Compaction: Telemetry and /status for context + decisions

[riatzukiza]

Epic: #46

Summary

Add observability for compaction and pruning decisions to make behavior transparent and debuggable.

Problem

• Hard to see which messages were kept/dropped and why. No easy token breakdown.

Proposal

• Log boundaries: preserved turns, summary index, message IDs included.
• Log pruning: tokens pruned/protected, count of parts affected.
• Provide /status (or extend existing status) to show token usage by category (user/assistant/tool/reasoning) and compaction readiness.

Code Refs

• packages/opencode/src/session/compaction.ts (boundary and prune points)
• packages/opencode/src/session/message-v2.ts (message/part conversions)

Tasks

• Implement structured logs (with service=, ids, token counts)
• Implement /status details for compaction/prune state

Acceptance Criteria

• Developers can see boundary/prune decisions and token breakdowns without reading code.

[riatzukiza]

Related Issues & Context

This Telemetry and Status issue is essential for debugging and monitoring ALL other compaction improvements:

Cross-Cutting Dependencies:

• Compaction: Smarter boundary selection and cross-session guard #47 - Boundary Selection: Need visibility into which messages/parts were kept vs dropped and why
• Compaction: Preserve rules/constraints across compaction #48 - Rules Preservation: Need to verify rules were successfully included in summaries
• Compaction: Structured BUILD-mode handoff summary prompt #49 - Structured Handoff: Need to see handoff quality and token costs
• Compaction: Pruning/soft compaction and selective tool-output deletion #50 - Pruning: Need visibility into pruning decisions and token reclamation
• Compaction: Concurrency/locking robustness and retries #53 - Concurrency: Need to detect race conditions and partial state issues
• Compaction: Provider options flow and accurate usage accounting #51 - Usage Accounting: Token breakdowns are core to cost tracking

Strategic Importance:

• Epic: Subagent State Handoff & Context Boundary Resilience #56 - Subagent State Handoff: This telemetry infrastructure becomes the foundation for subagent monitoring. Same visibility needed for:
  • What context was inherited by subagents?
  • What state was lost in subagent failures?
  • Why did subagents get stuck?

User Experience:

The status command extension will be CRITICAL for:

• Developers debugging compaction behavior
• Users understanding why context was lost
• Support diagnosing subagent failure patterns

Technical Infrastructure:

The structured logging patterns established here become the template for:

• Subagent state change logging
• Context boundary crossing events
• Error propagation across contexts

Implementation Priority:

Telemetry should be implemented EARLY because:

1. Debugging Foundation: All other compaction work needs visibility
2. Validation: Need to verify fixes are working correctly
3. User Trust: Transparency builds confidence in compaction behavior

Cross-Cutting Impact:

This issue provides the observability backbone for BOTH:

1. Compaction debugging - understanding context transfer within sessions
2. Subagent monitoring - understanding context transfer between agents

Without proper telemetry, both compaction and subagent issues remain black boxes that are impossible to debug effectively.