cats 0.1.21

Coding Agent ToolS - A comprehensive toolkit for building AI-powered coding agents
Documentation 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274

# Changelog

All notable changes to this project will be documented in this file.

The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/),
and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).

## [Unreleased]

## [0.1.21] - 2026-05-02

### Fixed
- **edit: "multiple matches" error now shows line numbers and context** — when
  `oldString` appears more than once the error now reports the count, line number
  of every occurrence, and 2–3 lines of surrounding context so the LLM can
  immediately construct a unique `oldString`. The message also explicitly warns
  against falling back to bash `>>` append (which causes duplicate content).

## [0.1.20] - 2026-04-29

### Fixed
- **executor: compact log shows `__truncated` marker instead of `{}`** — when tool call
  arguments fail JSON parsing (e.g. truncated by max_tokens), the compact log now records
  `{"__truncated": true, "__len": N}` so the log clearly shows a truncation occurred
  rather than silently displaying empty arguments.
- **executor: EOF parse errors produce actionable message** — when the LLM response is
  truncated mid-JSON (detected via `EOF while parsing`), the error returned to the LLM now
  reads: `"tool call was truncated (response too large — N bytes). Break the content into
  smaller pieces or use bash with heredoc."` Applies to all tools (`write`, `bash`,
  `edit`, etc.) via both `execute_tool_calls` and `execute_tool_calls_structured`.

## [0.1.19] - 2026-04-29

### Fixed
- **multiedit: atomic failure message** — when one edit in a `multiedit` batch fails, the error now
  explicitly states that no changes were written to disk and instructs the LLM to re-submit all edits
  with a corrected `oldString` rather than assuming earlier edits were applied.
- **edit/multiedit: oldString not found hint** — error message now tells the LLM to use the `read`
  tool to fetch exact current file content instead of constructing `oldString` from memory.

## [0.1.18] - 2026-04-29

### Fixed
- **grep: panic on multibyte characters at line truncation boundary** — `&line[..MAX_LINE_LENGTH]`
  sliced by byte index. Now uses `char_indices()` to find the nearest valid char boundary.
- **result_handler: panic on multibyte characters at result truncation boundary** — `&result_message[..max_size_bytes]`
  sliced by byte index. Now uses `char_indices()` to find the nearest valid char boundary.

## [0.1.17] - 2026-04-29

### Fixed
- **read: panic on multibyte characters at line truncation boundary** — `&line[..MAX_LINE_LENGTH]`
  sliced by byte index which could fall inside a multibyte UTF-8 character (e.g. em dash `—`).
  Now uses `char_indices()` to find the nearest valid char boundary before truncating.

## [0.1.16] - 2026-04-26

### Fixed
- **edit/read/write/multiedit: camelCase field names rejected by validate_args** — `filePath`, `oldString`,
  `newString` were silently blocked before reaching the `parse_*_args` functions that already handled them.
  Now `validate_args` accepts both snake_case and camelCase variants for all four tools.
- **multiedit: camelCase fields in edits array** — `EditOperation` deserialization failed with
  `missing field 'old_string'` when LLMs sent `oldString`/`newString`/`replaceAll`. Added
  `#[serde(alias)]` attributes to accept both naming conventions.
- **edit: CRLF line-ending corruption in fallback strategies** — `line_trimmed_replacer`,
  `whitespace_normalized_replacer`, and `trimmed_boundary_replacer` used `str::lines()` (strips `\r`)
  then reconstructed blocks with `join("\n")`, making `str::find` fail on CRLF files. Switched to
  `split('\n')` throughout so `\r` is preserved in reconstructed substrings.
- **edit: multiple-match early exit blocked fallback strategies** — `replace_in_content` returned
  `Err("Found multiple matches")` immediately when the exact string appeared more than once, before
  trying any fuzzy fallback. The error is now only raised after all strategies are exhausted without
  finding a unique match (mirrors OpenCode behaviour).
- **read: offset was 0-based instead of 1-based** — LLMs trained on OpenCode expect `offset=1` to
  mean "start at line 1", but cats treated it as "skip 1 line". Changed to 1-based: `skip = offset - 1`.

