1use crate::agent::inference::{tool_metadata_for_name, ToolDefinition, ToolFunction};
2use serde_json::Value;
3
4fn make_tool(name: &str, description: &str, parameters: Value) -> ToolDefinition {
5 ToolDefinition {
6 tool_type: "function".into(),
7 function: ToolFunction {
8 name: name.into(),
9 description: description.into(),
10 parameters,
11 },
12 metadata: tool_metadata_for_name(name),
13 }
14}
15
16pub fn get_tools() -> Vec<ToolDefinition> {
18 let os = std::env::consts::OS;
19 let mut tools = vec![
20 make_tool(
21 "shell",
22 &format!(
23 "Execute a command in the host shell ({os}). \
24 Use this ONLY for building, testing, or advanced system operations that have no dedicated Hematite tool. \
25 FORBIDDEN: Never use shell to run `mkdir`, `rm`, `cat`, `head`, `tail`, or `write-file` equivalents. \
26 Use the dedicated surgical tools (create_directory, read_file, tail_file) instead. \
27 Output is capped at 64KB. Prefer non-interactive commands."
28 ),
29 serde_json::json!({
30 "type": "object",
31 "properties": {
32 "command": {
33 "type": "string",
34 "description": "The command to run"
35 },
36 "reason": {
37 "type": "string",
38 "description": "For risky shell calls, explain what this command is verifying or changing."
39 },
40 "timeout_secs": {
41 "type": "integer",
42 "description": "Optional timeout in seconds (default 60)"
43 }
44 },
45 "required": ["command"]
46 }),
47 ),
48 make_tool(
49 "run_code",
50 "Execute a short JavaScript/TypeScript or Python snippet in a sandboxed subprocess. \
51 No network access, no filesystem escape, hard 10-second timeout. \
52 Use this to verify logic, test algorithms, compute values, or test functions \
53 when you need real output rather than a guess. \
54 ALWAYS include the `language` field — there is no default. \
55 \
56 JAVASCRIPT/TYPESCRIPT (language: \"javascript\"): \
57 Runs via Deno, NOT Node.js. `require()` does not exist — never use it. \
58 URL imports (e.g. from 'https://deno.land/...') are blocked — network is off. \
59 Use built-in Web APIs only: `crypto.subtle`, `TextEncoder`, `URL`, `atob`/`btoa`, etc. \
60 SHA-256 example: \
61 const buf = await crypto.subtle.digest('SHA-256', new TextEncoder().encode('hello')); \
62 console.log([...new Uint8Array(buf)].map(b=>b.toString(16).padStart(2,'0')).join('')); \
63 \
64 PYTHON (language: \"python\"): \
65 Standard library is available. `hashlib`, `json`, `math`, `datetime`, `re`, `itertools` all work. \
66 `subprocess`, `socket`, `urllib`, `requests` are blocked. \
67 SHA-256 example: import hashlib; print(hashlib.sha256(b'hello').hexdigest()) \
68 \
69 Do NOT use this tool for PowerShell or shell scripting. This is strictly for high-precision computation in JavaScript, TypeScript, or Python only. \
70 Do NOT fall back to shell to run deno, python, or node — use this tool directly.",
71 serde_json::json!({
72 "type": "object",
73 "properties": {
74 "language": {
75 "type": "string",
76 "enum": ["javascript", "typescript", "python"],
77 "description": "The language to run. javascript/typescript requires Deno; python requires Python 3."
78 },
79 "code": {
80 "type": "string",
81 "description": "The code to execute. Keep it short and self-contained. Print results to stdout."
82 },
83 "timeout_seconds": {
84 "type": "integer",
85 "description": "Max execution time in seconds (default 10, max 60). Use higher values for longer computations."
86 }
87 },
88 "required": ["language", "code"]
89 }),
90 ),
91
92 make_tool(
93 "trace_runtime_flow",
94 "Return an authoritative read-only trace of Hematite runtime flow. \
95 Use this for architecture questions about keyboard input to final output, \
96 reasoning/specular separation, startup wiring, runtime subsystems, \
97 voice synthesis and Ctrl+T toggle, or \
98 session reset commands like /clear, /new, and /forget. Prefer this over guessing.",
99 serde_json::json!({
100 "type": "object",
101 "properties": {
102 "topic": {
103 "type": "string",
104 "enum": ["user_turn", "session_reset", "reasoning_split", "runtime_subsystems", "startup", "voice"],
105 "description": "Which verified runtime report to return. Use 'voice' for any question about Ctrl+T, voice toggle, or TTS pipeline. Use 'user_turn' for keyboard-to-output flow. Use 'session_reset' for /clear, /forget, /new. Use 'startup' for startup wiring. Use 'reasoning_split' for specular/thought routing. Use 'runtime_subsystems' for background subsystem overview."
