Panggilan tool read-only (read, grep, glob, dst) dengan argumen persis
sama menyisakan satu salinan penuh saja di context; entri lama diganti
placeholder tapi tool-call-nya sendiri tetap terlihat di riwayat. Tool
bersifat mutasi (write/edit/bash/dll) tidak pernah disentuh.
Jadikan tool_scope::READ_TOOLS pub supaya jadi satu-satunya sumber
klasifikasi read-only, dipakai ulang bukan didaftar dua kali.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Satukan tiga salinan logika resolusi context_window (Action::Compact,
spawn_turn, status bar) yang sempat melenceng satu sama lain.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Ganti tiga heuristik char-count (/3 di shortsend, /4 di loop turn, /4
di status bar) yang saling tidak konsisten dengan satu BPE tokenizer
nyata. tiktoken-rs membundel vocab lewat include_str! saat build, jadi
tidak ada akses jaringan saat runtime.
Fixture pendek sebelumnya (~8 token nyata per pesan lewat tiktoken)
tidak pernah melebihi target 700 token, jadi shape_messages tidak
pernah men-drop satu pesan pun -- satu test gagal, satu test lain
lulus secara vakum. Pakai fixture lebih panjang (~49 token/pesan)
yang diverifikasi melebihi target dengan nyaman.
AppStateRest::new() men-spawn thread ignore::Walk untuk mention_index
tanpa syarat, padahal mode --attach cuma dipakai untuk render lokal —
handle_key dan logika mention berjalan di sisi daemon lewat IPC, jadi
index di client attach tidak pernah dipakai. Ini bikin setiap
--attach melakukan full workspace walk (sampai 50.000 entry) sia-sia.
Pindahkan thread-spawn itu ke method terpisah
spawn_mention_index_build(), dipanggil eksplisit dari
run_single_process() dan run_daemon() setelah AppStateRest::new(),
tapi sengaja tidak dipanggil dari run_attach().
Cursor bisa berpindah (Left/Right) tanpa menutup dropdown mention,
sehingga mention_start jadi stale relatif ke cursor saat Enter
ditekan. select_autocomplete() lalu memanggil replace_range dengan
start > end dan panic (crash seluruh TUI, termasuk daemon). Tambah
guard: jika cursor < mention_start atau mention_start > buffer.len(),
tutup dropdown dan kembalikan false alih-alih menyambung range yang
tidak valid. Tambah regression test yang mereproduksi skenario ini.
Menambahkan field mention_index: MentionIndex ke ToolCtx/ToolCtxBuilder dan
AppStateRest, mengikuti pola dir_cache yang sudah ada. AppStateRest::new()
sekarang men-spawn thread background yang men-walk setiap workspace root
via ignore::Walk, membangun daftar path file (dengan prefix [N] untuk root
selain yang pertama) dan mengisi mention_index — data ini yang akan dipakai
fitur autocomplete @file-mention.
Menambahkan MentionIndex (indeks path file thread-safe untuk fitur
autocomplete @file-mention), enum AutocompleteKind untuk membedakan
dropdown slash-command dan file-mention, serta method baru pada
InputState: mention_query_at_cursor untuk deteksi token @mention di
posisi cursor, dan open_mention_autocomplete untuk fuzzy-match file
via nucleo-matcher. select_autocomplete kini kind-aware: menyisipkan
path file ke posisi mention alih-alih mengganti seluruh buffer.
- Added functionality to generate lessons based on code reviews, including prompts and instructions for the reviewer.
- Updated the `.gitignore` to exclude lesson documentation files.
- Removed the `/lesson` command from the help menu and command parsing.
- Enhanced the status bar to display a message when a lesson is being generated.
- Introduced a new `lesson_running` state to track lesson generation progress.