### Added
- **edit: EscapeNormalizedReplacer** — new fallback strategy that unescapes literal `\n`, `\t`, `\r`, `\"`
  sequences LLMs sometimes emit due to JSON double-escaping, then retries the replacement.
- **edit: IndentationFlexibleReplacer** — new fallback strategy that strips minimum common indentation
  from both the search string and candidate blocks before comparing, tolerating 2-space vs 4-space
  mismatches and copy-paste indentation drift.
- **edit: BlockAnchorReplacer** — new fallback strategy that locates blocks by matching trimmed
  first/last lines (anchors), then scores middle-line similarity using Levenshtein distance (via the
  existing `edit-distance` crate). Accepts the best candidate above a 30% threshold when multiple
  anchor matches exist, enabling fuzzy matching on slightly-stale code blocks.

## [0.1.15] - 2026-04-15

### Added
- **Cooperative Cancellation for Tool Execution**: Added `CancelSignal` (Arc<AtomicBool>) infrastructure so running tools (especially bash) can be killed immediately when the user presses ESC
  - New `CancelSignal` struct with `request_cancel()`, `reset()`, and `is_cancelled()` methods
  - Added `cancel_signal` field to `ToolState` with accessor methods (`request_cancel()`, `reset_cancel()`, `is_cancelled()`, `cancel_signal()`)
  - Re-exported `CancelSignal` from `lib.rs` for downstream consumers
  - BashTool now clones the cancel signal before spawning child processes to avoid holding the Mutex lock
  - Added cancel checks in BashTool stdout and stderr read loops — on cancel, kills child process and returns early
  - Added cancel check before starting stderr read (skip if already cancelled)
  - On cancel, returns a user-friendly result: "Command cancelled by user (ESC)" with exit_code -1

## [0.1.14] - 2026-04-15

### Fixed
- **Edit Tool OOB Panics**: Prevented out-of-bounds panics in `whitespace_normalized_replacer` and `trimmed_boundary_replacer` when dealing with edge cases in text replacement
- **Edit Tool Empty Lines Panic**: Prevented panic in `line_trimmed_replacer` when search_lines becomes empty after trailing newline trim

## [0.1.13] - 2026-04-15

### Fixed
- **Grep/Glob Tool Silent Failures**: Fixed silent failures when LLM sends empty string values for `path` and `pattern` parameters
  - Made `path` parameter required (non-optional) in both `grep` and `glob` tools, matching the `list` tool pattern
  - Added explicit `grep` and `glob` handlers in `converter.rs` that filter out empty string values
  - Updated `validate_args` to check for non-empty pattern values (not just presence)
  - Updated `parse_grep_args` and `parse_glob_args` to filter empty patterns and paths with `.filter(|s| !s.is_empty())`
  - Error messages now clearly state "path parameter is required. Send '.' for current directory" instead of confusing "pattern argument" errors
  - Affects GLM-5 and GLM-5-Turbo models which were sending `{"path":"","pattern":"..."}` causing the pattern to be lost during JSON deserialization

## [0.1.12] - 2026-06-01

### Fixed
- **Grep/Glob Tool Silent Failures**: Fixed silent failures when LLM sends empty string values for `path` and `pattern` parameters
  - Made `path` parameter required (non-optional) in both `grep` and `glob` tools, matching the `list` tool pattern
  - Added explicit `grep` and `glob` handlers in `converter.rs` that filter out empty string values
  - Updated `validate_args` to check for non-empty pattern values (not just presence)
  - Updated `parse_grep_args` and `parse_glob_args` to filter empty patterns and paths with `.filter(|s| !s.is_empty())`
  - Error messages now clearly state "path parameter is required. Send '.' for current directory" instead of confusing "pattern argument" errors
  - Affects GLM-5 and GLM-5-Turbo models which were sending `{"path":"","pattern":"..."}` causing the pattern to be lost during JSON deserialization