106 },
107 "input": {
108 "type": "string",
109 "description": "Optional user input to label a normal user-turn trace"
110 },
111 "command": {
112 "type": "string",
113 "enum": ["/clear", "/new", "/forget", "all"],
114 "description": "Optional reset command when topic=session_reset"
115 }
116 },
117 "required": ["topic"]
118 }),
119 ),
120 make_tool(
121 "describe_toolchain",
122 "Return an authoritative read-only description of Hematite's actual tool surface and investigation strategy. \
123 Use this for tooling-discipline questions, best-tool selection, or read-only plans for tracing runtime behavior. \
124 Prefer this over improvising tool names or investigation steps from memory.",
125 serde_json::json!({
126 "type": "object",
127 "properties": {
128 "topic": {
129 "type": "string",
130 "enum": ["read_only_codebase", "user_turn_plan", "voice_latency_plan", "host_inspection_plan", "all"],
131 "description": "Which authoritative toolchain report to return"
132 },
133 "question": {
134 "type": "string",
135 "description": "Optional user question to label or tailor the read-only investigation plan"
136 }
137 }
138 }),
139 ),
140 make_tool(
141 "inspect_host",
142 "Return a structured read-only inspection of the current machine and environment. \
143 Prefer this over raw shell for questions about OS configuration (firewall, power, uptime), plain-English system health reports, silicon health and high-fidelity hardware telemetry (NVIDIA clocks/fans/power, CPU frequency averaging), installed developer tools, PATH issues, package-manager and environment health, network state, service state, running processes, desktop items, Downloads size, listening ports, repo health, or directory/disk summaries. \
144 For high-performance hardware testing, use topic=disk_benchmark to measure real-time kernel disk queue intensity. \
145 For remediation questions phrased like 'how do I fix cargo not found', 'how do I fix port 3000 already in use', or 'how do I fix LM Studio not reachable', use topic=fix_plan instead of diagnosis-only topics like env_doctor, path, or ports. \
146 Use topic=summary for a compact host snapshot, topic=toolchains for common dev tool versions, topic=path for PATH analysis, topic=env_doctor for package-manager and PATH health, topic=fix_plan for structured remediation plans, topic=network for adapters/IPs/gateways/DNS, topic=services for service status and startup mode, \
147 topic=processes for top processes by memory/cpu and real-time disk/network I/O stats (look for [I/O R:N/W:N] tags to identify disk-heavy processes), \
148 topic=desktop or topic=downloads for known folders, topic=ports for listening endpoints, topic=repo_doctor for a structured workspace health report, \
149 topic=log_check for recent critical/error events from system event logs or journalctl, topic=startup_items for programs and services that run at boot (registry Run keys and startup folders on Windows; systemd enabled units on Linux), \
150 topic=health_report for a plain-English tiered system health verdict (disk, RAM, tools, recent errors), \
151 topic=storage for all drives with capacity/free space plus large developer cache directories, \
152 topic=hardware for CPU model/cores, RAM size/speed, GPU name/driver, motherboard, BIOS, and display configuration, \
153 topic=updates for Windows Update status (last install date, pending update count, WU service state), \
154 topic=security for Windows Defender real-time protection status, last scan date, signature age, firewall profile states, Windows activation, and UAC state, \
155 topic=pending_reboot to check whether a system restart is required and why (Windows Update, CBS, file rename operations), \
156 topic=disk_health for physical drive health via Get-PhysicalDisk and SMART failure prediction, \
157 topic=battery for charge level, status, estimated runtime, and wear level (laptops only — reports no battery on desktops), \
158 topic=recent_crashes for BSOD and unexpected shutdown events plus application crash/hang events from the Windows event log, \
159 topic=scheduled_tasks for all non-disabled scheduled tasks including name, path, last run time, and executable, \
160 topic=dev_conflicts for cross-tool environment conflict detection (Node.js version managers, Python 2 vs 3 ambiguity, conda env shadowing, Rust toolchain path conflicts, Git identity/signing config, duplicate PATH entries), \
161 topic=bitlocker for drive encryption status (BitLocker on Windows, LUKS on Linux), \
162 topic=ad_user for Active Directory / Managed Identity details (SID, group memberships, domain role), \
163 topic=user_accounts for Local User and Group diagnostics (Built-in Administrators, local account state), \
164 topic=rdp for Remote Desktop configuration, port, and active sessions, \
165 topic=shadow_copies for Volume Shadow Copies (VSS) and system restore points, \
166 topic=pagefile for Windows page file configuration and current usage, \
167 topic=windows_features for enabled Windows optional features (IIS, Hyper-V, etc.), \
168 topic=printers for installed printers and active print jobs, \
169 topic=winrm for Windows Remote Management (WinRM) and PS Remoting status, \
170 topic=network_stats for adapter throughput (RX/TX), errors, and dropped packets, \
171 topic=udp_ports for active UDP listeners and notable port annotations, \
172 topic=gpo for applied Group Policy Objects, topic=certificates for local personal certificates, topic=integrity for Windows component store health (SFC/DISM state), topic=domain for Active Directory and domain join status, \
