo test pass with no regressions

Depends on

done .assign-bug-provider-slash

Required by

done .flip-bug-provider-slash

Log

2026-05-04T19:19:42.124087731+00:00 Lightweight assignment: agent=Careful Programmer (02e87968), exec_mode=full, context_scope=task, reason=This is a correctness-focused code bug with an integration test regression, so the careful programmer is the best fit for implementing and validating the routing fix.
2026-05-04T19:19:46.232003409+00:00 Spawned by coordinator --executor claude --model opus
2026-05-04T19:19:57.181193361+00:00 Starting investigation of provider routing test failure
2026-05-04T19:25:36.122965656+00:00 Root cause: native_cfg.provider value 'openai' (legacy alias) flows through provider_name resolution unnormalized. The canonical-tag commit ed04790b7 normalized spec_provider via provider_to_native_provider but missed the native_cfg.provider and endpoint_provider_override paths. Fix: normalize both via provider_to_native_provider so 'openai' becomes 'oai-compat' (the canonical internal tag).
2026-05-04T19:35:40.528919334+00:00 Validated: integration_provider tests all pass (24/24). cargo build clean. The only failing test (config::tests::test_load_merged_no_global_file) is pre-existing and unrelated — it's about default model selection (codex vs claude), not provider naming. Verified by stashing my changes and running it on main — same failure.
2026-05-04T19:36:21.125621247+00:00 Committed: 5994bde4e — pushed to origin/wg/agent-2404/bug-provider-slash
2026-05-04T19:37:41.335272383+00:00 Task pending eval (agent reported done; awaiting `.evaluate-*` to score)
2026-05-04T19:38:54.378918673+00:00 PendingEval → Done (evaluator passed; downstream unblocks)