## [0.1.11] - 2026-03-10

### Fixed
- **List Tool Path Parameter**: Made `path` parameter required to prevent GLM-5 from sending null/empty values
  - Previously, optional path parameter caused issues with certain LLMs (GLM-5) that would send null or empty strings
  - Path is now a required parameter, ensuring explicit directory specification
  - Improves reliability when used with various LLM providers

### Changed
- **List Tool Debug Logging**: Replaced `tracing` crate with simple `debug!` macro for debug logging
  - Reduces dependency overhead for basic debug output
  - Simplifies the codebase by removing unnecessary tracing infrastructure

## [0.1.10] - 2026-03-10

### Fixed
- **Bash Tool Empty Workdir Parameter Bug**: Empty string workdir parameters are now treated as `None` instead of being processed as valid paths
  - Previously, when LLM provided `{"workdir": ""}`, the tool would fail with "No such file or directory" error
  - Added `.filter(|s| !s.is_empty())` in `parse_bash_args()` to convert empty strings to `None`
  - This prevents command execution failures and agent confusion when empty workdir is provided
  - Related to fix in 0.1.7 for list tool - same pattern applied to bash tool

## [0.1.9] - 2026-03-10

### Fixed
- **CLI Tool Argument Parsing**: Fixed command-line interface to properly parse JSON arguments
  - CLI now parses JSON arguments from strings like `{"path":"/home/leo"}` and converts them to named args
  - Previously, JSON strings were treated as positional args, causing tools to ignore them
  - This fix enables proper CLI usage: `cats list '{"path":"/home/leo"}'`
  - Combined with list tool JSON converter handler for complete fix

## [0.1.8] - 2026-03-10

### Fixed
- **JSON Converter for List Tool**: Added explicit handler for `list` tool in JSON-to-ToolArgs conversion
  - Previously, the `list` tool fell into the default case which could cause parameter conversion issues
  - Now properly handles empty path strings by filtering them out (treating empty as None)
  - Ensures robust parameter handling when LLM calls the list tool via function calling APIs
  - Related to fix in 0.1.7 - provides defense-in-depth for empty path parameter handling

## [0.1.7] - 2026-03-10

### Fixed
- **List Tool Empty Path Parameter Bug**: Empty string path parameters are now treated as `None` instead of being processed as valid paths
  - Previously, when LLM provided `{"path": ""}`, the tool would list the current working directory instead of the intended directory
  - Added `.filter(|s| !s.is_empty())` in `parse_list_args()` to convert empty strings to `None`
  - This prevents agent confusion and repeated tool calls when empty paths are provided
  - Fixes issue where agent would appear "stuck" repeatedly listing the wrong directory

## [0.1.6] - 2026-02-27

### Changed
- **TLS Backend**: Switched from native-tls to rustls for cross-compilation support
  - `reqwest` now uses `rustls-tls-webpki-roots` instead of default native-tls
  - Enables static linking with musl for portable binaries
  - No OpenSSL dependency required

## [0.1.5] - 2026-02-23

### Fixed
- **Working Directory Resolution**: Tools now read working directory from `ToolState` at execution time instead of capturing it at instantiation
  - This fixes the issue where tools would use the wrong directory when invoked from a different context
  - Mirrors OpenCode's approach using async context
  - Affected tools: `bash`, `read`, `write`, `edit`, `multiedit`, `glob`, `grep`, `list`

### Added
- `ToolRegistry::set_working_directory()` - Set the working directory for all tools
- `ToolRegistry::get_working_directory()` - Get the current working directory from tool state

## [0.1.4] - 2026-02-21

### Changed
- **BREAKING**: Default feature changed from `old` to `opencode`
- `create_tool_registry_with_open_window_size()` now takes `Option<usize>` instead of `usize`