173 topic=device_health for identifying malfunctioning hardware with ConfigManager error codes (Yellow Bangs), topic=drivers for auditing active system drivers and their states, topic=peripherals for enumerating connected USB, input, and display hardware, \
174 topic=sessions for auditing active and disconnected user logon sessions, \
175 topic=ad_user for specific Active Directory user identity, SID, and group membership auditing, \
176 topic=dns_lookup for precision DNS record queries (SRV, MX, TXT), \
177 topic=hyperv for local Hyper-V VM inventory and real-time load, \
178 topic=ip_config for detailed adapter configuration and DHCP lease state, \
179 topic=disk_benchmark for high-performance silicon-aware stress testing, \
180 and topic=directory or topic=disk for arbitrary paths.",
181 serde_json::json!({
182 "type": "object",
183 "properties": {
184 "topic": {
185 "type": "string",
186 "enum": ["summary", "toolchains", "path", "env_doctor", "fix_plan", "network", "services", "processes", "desktop", "downloads", "directory", "disk", "ports", "repo_doctor", "log_check", "startup_items", "health_report", "storage", "hardware", "updates", "security", "pending_reboot", "disk_health", "battery", "recent_crashes", "scheduled_tasks", "dev_conflicts", "os_config", "bitlocker", "rdp", "shadow_copies", "pagefile", "windows_features", "printers", "winrm", "network_stats", "udp_ports", "gpo", "certificates", "integrity", "domain", "device_health", "drivers", "peripherals", "disk_benchmark", "permissions", "login_history", "registry_audit", "share_access", "thermal", "activation", "patch_history", "ad_user", "dns_lookup", "hyperv", "ip_config"],
187 "description": "Which structured host inspection to run. Use topic=ad_user for domain identity audit, topic=dns_lookup for SRV/MX records, topic=hyperv for VM load, and topic=ip_config for detailed adapter info."
188 },
189 "name": {
190 "type": "string",
191 "description": "Optional when topic=processes or topic=services. Case-insensitive substring filter for process or service names."
192 },
193 "issue": {
194 "type": "string",
195 "description": "Optional when topic=fix_plan. Plain-English issue description such as 'cargo not found', 'port 3000 already in use', or 'LM Studio not reachable on localhost:1234'."
196 },
197 "path": {
198 "type": "string",
199 "description": "Required when topic=directory. Optional for topic=disk or topic=repo_doctor. Absolute or relative path to inspect."
200 },
201 "port": {
202 "type": "integer",
203 "description": "Optional when topic=ports or topic=fix_plan. Filter the result to one listening TCP port or anchor a port-conflict fix plan."
204 },
205 "max_entries": {
206 "type": "integer",
207 "description": "Optional cap for listed entries. Defaults to 10 and is capped internally."
208 }
209 }
210 }),
211 ),
212 make_tool(
213 "resolve_host_issue",
214 "A safe, bounded tool for remediating OS and environment issues automatically with user approval. \
215 Use this to fix missing dependencies, restart stuck services, or clear disk space instead of using raw shell. \
216 The user will be prompted to approve the action. Keep targets exact.",
217 serde_json::json!({
218 "type": "object",
219 "properties": {
220 "action": {
221 "type": "string",
222 "enum": ["install_package", "restart_service", "clear_temp"],
223 "description": "The type of remediation to perform."
224 },
225 "target": {
226 "type": "string",
227 "description": "The specific target (e.g., 'python' for install_package, or 'docker' for restart_service). Optional for clear_temp."
228 }
229 },
230 "required": ["action"]
231 }),
232 ),
233 make_tool(
234 "run_hematite_maintainer_workflow",
235 "Run one of Hematite's known maintainer or release workflows with explicit approval. \
236 Prefer this over raw shell when the user explicitly asks to run one of Hematite's own scripts such as `clean.ps1`, `scripts/package-windows.ps1`, or `release.ps1`. \
237 Use workflow=clean for cleanup, workflow=package_windows for rebuilding the local Windows portable or installer, and workflow=release for the normal version bump/tag/push/publish flow. \
238 Keep this tool constrained to Hematite's own known workflows instead of inventing ad hoc shell commands or pretending to run arbitrary project scripts.",
239 serde_json::json!({
240 "type": "object",
241 "properties": {
242 "workflow": {
243 "type": "string",
244 "enum": ["clean", "package_windows", "release"],
245 "description": "Which known Hematite maintainer workflow to run."
246 },
247 "deep": {
248 "type": "boolean",
249 "description": "For workflow=clean. Also remove heavy build/runtime artifacts such as target/ and vein.db."
250 },
251 "reset": {
252 "type": "boolean",
253 "description": "For workflow=clean. Reset PLAN/TASK state in addition to normal cleanup."
254 },
255 "prune_dist": {
256 "type": "boolean",
257 "description": "For workflow=clean. Keep only the current Cargo.toml version under dist/."
258 },
259 "installer": {
260 "type": "boolean",
261 "description": "For workflow=package_windows. Also build the Windows installer."
262 },
263 "add_to_path": {
264 "type": "boolean",
265 "description": "For workflow=package_windows or workflow=release. Update the user PATH to the rebuilt portable."
266 },
267 "version": {
268 "type": "string",
269 "description": "For workflow=release. Exact semantic version such as 0.4.5."
