fix(opencode): route GPT models and discover OSS fallbacks

[dgokeeffe]

Issue and stack

Closes #85.

Depends on #217, #333, and #223. Review only the OpenCode/discovery incremental range:

dgokeeffe/ucode@94fe107...816588f

Do not merge before #223. Once prerequisites merge, the visible PR diff collapses to the eight-file OpenCode unit.

---

Unit 04 — OpenCode GPT routing and OSS fallback discovery

Objective and user-visible behavior

Route GPT/Codex models through OpenCode's Databricks Responses provider and discover validated OSS serving endpoints when UC model services do not expose them.

Candidate: review/opencode-routing (816588f), mandatory PR target review/pi-gpt-reasoning (94fe107), existing PR #239.

Exact scope

Production: src/ucode/agents/opencode.py, src/ucode/cli.py, src/ucode/databricks.py.
Tests/fixtures: tests/conftest.py, tests/test_agent_opencode.py, tests/test_cli.py, tests/test_databricks.py, tests/test_e2e.py.
Non-goals: Pi behavior, context policy, dynamic OSS capability metadata, MLflow repair proxy, cache behavior.

Before / after reproduction

git diff --stat 94fe107..816588f
# 8 files changed, 480 insertions(+), 20 deletions(-)

git show review/opencode-routing:src/ucode/agents/opencode.py | rg 'databricks-openai|openai-responses'
# databricks-openai provider and model qualification are present

Focused validation:

uv run --frozen pytest tests/test_agent_opencode.py tests/test_cli.py tests/test_databricks.py tests/test_e2e.py -q
# 469 passed, 29 skipped
uv run --frozen ruff check .
# All checks passed

Full suite:

uv run --frozen pytest -q
# 2 failed, 1649 passed, 36 skipped

Failures are installed Claude/Pi User-Agent capture tests. The branch predates the independent config-dir unit; integrated dev passes the Pi capture test and independently reproduces the Claude capture failure.

Live/discovery evidence

No token-bearing output is retained. Fixture coverage reproduces UC-first discovery, serving-endpoint fallback, dialect filtering, validated GLM/Kimi allowlisting, GPT provider qualification, default routing, and per-model E2E configuration with workspace-gated skips.

Impact map

• Agent: OpenCode.
• Providers: databricks-openai Responses and existing databricks-oss MLflow provider.
• Discovery/config: shared state gains OpenCode GPT list and fallback OSS IDs; OpenCode config qualifies defaults by provider.
• Compatibility: native Claude/Gemini endpoints and unsupported chat models remain excluded from OSS fallback.
• Pi/MLflow proxy/MCP/skills: unchanged.

Rollback and residual risk

Revert 816588f; OpenCode loses GPT routing and serving-endpoint OSS fallback but existing native-family behavior remains. Residual risk is future endpoint metadata drift; unit 05 adds structured capability validation separately.

Hygiene

git diff --check review/pi-gpt-reasoning..review/opencode-routing passes. PR #239 must target unit 02; its target-relative diff is the review contract. No generated files, credentials, uv.lock, .pi-subagents/, or goal.md; no merge markers or unresolved index entries.

[dgokeeffe]

This stack has been rebuilt on current main with one clean commit per layer. The incremental OpenCode diff is linked in the description; the rejected proxy and broad OSS cohort are gone. Local focused validation is 480 passing with Ruff clean. A maintainer CI approval/review is now the remaining gate.

[dgokeeffe]

Post-review update: OpenCode GPT entries now carry shared context/output limits and prefer the newest eligible GPT; foundation endpoint parsing is defensive against malformed records; CLI fallback tests are deterministic; and the live e2e fixture now populates OpenAI/OSS models and isolates Pi settings paths. Round-3 reviewers found no production blocker; final validation is 588 passed/29 skipped focused and 1062 passed/36 skipped excluding environment-sensitive installed-agent capture tests.

[dgokeeffe]

Rebased this PR stack onto current main (67429ec) and force-pushed the existing branch. New head: 816588f.

Hard stack dependency

This is 3 of 3 and must land after #223:

1. feat(pi): add validated GLM/Kimi MLflow provider #217 — Pi MLflow OSS provider
2. fix(pi): omit unsupported GPT off-reasoning effort #223 — shared Claude/GPT metadata + Pi GPT thinking-off fix
3. fix(opencode): route GPT models and discover OSS fallbacks #239 — OpenCode GPT routing + OSS discovery fallback (this PR)

This is a code dependency, not just a review preference: #239's preferred_gpt_model uses _normalized_foundation_model_id, and its per-model overlay uses gpt_model_token_limits; both are introduced by #223. #239 cannot safely apply directly to main before #223.

The helper and its tests now live in this PR with their only consumer (src/ucode/agents/opencode.py), rather than leaking into #223.

If #217/#223 are squash-merged, please rebase this branch onto the updated main before merging.

Validation

• CI-parity suite: 1,648 passed, 6 skipped
• Only local failures: the two pre-existing CLI capture-server tests (TestClaudeUserAgent, TestPiUserAgent), which skip in CI's unit-test job
• ruff check and ruff format --check: clean
• No committed uv.lock churn or conflict markers

This PR adds lower-level OpenCode GPT routing, but managed-config OpenCode GPT mapping and admin allowlist intersection are deliberately deferred to #290.