fix(tools): filter compaction summaries from session_search bookends and cap content length

[liuhao1024]

What does this PR do?

Filters context-compaction handoff summaries from session_search discovery bookends and caps message content length to prevent prompt bloat.

Related Issue

Fixes #43175

Type of Change

• 🐛 Bug fix (non-breaking change that fixes an issue)

Changes Made

• tools/session_search_tool.py: Add _COMPACTION_PREFIXES constant and _is_compaction_summary() helper to detect compaction handoff messages. Filter these from bookend_start and bookend_end in _discover(). Cap bookend content to 1200 chars and window message content to 4000 chars via new max_content_len parameter on _shape_message(). Add content_truncated and original_content_chars metadata when truncation occurs.
• tests/tools/test_session_search.py: Add 6 regression tests — prefix detection, bookend_start filtering, bookend_end filtering, bookend content capping, window content capping, and legacy [CONTEXT SUMMARY] prefix handling.

How to Test

1. pytest tests/tools/test_session_search.py -v — all 52 tests should pass (46 existing + 6 new)
2. Create a test session with a [CONTEXT COMPACTION — REFERENCE ONLY] message as the first user message, followed by normal messages
3. Run session_search(query=<term in later message>) — the compaction summary should NOT appear in bookend_start
4. Verify that long bookend/window messages are capped (bookend ≤1200 chars, window ≤4000 chars) with content_truncated: true metadata

Checklist

Code

• I've read the Contributing Guide
• My commit messages follow Conventional Commits (fix(scope):, feat(scope):, etc.)
• I searched for existing PRs to make sure this isn't a duplicate
• My PR contains only changes related to this fix/feature (no unrelated commits)
• I've run pytest tests/tools/test_session_search.py -q and all tests pass
• I've added tests for my changes (required for bug fixes, strongly encouraged for features)
• I've tested on my platform: macOS

Documentation & Housekeeping

• I've updated relevant documentation (README, docs/, docstrings) — or N/A
• I've updated cli-config.yaml.example if I added/changed config keys — or N/A
• I've updated CONTRIBUTING.md or AGENTS.md if I changed architecture or workflows — or N/A
• I've considered cross-platform impact (Windows, macOS) per the compatibility guide — or N/A
• I've updated tool descriptions/schemas if I changed tool behavior — or N/A

Code Intelligence

• Analyzed: tools/session_search_tool.py (_discover, _shape_message, _is_compaction_summary)
• Analyzed: hermes_state.py (get_anchored_view — bookend SQL queries)
• Blast radius: LOW — changes are additive (filtering + capping) in a single tool file; no control flow changes to the underlying DB queries
• Related patterns: Content truncation follows the same pattern as session_search scroll mode's existing read truncation

[teknium1]

Thanks for targeting a real current-main retrieval bloat path: _discover() still returns unbounded bookends and windows at tools/session_search_tool.py:599-601.

Problems

• The new prefix-only predicate misses merged handoffs. Current compaction explicitly recognizes summaries after _MERGED_SUMMARY_DELIMITER in agent/context_compressor.py:2261-2270; those messages would still enter bookends.
• Current title-match discovery has a separate response construction at tools/session_search_tool.py:487-489, using unbounded _shape_message() calls. The proposed _discover() change does not cover that path.

Suggested changes

• Centralize bounded, compaction-aware shaping and use it for both normal and title-match discovery results.
• Add tests for merged-summary bookends and a title query resolving to a session with a long handoff.

Automated hermes-sweeper review.

[teknium1]

This only detects a handoff when the prefix begins the message. Current compaction also recognizes merged summaries whose handoff starts after _MERGED_SUMMARY_DELIMITER (agent/context_compressor.py:2261-2270); those can still be returned in bookends. Please cover that merged form as well.

[teknium1]

Merged via #69334 (commit 9bb253d). Your commit was cherry-picked with authorship preserved, rebased over the schema-v23 FTS work. Thanks!