270 },
271 "bump": {
272 "type": "string",
273 "enum": ["patch", "minor", "major"],
274 "description": "For workflow=release. Ask release.ps1 to calculate the next version."
275 },
276 "push": {
277 "type": "boolean",
278 "description": "For workflow=release. Push main and the new tag."
279 },
280 "skip_installer": {
281 "type": "boolean",
282 "description": "For workflow=release. Skip the Windows installer build."
283 },
284 "publish_crates": {
285 "type": "boolean",
286 "description": "For workflow=release. Publish hematite-cli to crates.io after a successful push."
287 },
288 "publish_voice_crate": {
289 "type": "boolean",
290 "description": "For workflow=release. Publish hematite-kokoros first, then hematite-cli."
291 }
292 },
293 "required": ["workflow"]
294 }),
295 ),
296 make_tool(
297 "run_workspace_workflow",
298 "Run an approval-gated workflow or script in the locked project workspace root. \
299 Use this for the current project's build, test, lint, fix, package.json scripts, just/task/make targets, explicit local script paths, or an exact workspace command. \
300 FORBIDDEN: The `command` field MUST be a real executable shell command (e.g. `npm install`, `cargo build`). \
301 NEVER put natural language, user-requests, or conversational intent into the `command` field. \
302 This tool is for the active workspace, not for Hematite's own maintainer scripts.",
303 serde_json::json!({
304 "type": "object",
305 "properties": {
306 "workflow": {
307 "type": "string",
308 "enum": ["build", "test", "lint", "fix", "package_script", "task", "just", "make", "script_path", "command"],
309 "description": "Which workspace workflow to run."
310 },
311 "name": {
312 "type": "string",
313 "description": "Required for workflow=package_script, task, just, or make. The script or target name."
314 },
315 "path": {
316 "type": "string",
317 "description": "Required for workflow=script_path. Relative path to a script inside the locked workspace root."
318 },
319 "command": {
320 "type": "string",
321 "description": "Required for workflow=command. Exact command to execute from the locked workspace root."
322 },
323 "timeout_ms": {
324 "type": "integer",
325 "description": "Optional timeout override in milliseconds."
326 }
327 },
328 "required": ["workflow"]
329 }),
330 ),
331 make_tool(
332 "read_file",
333 "Read the contents of a file. For large files, use 'offset' and 'limit' to navigate.",
334 serde_json::json!({
335 "type": "object",
336 "properties": {
337 "path": {
338 "type": "string",
339 "description": "Path to the file, relative to the project root"
340 },
341 "offset": {
342 "type": "integer",
343 "description": "Starting line number (0-indexed)"
344 },
345 "limit": {
346 "type": "integer",
347 "description": "Number of lines to read"
348 }
349 },
350 "required": ["path"]
351 }),
352 ),
353 make_tool(
354 "lsp_definitions",
355 "Get the precise definition location (file:line:char) for a symbol at a specific position. \
356 Use this to jump to function/struct source code accurately.",
357 serde_json::json!({
358 "type": "object",
359 "properties": {
360 "path": { "type": "string", "description": "File path" },
361 "line": { "type": "integer", "description": "0-indexed line" },
362 "character": { "type": "integer", "description": "0-indexed character" }
363 },
364 "required": ["path", "line", "character"]
365 }),
366 ),
367 make_tool(
368 "lsp_references",
369 "Find all locations where a symbol is used across the entire workspace. \
370 Use this to understand the impact of a refactor or discover internal API users.",
371 serde_json::json!({
372 "type": "object",
373 "properties": {
374 "path": { "type": "string", "description": "File path" },
375 "line": { "type": "integer", "description": "0-indexed line" },
376 "character": { "type": "integer", "description": "0-indexed character" }
377 },
378 "required": ["path", "line", "character"]
379 }),
380 ),
381 make_tool(
382 "lsp_hover",
383 "Get hover information (documentation, function signature, type details) for a symbol. \
384 Use this for rapid spatial awareness without opening every file.",
385 serde_json::json!({
386 "type": "object",
387 "properties": {
388 "path": { "type": "string", "description": "File path" },
389 "line": { "type": "integer", "description": "0-indexed line" },
390 "character": { "type": "integer", "description": "0-indexed character" }
391 },
392 "required": ["path", "line", "character"]
393 }),
394 ),
395 make_tool(
396 "lsp_rename_symbol",
397 "Rename a symbol project-wide using the Language Server. Ensures all references are updated safely.",
398 serde_json::json!({
399 "type": "object",
400 "properties": {
401 "path": { "type": "string", "description": "File path" },
402 "line": { "type": "integer", "description": "0-indexed line" },
403 "character": { "type": "integer", "description": "0-indexed character" },
404 "new_name": { "type": "string", "description": "The new name for the symbol" }
405 },
406 "required": ["path", "line", "character", "new_name"]
407 }),
408 ),
409 make_tool(
410 "lsp_get_diagnostics",
411 "Get a list of current compiler errors and warnings for a specific file. \
412 Use this to verify your code compiles and and to find exactly where errors are located.",
413 serde_json::json!({
414 "type": "object",
415 "properties": {
416 "path": { "type": "string", "description": "File path" }
