fix(opencode): accumulate step tokens for compaction

[Skyline-23]

What does this PR do?

• Accumulates step-level usage tokens across steps so compaction uses the cumulative token count instead of only the last step.
• Fixes Codex model response latency increases significantly as conversation grows #9045

How did you verify your code works?

• bun test --timeout 30000 --bail (failed: ConnectionRefused to https://models.dev/api.json)

[github-actions]

Thanks for your contribution!

This PR doesn't have a linked issue. All PRs must reference an existing issue.

Please:

1. Open an issue describing the bug/feature (if one doesn't exist)
2. Add Fixes #<number> or Closes #<number> to this PR description

See CONTRIBUTING.md for details.

[github-actions]

The following comment was made by an LLM, it may be inaccurate:

Potential Duplicate/Related PRs Found:

1. PR feat(session): add custom compaction thresholds along with prevention of continuous compaction #10123 - "feat(session): add custom compaction thresholds along with prevention of continuous compaction"

   • Related to token overflow handling and compaction triggering mechanisms

2. PR feat(opencode): add context overflow prevention with configurable thresholds #8810 - "feat(opencode): add context overflow prevention with configurable thresholds"

   • Addresses context overflow prevention with thresholds

3. PR fix(session): prevent context overflow by adding safety margin to compaction check #6562 - "fix(session): prevent context overflow by adding safety margin to compaction check"

   • Fixes overflow prevention in compaction logic

4. PR feat(opencode): trigger compaction earlier and add multi-file read #9656 - "feat(opencode): trigger compaction earlier and add multi-file read"

   • Related to compaction triggering conditions

5. PR feat(compaction): Hybrid compaction pipeline with deterministic extraction + LLM #7104 - "feat(compaction): Hybrid compaction pipeline with deterministic extraction + LLM"

   • Core compaction pipeline implementation

These PRs all relate to token management, compaction triggering, and overflow prevention. PR #10123 seems most closely related as it deals with custom compaction thresholds and token overflow. Recommend reviewing these to ensure PR #10215's token accumulation logic integrates properly with existing compaction mechanisms.

[Copilot]

Pull request overview

This pull request fixes token accumulation in the OpenCode session processor to properly track cumulative token usage across multi-step loops and enable correct overflow detection for compaction triggers.

Changes:

• Accumulate step-level token usage instead of overwriting per-step values in the assistant message
• Use cumulative tokens for overflow checks to properly trigger compaction in multi-step scenarios

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.