refactor(context_compressor): extract tool-result summarizers (LB5a)

[andrexibiza]

Summary

God-file Feature Package (tracker #78647): agent/context_compressor.py slice LB5a — extract the tool-result one-line summarizers into agent/context_compressor_summarizers.py with an is-identical re-export seam.

Slice

• God-file: agent/context_compressor.py (6,883 lines @ pin 6e9cae6ac4b)
• Window: 1114–1289 @ pin (459–634 at LB4)
• Golden sha256: efd0867b107f3f25cc0d4d2e4b3cb228be4bd8eef29c26026a593d19749939ec — byte-exact, verified by 5/5 blind reviewers (round 2)
• Module: agent/context_compressor_summarizers.py (3 members)
• Deps: one-way on skill-prune module; logger via lazy in-function round-trip import (auth s2 seam pattern) — no import-time cycle
• Seam: godfile re-exports all 3 members; cc.<m> is cs.<m> verified at runtime

Round-2 fix (blind Pass E caught it)

test_summarize_tool_result_type_safety.py patched agent.context_compressor._summarize_tool_result_unguarded, but the call resolves via the summarizers module global after extraction — the mock never fired. Per extraction doctrine, test patch targets move with the code: retargeted to agent.context_compressor_summarizers._summarize_tool_result_unguarded. Dead typing imports trimmed.

Verification

• 13/13 type-safety + 60/60 full family
• 5/5 blind re-review APPROVED (round 2; round-1 CRITICAL confirmed fixed)
• ruff clean · git diff --check clean · LF-only · DCO signed

Part of #78645
Part of #78647

[andrexibiza]

Interlock: Part of #78645 + #78647. 5/5 blind APPROVED (round 2) receipts in worktree tmp/PASS_*_R2_VERDICT.md.

[andrexibiza]

Interlock completion for the context-compressor shard campaign:

• Current shard PR: refactor(context_compressor): extract durable session guards mixin (LB6) #81074
• Lock issue: Shard agent/context_compressor.py (god-file decomposition) #78645
• Campaign epic: [COMPLETE] Large-file decomposition: 20/20 done #78647
• This node is part of the same coordination graph; no issue is being closed by this refactor.

The full related set is explicitly recorded in #81074, including the prior compressor shards, same-surface colliders/fixers, and their referenced issue nodes. The residual agent/context_compressor.py remains alive until the 2k Law is satisfied.

[Enough1122]

AI code review — automated review for reference, author can ignore or act on any point.

refactor(context_compressor): extract tool-result summarizers (LB5a)

1. Heavy overlap with sibling extraction PRs — this PR and refactor(context_compressor): extract threshold coercion mixin (LB7) #80644 both move budget / media / skill_prune / summarizers / text_utils out of context_compressor.py (this one omits the threshold module that refactor(context_compressor): extract threshold coercion mixin (LB7) #80644 adds). If both land independently they will conflict on the same hunks. Since the series (refactor(context_compressor): extract content/media strip helpers (LB4) #80636 LB4, refactor(context_compressor): extract tool-result summarizers (LB5a) #80645 LB5a, refactor(context_compressor): extract durable session guards mixin (LB6) #81074 LB6, refactor(context_compressor): extract threshold coercion mixin (LB7) #80644 LB7) is clearly staged, coordinating so each stage builds on the merged previous one would avoid merge churn and duplicated seam tests.
2. Patch-target migration is handled correctly — tests/agent/test_summarize_tool_result_type_safety.py moves the mock target to agent.context_compressor_summarizers._summarize_tool_result_unguarded; the name is still re-exported on agent.context_compressor, so both old and new patch paths resolve to the same object. Worth keeping the re-export stable since runtime code may also reference it.
3. The long _summarize_tool_result tool-name if/elif chain moved mechanically — the seam tests cover the moved functions in isolation; recommend one compress()-level test that exercises a summarized tool result end-to-end (e.g. a terminal call with a large output) to prove the extraction didn't alter which branch fires.
4. Minor: the new module's top-level imports from sibling leaves while context_compressor imports it mid-file with # noqa: E402 — the import graph is exercised by cycle-free tests in the other seams, but a comment documenting the allowed dependency direction would help future edits.