417 },
418 "required": ["path"]
419 }),
420 ),
421 make_tool(
422 "vision_analyze",
423 "Send an image file (screenshot, diagram, or UI mockup) to the multimodal vision model for technical analysis. \
424 Use this to identify UI bugs, confirm visual states, or understand architectural diagrams.",
425 serde_json::json!({
426 "type": "object",
427 "properties": {
428 "path": { "type": "string", "description": "Absolute or relative path to the image file." },
429 "prompt": { "type": "string", "description": "The specific question or analysis request for the vision model." }
430 },
431 "required": ["path", "prompt"]
432 }),
433 ),
434 make_tool(
435 "patch_hunk",
436 "Replace a specific line range [start_line, end_line] with new content. \
437 This is the most precise way to edit code and avoids search string failures.",
438 serde_json::json!({
439 "type": "object",
440 "properties": {
441 "path": { "type": "string", "description": "File path" },
442 "start_line": { "type": "integer", "description": "Starting line (1-indexed)" },
443 "end_line": { "type": "integer", "description": "Ending line (inclusive)" },
444 "replacement": { "type": "string", "description": "The new content for this range" }
445 },
446 "required": ["path", "start_line", "end_line", "replacement"]
447 }),
448 ),
449 make_tool(
450 "multi_search_replace",
451 "Replace multiple existing code blocks in a single file with new content. \
452 Each hunk specifies an EXACT 'search' string and a 'replace' string. \
453 The 'search' string MUST exactly match the existing file contents (including whitespace). \
454 This is the safest and most reliable way to make multiple structural edits.",
455 serde_json::json!({
456 "type": "object",
457 "properties": {
458 "path": { "type": "string", "description": "File path" },
459 "hunks": {
460 "type": "array",
461 "items": {
462 "type": "object",
463 "properties": {
464 "search": { "type": "string", "description": "Exact existing text to find and replace" },
465 "replace": { "type": "string", "description": "The new replacement text" }
466 },
467 "required": ["search", "replace"]
468 }
469 }
470 },
471 "required": ["path", "hunks"]
472 }),
473 ),
474 make_tool(
475 "write_file",
476 "Write content to a file, creating it (and any parent dirs) if needed. \
477 Overwrites existing files. \
478 SOVEREIGN PATHING: For files in common areas, use `@DESKTOP/file.txt`, `@DOCUMENTS/file.txt`, `@DOWNLOADS/file.txt`, or `@HOME/file.txt` to ensure 100% path accuracy.",
479 serde_json::json!({
480 "type": "object",
481 "properties": {
482 "path": { "type": "string", "description": "File path" },
483 "content": { "type": "string", "description": "Full file content to write" }
484 },
485 "required": ["path", "content"]
486 }),
487 ),
488 make_tool(
489 "create_directory",
490 "Authoritatively create a new directory (and any parent dirs) if they do not exist. \
491 Use this instead of raw shell (mkdir) for all filesystem organization. \
492 Supports both relative paths and absolute paths. \
493 SOVEREIGN PATHING: For directories in common areas, use `@DESKTOP/folder`, `@DOCUMENTS/folder`, `@DOWNLOADS/folder`, or `@HOME/folder` to ensure 100% path accuracy.",
494 serde_json::json!({
495 "type": "object",
496 "properties": {
497 "path": { "type": "string", "description": "Relative or absolute directory path" }
498 },
499 "required": ["path"]
500 }),
501 ),
502 make_tool(
503 "research_web",
504 "Perform a zero-cost technical search using DuckDuckGo. \
505 Use this to find documentation, latest API changes, or solutions to complex errors \
506 when your internal knowledge is insufficient. Returns snippets and URLs.",
507 serde_json::json!({
508 "type": "object",
509 "properties": {
510 "query": { "type": "string", "description": "The technical search query" }
511 },
512 "required": ["query"]
513 }),
514 ),
515 make_tool(
516 "fetch_docs",
517 "Fetch a URL and convert it to clean Markdown. Use this to 'read' the documentation \
518 links found via research_web. This tool uses a proxy to bypass IP blocks.",
519 serde_json::json!({
520 "type": "object",
521 "properties": {
522 "url": { "type": "string", "description": "The URL of the documentation to fetch" }
523 },
524 "required": ["url"]
525 }),
526 ),
527 make_tool(
528 "edit_file",
529 "Edit a file by replacing an exact string with another. \
530 The 'search' string does NOT need perfectly matching indentation (it is fuzzy), \
531 but the non-whitespace text must match exactly. Use this for targeted edits.",
532 serde_json::json!({
533 "type": "object",
534 "properties": {
535 "path": { "type": "string", "description": "File path" },
536 "search": {
537 "type": "string",
538 "description": "The exact text to find (must match whitespace/indentation precisely)"
539 },
540 "replace": {
541 "type": "string",
542 "description": "The replacement text"
543 }
544 },
545 "required": ["path", "search", "replace"]
546 }),
547 ),
548 make_tool(
549 "auto_pin_context",
550 "Select 1-3 core files to 'Lock' into prioritized memory. \
551 Use this to ensure the most important architecture files \
552 are always visible during complex refactorings.",
553 serde_json::json!({
554 "type": "object",
555 "properties": {
556 "paths": {
557 "type": "array",
558 "items": { "type": "string" }
559 },
560 "reason": { "type": "string" }
561 },