### Added
- OpenCode-compatible tool set is now the default:
  - `bash` - Execute shell commands
  - `read` - Read file contents with windowing
  - `write` - Write content to files
  - `edit` - Edit files with str_replace
  - `glob` - Find files by pattern
  - `grep` - Search file contents
  - `ls` - List directory contents
  - `webfetch` - Fetch web content (optional, requires `reqwest`)
  - `question` - Ask user questions
  - `task` - Launch sub-agents (placeholder)
  - `todo` - Todo list management
  - `patch` - Patch files (placeholder)
  - `lsp` - LSP integration (placeholder)
  - `skill` - Skill execution (placeholder)

### Removed
- `old` toolset is no longer the default (still available via `default = ["old"]` feature)

## [0.1.1] - 2025-10-29

### Fixed
- CLI now dynamically lists all available tools from registry instead of hardcoded subset
- Removed non-existent tools (`edit`, `insert`) from CLI help output
- Added missing tools to CLI help: `run_command`, all editing tools (`create_file`, `replace_text`, `insert_text`, `delete_text`, `delete_line`, `overwrite_file`, `delete_function`), file management tools (`delete_path`, `move_path`, `copy_path`, `create_directory`), and `classify_task`
- CLI help output now stays automatically synchronized with the tool registry

## [0.1.0] - 2025-10-26

### Added
- Initial public release of CATS (Coding Agent ToolS)
- Extracted from Simpaticoder monorepo as independent crate
- Complete tool suite for AI coding agents:
  - File Navigation tools: `open`, `goto`, `scroll_up`, `scroll_down`
  - Search tools: `find_file`, `search_file`, `search_dir`
  - Editing tools: `create_file`, `replace_text`, `insert_text`, `delete_text`, `delete_line`, `overwrite_file`, `delete_function`
  - File Management tools: `delete_path`, `move_path`, `copy_path`, `create_directory`
  - Execution tools: `run_command`
  - Utility tools: `_state`, `filemap`, `submit`, `classify_task`
  - Optional `count_tokens` tool (requires `tiktoken` feature)
- Comprehensive documentation and examples
- Dual-license: MIT OR Apache-2.0
- CI/CD pipeline for automated testing
- Cross-platform support (Linux, macOS, Windows)

### Changed
- Renamed from `simpaticoder-tools` to `cats`
- Updated all internal references and documentation
- Published to crates.io as standalone crate
- Repository: https://github.com/podtan/cats

### Notes
- This is the first independent release
- Formerly part of the Simpaticoder project
- API surface is stable and production-ready
- No backward compatibility with `simpaticoder-tools` crate name

[Unreleased]: https://github.com/podtan/cats/compare/v0.1.15...HEAD
[0.1.15]: https://github.com/podtan/cats/compare/v0.1.14...v0.1.15
[0.1.14]: https://github.com/podtan/cats/compare/v0.1.13...v0.1.14
[0.1.13]: https://github.com/podtan/cats/compare/v0.1.12...v0.1.13
[0.1.12]: https://github.com/podtan/cats/compare/v0.1.11...v0.1.12
[0.1.11]: https://github.com/podtan/cats/compare/v0.1.10...v0.1.11
[0.1.10]: https://github.com/podtan/cats/compare/v0.1.9...v0.1.10
[0.1.9]: https://github.com/podtan/cats/compare/v0.1.8...v0.1.9
[0.1.8]: https://github.com/podtan/cats/compare/v0.1.7...v0.1.8
[0.1.7]: https://github.com/podtan/cats/compare/v0.1.6...v0.1.7
[0.1.6]: https://github.com/podtan/cats/releases/tag/v0.1.6
[0.1.5]: https://github.com/podtan/cats/releases/tag/v0.1.5
[0.1.4]: https://github.com/podtan/cats/releases/tag/v0.1.4
[0.1.1]: https://github.com/podtan/cats/releases/tag/v0.1.1
[0.1.0]: https://github.com/podtan/cats/releases/tag/v0.1.0