refactor: rename txt to text, document to file, and broaden usage tracking#800
Merged
Merged
Conversation
Broaden usage tracking from delegation-only sub-agents to all tool calls, adding input/output fields to the usage schema. Rename SubAgentDetailsDialog to ToolDetailsDialog and update validators, types, translations, and OpenAPI spec with backwards-compatible deprecation aliases.
- Update chat agent instructions to reference "file agent" instead of "document agent" - Fix tool-details-dialog display name mapping (document_assistant → file_assistant) - Remove text/csv from text tool mimeTypes to avoid overlap with excel tool - Replace text/* wildcard with explicit file extensions and MIME types - Add missing MIME types for .sql, .graphql extensions - Remove unused model parameter from Python vision extractors - Update stale comments in generate_response.ts and create_delegation_tool.ts
![greptile-apps[bot]](https://avatars.githubusercontent.com/in/867647?s=60&v=4){kind=small}
There was a problem hiding this comment.
Your free trial has ended. If you'd like to continue receiving code reviews, you can add a payment method here.
Contributor
📝 WalkthroughWalkthroughThis PR adds a Estimated code review effort🎯 3 (Moderate) | ⏱️ ~45 minutes Possibly related PRs
🚥 Pre-merge checks | ✅ 2 | ❌ 1❌ Failed checks (1 warning)
✅ Passed checks (2 passed)
✏️ Tip: You can configure your own custom pre-merge checks in the settings. ✨ Finishing Touches
🧪 Generate unit tests (beta)
📝 Coding Plan
Comment Tip Migrating from UI to YAML configuration.Use the |
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
Contributor
There was a problem hiding this comment.
Actionable comments posted: 4
Caution
Some comments are outside the diff and can’t be posted inline due to platform limitations.
⚠️ Outside diff range comments (6)
services/crawler/app/services/vision/openai_client.py (1)
289-294: 🧹 Nitpick | 🔵 TrivialConsider reusing the
VisionClientinstance instead of creating a newAsyncOpenAIclient per call.A new
AsyncOpenAIclient is instantiated on every call toprocess_pages_with_llm, while theVisionClientclass at the bottom of this file maintains a singleton pattern with_get_client(). This is likely intentional (different timeout: 180s vs 120s), but if call frequency is high, reusing a client or extracting this to a shared helper could reduce connection overhead.🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@services/crawler/app/services/vision/openai_client.py` around lines 289 - 294, process_pages_with_llm currently creates a new AsyncOpenAI client on each call (AsyncOpenAI instantiation with timeout=180.0) which duplicates connection overhead instead of reusing the VisionClient singleton; refactor to reuse a shared client by either (a) adding a parameterized client getter on VisionClient (e.g., VisionClient._get_client(timeout=180)) that returns a cached AsyncOpenAI instance per timeout or (b) extracting a shared helper factory used by process_pages_with_llm to obtain the client; update process_pages_with_llm to call that helper instead of instantiating AsyncOpenAI directly and preserve the intended resolved_model logic (resolved_model = model or settings.get_fast_model()) and the different timeout behavior.services/crawler/app/services/file_parser_service.py (1)
54-74: 🧹 Nitpick | 🔵 TrivialUpdate docstring to document the new
modelparameter.The
modelparameter was added toparse_pdf_with_visionbut the docstring (lines 65-74) doesn't document it. The same applies to the other vision methods (parse_docx_with_vision,parse_pptx_with_vision,parse_file_with_vision).📝 Suggested docstring update for parse_pdf_with_vision
Args: file_bytes: Raw PDF bytes filename: Filename for logging user_input: Optional user instruction for AI extraction process_images: Whether to extract and describe embedded images ocr_scanned_pages: Whether to OCR pages with low text content + model: Optional LLM model name for text processing (defaults to fast model) Returns: Extraction result with full_text and metadata🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@services/crawler/app/services/file_parser_service.py` around lines 54 - 74, The docstrings for parse_pdf_with_vision, parse_docx_with_vision, parse_pptx_with_vision, and parse_file_with_vision are missing documentation for the new model parameter; update each method's docstring to add a brief "model" entry in the Args section (e.g., model: Optional[str] = None — name of the vision/LLM model to use for extraction/analysis, defaults to the service-configured model), keeping wording consistent with existing parameter docs and noting default/behavior when None.services/platform/convex/agent_tools/files/helpers/analyze_text.ts (1)
1-6: 🧹 Nitpick | 🔵 TrivialUpdate outdated documentation comment.
The comment on line 2 mentions "using the fast model," but this PR changes the function to accept an explicit
modelparameter, making the model configurable by the caller. Consider updating the documentation to reflect this.📝 Suggested documentation update
/** - * Helper for analyzing text files using the fast model. + * Helper for analyzing text files using a specified LLM model. * Handles encoding detection, chunking for large files, and LLM analysis. * Uses ctx.storage.get() for direct Convex storage access (like analyze_image.ts). * Uses Agent framework with saveMessages: 'none' to avoid creating visible thread messages. */🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@services/platform/convex/agent_tools/files/helpers/analyze_text.ts` around lines 1 - 6, Update the top-of-file documentation to remove the hardcoded "fast model" reference and document the new configurable model parameter (e.g., model) accepted by the analyzeText function; mention that the caller supplies the model and that the helper will use that model for LLM analysis, and keep existing notes about encoding detection, chunking, ctx.storage.get(), and saveMessages: 'none' intact.services/platform/convex/lib/agent_chat/internal_actions.ts (1)
574-595:⚠️ Potential issue | 🟠 MajorRegression: attachment context is dropped before generation.