562 "required": ["paths", "reason"]
563 }),
564 ),
565 make_tool(
566 "list_pinned",
567 "List all files currently pinned in the model's active context.",
568 serde_json::json!({
569 "type": "object",
570 "properties": {}
571 }),
572 ),
573 make_tool(
574 "list_files",
575 "List files in a directory, optionally filtered by extension.",
576 serde_json::json!({
577 "type": "object",
578 "properties": {
579 "path": {
580 "type": "string",
581 "description": "Directory to list (default: current dir)"
582 },
583 "extension": {
584 "type": "string",
585 "description": "Only return files with this extension, e.g. 'rs', 'toml' (no dot)"
586 }
587 },
588 "required": []
589 }),
590 ),
591 make_tool(
592 "tail_file",
593 "Read the last N lines of a file — useful for log files, test output, \
594 build artifacts, and any large file where only the tail is relevant. \
595 Supports an optional grep filter to show only matching lines from the tail. \
596 Use this instead of read_file when you only need the end of a large file.",
597 serde_json::json!({
598 "type": "object",
599 "properties": {
600 "path": {
601 "type": "string",
602 "description": "Path to the file, relative to the project root"
603 },
604 "lines": {
605 "type": "integer",
606 "description": "Number of lines to return from the end (default: 50, max: 500)"
607 },
608 "grep": {
609 "type": "string",
610 "description": "Optional regex pattern — only return lines matching this pattern (applied before the tail slice)"
611 }
612 },
613 "required": ["path"]
614 }),
615 ),
616 make_tool(
617 "grep_files",
618 "Search file contents for a regex pattern. Supports context lines, files-only mode, \
619 and pagination. Returns file:line:content format by default.",
620 serde_json::json!({
621 "type": "object",
622 "properties": {
623 "pattern": {
624 "type": "string",
625 "description": "Regex pattern to search for (case-insensitive by default)"
626 },
627 "path": {
628 "type": "string",
629 "description": "Directory to search (default: current dir)"
630 },
631 "extension": {
632 "type": "string",
633 "description": "Only search files with this extension, e.g. 'rs'"
634 },
635 "mode": {
636 "type": "string",
637 "enum": ["content", "files_only"],
638 "description": "'content' (default) returns matching lines; 'files_only' returns only filenames"
639 },
640 "context": {
641 "type": "integer",
642 "description": "Lines of context before AND after each match (like rg -C)"
643 },
644 "before": {
645 "type": "integer",
646 "description": "Lines of context before each match (overrides context)"
647 },
648 "after": {
649 "type": "integer",
650 "description": "Lines of context after each match (overrides context)"
651 },
652 "head_limit": {
653 "type": "integer",
654 "description": "Max hunks (or files in files_only) to return (default: 50)"
655 },
656 "offset": {
657 "type": "integer",
658 "description": "Skip first N hunks/files - for pagination (default: 0)"
659 }
660 },
661 "required": ["pattern"]
662 }),
663 ),
664 make_tool(
665 "git_commit",
666 "Stage all changes (git add -A) and create a commit. You MUST use 'Conventional Commits' (e.g. 'feat: description').",
667 serde_json::json!({
668 "type": "object",
669 "properties": {
670 "message": { "type": "string", "description": "Commit message (Conventional Commit style)" }
671 },
672 "required": ["message"]
673 }),
674 ),
675 make_tool(
676 "git_push",
677 "Push current branched changes to the remote origin. Requires an existing remote connection.",
678 serde_json::json!({
679 "type": "object",
680 "properties": {},
681 "required": []
682 }),
683 ),
684 make_tool(
685 "git_remote",
686 "View or manage git remotes. Use this for onboarding to GitHub/GitLab services.",
687 serde_json::json!({
688 "type": "object",
689 "properties": {
690 "action": {
691 "type": "string",
692 "enum": ["list", "add", "remove"],
693 "description": "Operation to perform"
694 },
695 "name": { "type": "string", "description": "Remote name (e.g. origin)" },
696 "url": { "type": "string", "description": "Remote URL (for 'add' action)" }
697 },
698 "required": ["action"]
699 }),
700 ),
701 make_tool(
702 "git_onboarding",
703 "High-level wizard to connect this repository to a remote host (GitHub/GitLab). \
704 Handles adding the remote and performing the initial tracking push in one step.",
705 serde_json::json!({
706 "type": "object",
707 "properties": {
708 "url": { "type": "string", "description": "The remote repository URL (HTTPS or SSH)" },
709 "name": { "type": "string", "description": "The remote name (default: origin)" },
710 "push": { "type": "boolean", "description": "Whether to perform an initial push to establish tracking (default: false)" }
711 },
712 "required": ["url"]
713 }),
714 ),
715 make_tool(
716 "verify_build",
717 "Run project verification for build, test, lint, or fix workflows. \
718 Prefer per-project verify profiles from `.hematite/settings.json`, and fall back to \
719 auto-detected defaults when no profile is configured. Returns BUILD OK or BUILD FAILED \
720 with command output. ALWAYS call this after scaffolding a new project or making structural changes.",
721 serde_json::json!({
722 "type": "object",
723 "properties": {
724 "action": {
725 "type": "string",
726 "enum": ["build", "test", "lint", "fix"],
727 "description": "Which verification action to run. Defaults to build."
