fix(compression): skip empty summary windows

[HOYALIM]

Problem

Fixes #59496. After the latest persisted handoff summary is removed from the compression window, the remaining window can be empty. The compressor still proceeded to _generate_summary([]), spending one auxiliary model request and then surfacing the empty-window failure path.

Solution

After rehydrating _previous_summary from the latest handoff, return early when there are no new turns left to summarize. The early return mirrors the existing no-compressible-window path: it records an ineffective compression attempt, sets savings to 0%, and avoids the model call.

Validation

• scripts/run_tests.sh tests/agent/test_context_compressor_summary_continuity.py -q
• scripts/run_tests.sh tests/agent/test_context_compressor.py tests/agent/test_context_compressor_cross_session_guard.py tests/agent/test_preflight_compression_gate.py tests/agent/test_context_compressor_summary_continuity.py -q
• .venv/bin/ruff check agent/context_compressor.py tests/agent/test_context_compressor_summary_continuity.py
• git diff --check
• uv lock --check

Impact

For the affected boundary case, _generate_summary calls drop from 1 to 0. That avoids one auxiliary context-engine request, its prompt tokens, and its latency for each no-new-turn compression trigger.

Notes

ty check agent/context_compressor.py tests/agent/test_context_compressor_summary_continuity.py still reports existing annotations/type issues in agent/context_compressor.py around nullable defaults and call_llm(**call_kwargs); this patch does not add new ty diagnostics.

[Copilot]

Pull request overview

This PR fixes an edge case in the context compression pipeline where, after rehydrating _previous_summary from a persisted handoff summary, the adjusted compression slice can become empty but the compressor still called _generate_summary([]), wasting an auxiliary model request and triggering the “summary generation failed → abort” path.

Changes:

• Add an early-return guard in ContextCompressor.compress() to treat a post-handoff-adjustment empty turns_to_summarize window as a no-op (record ineffective attempt, 0% savings, skip model call).
• Add a regression test ensuring the compressor does not call _generate_summary when the handoff summary sits at the end of the compression window, and that state counters are updated accordingly.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated no comments.

File	Description
agent/context_compressor.py	Adds a no-op guard after handoff summary rehydration to avoid calling the summarizer with an empty window.
tests/agent/test_context_compressor_summary_continuity.py	Adds coverage for the “handoff at window end → no new turns” boundary condition and asserts no model call occurs.

---

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

[alt-glitch]

This was generated by AI during triage.

Duplicate of #59526 — diff-verified: this PR inserts the identical if not turns_to_summarize: early-return guard in context_compressor.compress() (same _ineffective_compression_count/_last_compression_savings_pct = 0.0/return messages), differing only in log-message wording. Both Closes #59496. #59526 is the earlier, open, canonical PR; the extra regression test here doesn't change the mechanism. (Sibling #59737 is the same fix, also linked to #59526.)

[HOYALIM]

@HOYALIM Closing this as a duplicate of #59526, which was opened earlier for #59496 and carries the same guard plus regression coverage. Next step is to track #59526 as the canonical fix.