(CODUX-129) Harden OpenCode and Antigravity parsers

[numnx]

🧩 Task: Harden OpenCode and Antigravity parsers

Automated task PR opened by Code UX for T04.

📋 Summary

Sprint: Sprint 3 · Feat: Invocation Parsing all Providers · Task: T04

Implement extended invocation parsing across every local CLI provider so provider-backed invocations persist normalized user, assistant, reasoning, tool-call, and tool-result turns with accurate usage metadata, regression coverage, and updated documentation.

🤖 Model & Provider

Provider	Codex CLI
Model	gpt-5.5

⏱️ Timing

Started	2026-07-10 05:53:06 UTC
Finished	2026-07-10 06:02:56 UTC
Duration	9m 49s

📊 Token Usage

Metric	Value
Input tokens	99,196
Cached input tokens	2,080,512
Output tokens	19,359
Total tokens	2,199,067
Invocations	1
Tool calls	49
Active time	8m 2s

Included usage estimate (subscription/local login): $2.12

✅ QA Review

⏳ QA review runs after this PR is opened. Check the task card in the dashboard for the outcome once it lands.

📝 Full Task Prompt

## Objective
Improve OpenCode and Antigravity invocation parsing so their native streams and transcript stores produce normalized conversation turns and usage isolated to the current invocation.

## Scope
- Edit `src/infrastructure/providers/cli/provider-logs/opencode-log-parser.ts`
- Edit `src/infrastructure/providers/cli/provider-logs/antigravity-log-parser.ts`
- Edit `src/infrastructure/providers/cli/provider-transcripts.ts` only if transcript lookup patterns need to include additional known Antigravity paths
- Edit `tests/backend/infrastructure/providers/cli/opencode-log-parser.test.ts`
- Edit `tests/backend/infrastructure/providers/cli/antigravity-log-parser.test.ts`
- Verify integration behavior through `tests/backend/infrastructure/providers/cli/provider-usage.test.ts`
- Relevant symbols: `parseOpenCodeJsonLines`, `parseOpenCodeExport`, `subtractOpenCodeBaseline`, `parseAntigravityTranscript`, `parseAntigravityDatabase`, `parseAntigravityConversationId`

## Implementation Requirements
1. Extend OpenCode parsing for `session.created`, text, reasoning, tool, step-finish, assistant message, and current export payload shapes while keeping session id extraction stable.
2. Ensure OpenCode export usage remains authoritative when present and that `subtractOpenCodeBaseline` isolates follow-up runs from cumulative native session totals.
3. Extend Antigravity transcript parsing for transcript JSONL and overview-style entries that expose user, assistant, reasoning, tool call, and tool result content, while preserving database usage extraction from `gen_metadata` rows.
4. Add parser tests for OpenCode streaming turns, export payload extraction from noisy stdout, baseline subtraction, Antigravity transcript variants, conversation id detection, malformed rows, and `sinceIdx` database usage isolation.

## Constraints
- Do not treat OpenCode `run --format json` stdout token data as more authoritative than `opencode export` when export data is available.
- Do not re-report prior OpenCode or Antigravity cumulative session usage for resumed invocations.
- Do not throw on malformed transcript rows or missing local database files.
- Do not edit Codex, Gemini, Claude, or Qwen parser files in this task.

## Verification
- Run `pnpm exec vitest run tests/backend/infrastructure/providers/cli/opencode-log-parser.test.ts tests/backend/infrastructure/providers/cli/antigravity-log-parser.test.ts tests/backend/infrastructure/providers/cli/provider-usage.test.ts`.
- Run `pnpm run lint`.
- Success means OpenCode and Antigravity emit normalized turns when data exists, isolate cumulative usage for resumed sessions, and fall back cleanly when transcript or usage data is unavailable.

🌿 Branch Info

Base: feature/sprint-COWIN-3-feat-invocation-parsing-all-providers-implementation
Head: task/feature-sprint-c-t04-codex-6b1c8201-mreirsd4

---

🤖 Generated by Code UX