728 },
729 "profile": {
730 "type": "string",
731 "description": "Optional named verify profile from `.hematite/settings.json`."
732 },
733 "timeout_secs": {
734 "type": "integer",
735 "description": "Optional timeout override for this verification run."
736 }
737 }
738 }),
739 ),
740 make_tool(
741 "git_worktree",
742 "Manage Git worktrees - isolated working directories on separate branches. \
743 Use 'add' to create a safe sandbox for risky/experimental work, \
744 'list' to see all worktrees, 'remove' to clean up, 'prune' to remove stale entries.",
745 serde_json::json!({
746 "type": "object",
747 "properties": {
748 "action": {
749 "type": "string",
750 "enum": ["list", "add", "remove", "prune"],
751 "description": "Worktree operation to perform"
752 },
753 "path": {
754 "type": "string",
755 "description": "Directory path for the new worktree (required for add/remove)"
756 },
757 "branch": {
758 "type": "string",
759 "description": "Branch name for the worktree (add only; defaults to path basename)"
760 }
761 },
762 "required": ["action"]
763 }),
764 ),
765 make_tool(
766 "clarify",
767 "Ask the user a clarifying question when you genuinely cannot proceed without \
768 more information. Use this ONLY when you are blocked and cannot make a \
769 reasonable assumption. Do NOT use it to ask permission - just act.",
770 serde_json::json!({
771 "type": "object",
772 "properties": {
773 "question": {
774 "type": "string",
775 "description": "The specific question to ask the user"
776 }
777 },
778 "required": ["question"]
779 }),
780 ),
781 make_tool(
782 "manage_tasks",
783 "Manage the persistent task ledger in .hematite/TASK.md. Use this to track long-term goals across restarts.",
784 crate::tools::tasks::get_tasks_params(),
785 ),
786 make_tool(
787 "maintain_plan",
788 "Document the architectural strategy and session blueprint in .hematite/PLAN.md. Use this to maintain context across restarts.",
789 crate::tools::plan::get_plan_params(),
790 ),
791 make_tool(
792 "generate_walkthrough",
793 "Generate a final session report in .hematite/WALKTHROUGH.md including achievements and verification results.",
794 crate::tools::plan::get_walkthrough_params(),
795 ),
796 make_tool(
797 "swarm",
798 "Delegate high-volume parallel tasks to a swarm of background workers. \
799 Use this for large-scale refactors, multi-file research, or parallel documentation updates. \
800 You must provide a 'tasks' array where each task has an 'id', 'target' (file), and 'instruction'.",
801 serde_json::json!({
802 "type": "object",
803 "properties": {
804 "tasks": {
805 "type": "array",
806 "items": {
807 "type": "object",
808 "properties": {
809 "id": { "type": "string" },
810 "target": { "type": "string", "description": "Target file or directory" },
811 "instruction": { "type": "string", "description": "Specific task for this worker" }
812 },
813 "required": ["id", "target", "instruction"]
814 }
815 },
816 "max_workers": {
817 "type": "integer",
818 "description": "Max parallel workers (default 3, auto-throttled by hardware)",
819 "default": 3
820 }
821 },
822 "required": ["tasks"]
823 }),
824 ),
825 ];
826
827 let lsp_defs = crate::tools::lsp_tools::get_lsp_definitions();
828 tools.push(make_tool(
829 "lsp_search_symbol",
830 "Find the location (file/line) of any function, struct, or variable in the entire project workspace. \
831 This is the fastest 'Golden Path' for navigating to a symbol by name.",
832 serde_json::json!({
833 "type": "object",
834 "properties": {
835 "query": { "type": "string", "description": "The name of the symbol to find (e.g. 'initialize_mcp')" }
836 },
837 "required": ["query"]
838 }),
839 ));
840 for def in lsp_defs {
841 let name = def["name"].as_str().unwrap();
842 tools.push(ToolDefinition {
843 tool_type: "function".into(),
844 function: ToolFunction {
845 name: name.into(),
846 description: def["description"].as_str().unwrap().into(),
847 parameters: def["parameters"].clone(),
848 },
849 metadata: tool_metadata_for_name(name),
850 });
851 }
852
853 tools
854}
855
856pub async fn dispatch_builtin_tool(name: &str, args: &Value) -> Result<String, String> {
857 match name {
858 "shell" => crate::tools::shell::execute(args).await,
859 "run_code" => crate::tools::code_sandbox::execute(args).await,
860 "trace_runtime_flow" => crate::tools::runtime_trace::trace_runtime_flow(args).await,
861 "describe_toolchain" => crate::tools::toolchain::describe_toolchain(args).await,
