Compaction: Tests – boundary, rules, pruning, integration

[riatzukiza]

Epic: #46

Summary

Add unit and integration tests that prevent regressions in compaction.

Scope

• Unit: boundary selection (empty stream, multiple summaries, preserve last N turns)
• Unit: rules presence in summary
• Unit: pruning decisions (completed tool outputs pruning)
• Integration: BUILD-mode compaction -> continuation on a toy repo

Tasks

• Test harness to seed session with messages/parts programmatically
• Unit tests for new boundary function
• Unit tests for summary prompt includes mandatory sections
• Integration: simulate long BUILD, trigger prune + compact, verify continuation

Acceptance Criteria

• Tests fail on the historical bugs described in linked issues and pass after fixes.

[riatzukiza]

Related Issues & Context

This Testing issue provides the validation foundation for ALL other compaction improvements:

Test Coverage Dependencies:

• Compaction: Smarter boundary selection and cross-session guard #47 - Boundary Selection: Unit tests for empty slice, multiple summaries, preserved turns logic
• Compaction: Preserve rules/constraints across compaction #48 - Rules Preservation: Unit tests verifying rules section presence in summaries
• Compaction: Structured BUILD-mode handoff summary prompt #49 - Structured Handoff: Integration tests for BUILD-mode continuation across compaction
• Compaction: Pruning/soft compaction and selective tool-output deletion #50 - Pruning: Unit tests for pruning decisions and tool output selection

Cross-Cutting Validation:

• Compaction: Provider options flow and accurate usage accounting #51 - Usage Accounting: Tests for usage consistency between compaction and non-compaction flows
• Compaction: Telemetry and /status for context + decisions #52 - Telemetry: Tests for logging accuracy and status command output
• Compaction: Concurrency/locking robustness and retries #53 - Concurrency: Integration tests for concurrent compactions and abort scenarios

Strategic Importance:

• Epic: Subagent State Handoff & Context Boundary Resilience #56 - Subagent State Handoff: The test harness developed here becomes the foundation for subagent testing:
  • Session seeding for subagent test scenarios
  • Boundary testing for context inheritance
  • Integration tests for parent-child communication

Regression Prevention:

The integration test ("simulate long BUILD, trigger prune + compact, verify continuation") is CRITICAL for preventing regressions in:

• Session automatically compacted, destroying the entire working context anomalyco/opencode#2945 (working context destroyed)
• Model in BUILD mode does not have enough context to continue after compaction anomalyco/opencode#3031 (BUILD mode context loss)
• Agent no follow rules after compact session anomalyco/opencode#3099 (rules not followed)

Implementation Priority:

Tests should be developed IN PARALLEL with features because:

1. Validation: Need to verify each compaction improvement works correctly
2. Regression Prevention: Historical bugs must not reoccur
3. Confidence: Tests enable aggressive refactoring of compaction logic

Cross-Cutting Impact:

The test infrastructure established here serves BOTH:

1. Compaction validation - ensuring context transfer works within sessions
2. Subagent validation - ensuring context transfer works between agents

Long-term Value:

The session seeding and boundary testing capabilities become essential for:

• Future context transfer features
• Performance optimization testing
• User scenario simulation
• CI/CD pipeline validation

Without comprehensive test coverage, both compaction and subagent features remain fragile and prone to regression.