- Updated various components to handle the new lesson generation workflow, including subagent events and token usage tracking.
feat(engine): summarize agent completion progress in UI notifications
feat(main): enable and disable bracketed paste support in terminal
fix(app_config): store API key from file as fallback for Claude provider
refactor(workflow): remove unnecessary card separator in workflow panel
Overlay Todo dan Usage sebelumnya tidak punya trigger sama sekali di
jalur interaksi normal (cuma bisa lewat restore snapshot sesi) --
sekarang mengikuti pola /workflow yang sudah ada.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
unnested_or_patterns, collapsible_if, items_after_statements, dan
case_sensitive_file_extension_comparisons di auto.rs, engine.rs, dan
actions/mod.rs.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Pesan sistem bertanda [Hive-Mind Consensus] hanya di-push ke variabel lokal
run_agent_turn dan diarsipkan ke SQLite, tidak pernah masuk ke
rt.messages lewat TurnEvent — sehingga hive_mind_already_ran selalu
memindai daftar pesan yang kosong dan gerbang "converge sekali per sesi"
tidak pernah aktif. Tambahkan SessionRuntime.hive_mind_converged yang
diset dari event TurnEvent::SystemNote { kind: "hive_mind_converged" }
setelah konvergensi selesai, disalurkan lewat TurnCtx, dan dijadikan
sinyal utama di run_agent_turn (pemindaian pesan lama tetap sebagai
fallback defensif).
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
RunningGuard resets TEST_GEN_RUNNING/ARCH_REVIEW_RUNNING/SECURITY_REVIEW_RUNNING
via Drop so a subagent panic can no longer wedge that review kind disabled
for the rest of the process. Doc comments on run_subagent_with_retry and the
three spawn_background_* / spawn_all_background functions now describe the
abort_flag and overlap-guard behavior added in Task 7.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
- Introduced a new hive-mind architecture that allows the Core Intelligence to issue directives to anonymous processing nodes.
- Each node executes its directive and merges output into a collective state, visible to all nodes in real-time.
- Added support for dynamic cognitive cycles, enabling flexible task management.
- Implemented documentation generation for hive-mind runs, ensuring a durable record of decisions and actions.
- Refactored existing company pipeline tools to align with the new hive-mind structure, replacing division-specific prompts with a more generalized approach.
- Updated workflow rendering to accommodate hive-mind nodes and their system-assigned designations.
- Enhanced error handling and validation for cognitive cycle plans.
- Updated README.md to reflect the addition of 3 new built-in tools, bringing the total to 37.
- Revised architecture documentation to indicate the increase in tool count.
- Enhanced backend documentation with updated line counts for various modules.
- Modified data documentation to change edit log format from JSON to JSONL.
- Updated dependencies documentation to reflect version upgrades for several crates.
- Improved prompts for auto-reviewer, division implementer, planner, tester, and quality reviewer to enforce stricter coding standards regarding linter bypasses.
- Refactored code in various modules to improve clarity and performance, including updates to error handling and tool execution logic.
- Added comprehensive tests for IPC frame serialization and deserialization.
- Updated `crossterm` from version 0.28 to 0.29.
- Upgraded `reqwest` from version 0.12 to 0.13 and added "form" feature.
- Bumped `serde_yaml_ng` from version 0.9 to 0.10.
- Increased `dirs` version from 5 to 6.
- Updated `rusqlite` from version 0.32 to 0.40.
- Upgraded `infer` from version 0.16 to 0.19.
- Bumped `sha2` from version 0.10 to 0.11.
- Updated `rmcp` from version 1.8 to 2.2.
- Refactored SHA256 hash encoding in `rewind.rs`, `mod.rs`, and `rest.rs` to use `hex::encode` instead of formatting with `{:x}` for better clarity and consistency.
- Introduced prompts for various subagent roles: architecture reviewer, code quality reviewer, documentation maintainer, implementation team, testing team, and security reviewer.
- Implemented the auto-subagent orchestration in `auto.rs` to manage inline and background reviews.
- Created a division structure in `division.rs` to define roles and responsibilities for each subagent.
- Developed a company workflow orchestrator in `company.rs` to run the complete division pipeline, consolidating findings and generating executive summaries.
- Added logic to determine whether to run a full or quick pipeline based on request complexity.
