v0.4.1-fixes: AI hardening, model-agnostic prompts, codebase modularization - #17
Merged
Conversation
- Add max_rounds guard at top of tool loop — previously set but never checked - Fix oscillation detection: count frequency in sliding window instead of requiring consecutive identical turns. Catches A→B→A→B patterns that previously evaded the circuit breaker - Reduce aggressive_prune from 25% to 10% to preserve more context - Reduce context window shrink from 50% to 20% on overflow - Improve truncation message with actionable guidance (use line ranges, narrow queries) instead of generic warning - Remove unused consecutive_identical_tools field - Add tests: max_rounds enforcement, A-B oscillation detection, prune ratio Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Fix default ai_mode from invalid "manual" to "standard" (both constructors) - Fix BackTab cycle to use valid modes: standard → auto-accept → plan - Add mode to status bar: [AI:standard|Write] in both terminal and GUI - Add plan-mode guard on buffer_write — returns error suggesting create_plan - Add mode/profile context injection per-turn so model knows its constraints Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Remove dead terminal_list entry from executor (duplicate of shell_list) - Add comment documenting session-level tools handled in session.rs - Define AI_PROFILES constant and use in ai_set_profile + delegate tools - Change brittle assert_eq!(tools.len(), 98) to assert!(>= 90) Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Rewrite all 4 profile prompts: shorter, concrete rules, no <reasoning> blocks - Add READ-ONLY hard constraints (Claude Code pattern) to explorer/planner/reviewer - Add "When to Stop" section to system prompt with explicit completion criteria - Add "Mode Rules (ENFORCED)" section replacing advisory mode paragraph - Remove hardcoded profile names from delegation SOP - Drop Forced Reasoning Block requirement (Claude-ism that confuses other models) - Prompts now work across Claude, Gemini, OpenAI, DeepSeek, and smaller models Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
…tripping - gpt-4o-mini: 100→30, deepseek-*: 100→25 to prevent token drain - trim_messages: only strip orphaned tool messages after actual pruning, preventing destruction of valid tool history on non-pruned sessions - Add regression test for trim preservation Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Split monolithic editor/tests.rs into editor/tests/ directory: - command_tests (48), navigation_tests (55), visual_tests (34) - editing_tests (28), operator_tests (33), count_tests (21) - change_tests (14), search_tests (12), text_object_tests (15) - lsp_tests (21), shell_tests (15), mouse_tests (13) - buffer_tests (12), misc_tests (18) All 1066 tests pass. Shared helpers in tests/mod.rs. No file exceeds 800-line ceiling (max: navigation_tests at 757). Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
…modules Split key_handling.rs into key_handling/ directory: - mod.rs (363) — handle_key dispatcher, conversion fns, shared helpers - normal.rs (500) — handle_keymap_mode, handle_describe_key_await, handle_normal_mode - command.rs (336) — handle_command_mode, tab completion, self-test prompt - conversation.rs (265) — handle_conversation_input, submit_conversation_prompt - file_picker.rs (225) — handle_file_picker_mode, handle_file_browser_mode - insert.rs (154) — handle_insert_mode - command_palette.rs (110) — handle_command_palette_mode - visual.rs (81) — handle_visual_mode - git_status.rs (70) — handle_git_status_mode - search.rs (29) — handle_search_mode All tests pass. No file exceeds 500-line ceiling. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
tools.rs (2704 lines) → tools/ directory (7 modules + mod.rs) executor.rs (2050 lines) → executor/ directory (6 helpers + mod.rs) All 232 AI tests pass. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
…eys from main.rs main.rs (2477 → 1148 lines, 460 non-GUI): - terminal_loop.rs (564) — run_terminal_loop, cleanup_stale_mcp_sockets, run_headless_self_test - lsp_bridge.rs (433) — drain_lsp_intents, handle_lsp_event, try_complete_deferred, open_location - dap_bridge.rs (217) — drain_dap_intents, handle_dap_event, intent_to_dap_command - shell_keys.rs (157) — shell_dims_for_buffer, handle_shell_key, keypress_to_pty_bytes GUI code remains in main.rs under #[cfg(feature = "gui")] for a future split. All tests pass. make build (with GUI) succeeds. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- mod.rs (183) — struct definition, constructors, builders - run_loop.rs (178) — run() async loop, execute_shell(), update_cost() - context_mgmt.rs (149) — trim_messages, aggressive_prune, budget - handle_prompt.rs (1013) — handle_prompt(), collapse_transaction() - tests.rs (1295) — all session tests All 232 AI tests pass. Fields changed to pub(super) for cross-module access. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- #1: Replace .expect() with .unwrap_or_default() in provider HTTP clients - #2: Remove redundant Some("STOP") match arm in gemini.rs - #3/#17: Fix ErrorKind::Unknown → ErrorKind::Transport for HTTP/JSON errors across all three providers (claude, gemini, openai) - #6: Add finalize_transaction() helper, replace 10 collapse+clear patterns - #7: Cache transcript_path as String once, eliminate 8x to_string_lossy allocs - #15: Remove dead `let _ = editor` in executor benchmark fn - #18: Fix stale gemini-3.1-*/3.0-flash model names → real 2.5/2.0 names in both context_limits.rs and pricing.rs; deduplicate gemini-2.5-pro entry Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- ROADMAP: v0.4.0→v0.4.1, add Phase 3g-v2 section (6 file splits, 12 code smell fixes), update test table (1,324→1,590), add rendering dedup and packaging readiness to priority list - README: badges updated (tests 1,324→1,590, LOC ~26k→~76k) - CLAUDE.md: v0.4.1 modularization note in Phase 3 section Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
This was referenced Aug 5, 2026
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
core/editor/tests.rs(4,458 lines) → 14 test modulesmae/key_handling.rs(2,056 lines) → 10 mode-specific modulesai/tools.rs(2,704 lines) → 8 modules by categoryai/executor.rs(2,050 lines) → 7 modules by categoryai/session.rs(2,791 lines) → 5 modules (run_loop, context_mgmt, handle_prompt, tests)mae/main.rs(2,477 lines) → extracted terminal_loop, lsp_bridge, dap_bridge, shell_keysNet: -63 lines, +20,181/-11,308 with no new files beyond the module splits. All 1,484 tests pass.
Test plan
make cipasses (fmt + clippy + build + test)make buildsucceeds (includes GUI)🤖 Generated with Claude Code