862 "inspect_host" => crate::tools::host_inspect::inspect_host(args).await,
863 "resolve_host_issue" => crate::tools::host_inspect::resolve_host_issue(args).await,
864 "run_hematite_maintainer_workflow" => {
865 crate::tools::repo_script::run_hematite_maintainer_workflow(args).await
866 }
867 "run_workspace_workflow" => crate::tools::workspace_workflow::run_workspace_workflow(args).await,
868 "read_file" => crate::tools::file_ops::read_file(args).await,
869 "inspect_lines" => crate::tools::file_ops::inspect_lines(args).await,
870 "tail_file" => crate::tools::file_ops::tail_file(args).await,
871 "write_file" => crate::tools::file_ops::write_file(args).await,
872 "create_directory" => crate::tools::file_ops::create_directory(args).await,
873 "edit_file" => crate::tools::file_ops::edit_file(args).await,
874 "patch_hunk" => crate::tools::file_ops::patch_hunk(args).await,
875 "multi_search_replace" => crate::tools::file_ops::multi_search_replace(args).await,
876 "list_files" => crate::tools::file_ops::list_files(args).await,
877 "grep_files" => crate::tools::file_ops::grep_files(args).await,
878 "git_commit" => crate::tools::git::execute(args).await,
879 "git_push" => crate::tools::git::execute_push(args).await,
880 "git_remote" => crate::tools::git::execute_remote(args).await,
881 "git_onboarding" => crate::tools::git_onboarding::execute(args).await,
882 "verify_build" => crate::tools::verify_build::execute(args).await,
883 "git_worktree" => crate::tools::git::execute_worktree(args).await,
884 "health" => crate::tools::health::execute(args).await,
885 "research_web" => crate::tools::research::execute_search(args).await,
886 "fetch_docs" => crate::tools::research::execute_fetch(args).await,
887 "manage_tasks" => crate::tools::tasks::manage_tasks(args).await,
888 "maintain_plan" => crate::tools::plan::maintain_plan(args).await,
889 "generate_walkthrough" => crate::tools::plan::generate_walkthrough(args).await,
890 "clarify" => {
891 let q = args.get("question").and_then(|v| v.as_str()).unwrap_or("?");
892 Ok(format!("[clarify] {q}"))
893 }
894 "vision_analyze" => Err(
895 "Tool 'vision_analyze' must be dispatched by ConversationManager (it requires hardware engine access)."
896 .into(),
897 ),
898 other => {
899 if other.contains('.') || other.contains('/') || other.contains('\\') {
900 Err(format!(
901 "'{}' is a PATH, not a tool. You correctly identified the location, but you MUST use `read_file` or `list_files` (internal) or `powershell` (external) to access it.",
902 other
903 ))
904 } else if matches!(other.to_lowercase().as_str(), "hematite" | "assistant" | "ai") {
905 Err(format!(
906 "'{}' is YOUR IDENTITY, not a tool. Use list_files or read_file to explore the codebase.",
907 other
908 ))
909 } else if matches!(
910 other.to_lowercase().as_str(),
911 "thought" | "think" | "reasoning" | "thinking" | "internal"
912 ) {
913 Err(format!(
914 "'{}' is NOT a tool - it is a reasoning tag. Output your answer as plain text after your <think> block.",
915 other
916 ))
917 } else {
918 Err(format!("Unknown tool: '{}'", other))
919 }
920 }
921 }
922}
923
924pub fn get_mutation_label(name: &str, args: &Value) -> Option<String> {
925 match name {
926 "shell" => {
927 let cmd = args.get("command").and_then(|v| v.as_str()).unwrap_or("");
928 if cmd.contains("rm ") || cmd.contains("del ") {
929 Some("Destructive File Deletion".into())
930 } else if cmd.contains("mkdir ") {
931 Some("Directory Creation".into())
932 } else {
933 Some("Execute Shell Command".into())
934 }
935 }
936 "write_file" => {
937 let path = args.get("path").and_then(|v| v.as_str()).unwrap_or("file");
938 Some(format!("Create/Overwrite File: {}", path))
939 }
940 "create_directory" => {
941 let path = args
942 .get("path")
943 .and_then(|v| v.as_str())
944 .unwrap_or("folder");
945 Some(format!("Create Directory: {}", path))
946 }
947 "edit_file" | "patch_hunk" | "multi_search_replace" => {
948 let path = args.get("path").and_then(|v| v.as_str()).unwrap_or("file");
949 Some(format!("Surgical Code Mutation: {}", path))
950 }
951 "git_commit" => Some("Permanent Version History Commit".into()),
952 "git_push" => Some("Remote Origin Synchronisation (Push)".into()),
953 "resolve_host_issue" => Some("System-Level Host Remediation".into()),
954 "run_workspace_workflow" => Some("Automated Workspace Re-alignment".into()),
955 _ => None,
956 }
957}