fix: dedupe chunks across queries in a multi-query ctx_search call

[uwuclxdy]

What / Why / How

Root cause: ctx_search runs each query in queryList independently and pushes a ## ${q} section per query, with no state shared across the loop. A chunk that matches more than one query is printed verbatim under every matching query's section, spending context on duplicates.

Fix: dedupe chunks across the call's queries with a shared Set that keeps the first occurrence. It lives in a small exported helper, dedupeAcrossQueries, so tests can hit it directly. Neither SearchResult nor UnifiedSearchResult has a stable chunk id. The key is therefore the query-independent source + title + content tuple; the per-query snippet cannot serve, because extractSnippet windows it around each query's terms. Single-query output stays byte-identical: the dedup only runs when queryList.length > 1. A later query whose matches were all shown earlier now reads (all matches already shown above).

Affected platforms

• Claude Code
• Cursor
• VS Code Copilot (GitHub Copilot)
• JetBrains Copilot
• Gemini CLI
• Qwen Code
• OpenCode
• KiloCode
• Codex CLI
• OpenClaw (Pi Agent)
• Pi
• Kiro
• Antigravity
• Zed
• All platforms

Test plan

• npx tsc -b --noEmit clean.
• npx vitest run tests/core/search.test.ts green. Added ctx_search cross-query dedup, covering: a chunk matching two queries is emitted once with the first occurrence kept, and distinct chunks that share source and title are not falsely deduped.

Checklist

• Tests added/updated (TDD: red → green)
• npm test passes
• npm run typecheck passes
• Docs updated if needed (README, platform-support.md)
• No Windows path regressions (forward slashes only)
• Targets next branch (unless hotfix)

Cross-platform notes

Our CI runs on Ubuntu, macOS, and Windows.

• If touching file paths, verify forward-slash normalization on Windows
• If touching hook paths, verify no backslash separators
• Use path.join() / path.resolve(), never hardcode / separators
• Use event-based stdin reading — readFileSync(0) breaks on Windows
• Use os.tmpdir(), never hardcode /tmp