The
beforeGenerateHookreceivesattachments(passed bybuildHooksFromConfigat line 469), but the handler at line 574 does not destructure them, and line 594 returnspromptContent: undefinedunconditionally. This prevents attached files from being processed into prompt content, breaking file-attachment functionality.💡 Proposed fix (restore attachment-derived prompt content)
- const { threadId, promptMessage, contextMessagesTokens } = args; + const { threadId, promptMessage, attachments, contextMessagesTokens } = args; @@ - return { - promptContent: undefined, - contextExceedsBudget, - }; + const promptContent = + attachments && attachments.length > 0 + ? await processAttachments(ctx, attachments, promptMessage) + : undefined; + + return { + promptContent, + contextExceedsBudget, + };🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@services/platform/convex/lib/agent_chat/internal_actions.ts` around lines 574 - 595, The handler currently drops attachments by not reading args.attachments and always returning promptContent: undefined; fix it by destructuring attachments from args alongside threadId/promptMessage/contextMessagesTokens, then restore the attachment-derived prompt content before return (e.g., if attachments exist, produce promptContent from args.attachments using the same conversion logic used elsewhere / by buildHooksFromConfig and return that value instead of undefined), keeping the contextExceedsBudget logging via beforeGenerateDebugLog and preserving existing variables like threadId and contextMessagesTokens.services/platform/convex/lib/attachments/process_attachments.ts (1)
60-65: 🧹 Nitpick | 🔵 TrivialMake the config contract consistently require
model.
ProcessAttachmentsConfigmarksmodeloptional, while the exported function immediately strengthens it toProcessAttachmentsConfig & { model: string }. Making the interface itself required is clearer and removes a misleading public shape.Proposed refactor
export interface ProcessAttachmentsConfig { maxDocumentLength?: number; debugLog?: (message: string, data?: Record<string, unknown>) => void; toolName?: string; - model?: string; + model: string; } @@ export async function processAttachments( ctx: ActionCtx, attachments: FileAttachment[], userText: string | undefined, - config: ProcessAttachmentsConfig & { model: string }, + config: ProcessAttachmentsConfig, ): Promise<ProcessedAttachments> {Also applies to: 83-88
🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@services/platform/convex/lib/attachments/process_attachments.ts` around lines 60 - 65, ProcessAttachmentsConfig currently declares model as optional but the exported function expects a guaranteed model; change the interface ProcessAttachmentsConfig to require model: string (remove the optional mark) and update any related type intersections (e.g., the exported function signature that used ProcessAttachmentsConfig & { model: string }) to just use ProcessAttachmentsConfig. Ensure callers and any defaulting logic (referenced in the file around the exported function and related helper types) are updated to pass a model string or handle validation so the new required property is satisfied.services/platform/app/features/chat/components/message-info-dialog.tsx (1)
223-234:⚠️ Potential issue | 🟠 MajorThe list still uses the pre-rename tool formatter.
formatAgentNameonly knowsdocument_assistant, so afile_assistantrow will render as the raw identifier here whileToolDetailsDialognow titles it asFile. Reuse one formatter in both places.🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@services/platform/app/features/chat/components/message-info-dialog.tsx` around lines 223 - 234, The UI is using the old formatter formatAgentName in message-info-dialog.tsx which doesn't handle the renamed tool ids (e.g., file_assistant), so create or expose a single shared formatter (e.g., formatToolName or export the formatter used by ToolDetailsDialog) and replace formatAgentName usages with that shared function; update message-info-dialog.tsx (the render mapping that calls formatAgentName) to import and call the shared formatter so both the list and ToolDetailsDialog produce identical human-friendly titles for tools like document_assistant and file_assistant.
🤖 Prompt for all review comments with AI agents
Verify each finding against the current code and only fix it if needed.
Inline comments:
In `@services/platform/app/features/chat/components/message-info-dialog.tsx`:
- Around line 289-292: The nested ToolDetailsDialog can remain open because
selectedTool outlives the parent dialog; update the parent dialog's close path
to clear selection so the child cannot survive a parent close — specifically,
ensure the parent's onOpenChange handler (the prop controlling isOpen) calls
setSelectedTool(null) when open becomes false (in addition to or instead of only
relying on ToolDetailsDialog's onOpenChange), so that ToolDetailsDialog receives
isOpen={selectedTool !== null} and will always reset when the parent closes.
In `@services/platform/convex/custom_agents/system_defaults.ts`:
- Around line 85-91: Update the agent instructions and tool descriptions to
remove any reference to the ".csv" extension for the text tool: edit
FILE_AGENT_INSTRUCTIONS in services/platform/convex/agents/file/agent.ts to stop
instructing the agent to treat .csv via the text tool, update the textTool
description in services/platform/convex/agent_tools/files/text_tool.ts (both the
header and the detailed description) to list only plain text-like extensions
(e.g., .txt, .md) or generic "text" instead of ".csv", and change the error
message in services/platform/convex/agent_tools/files/helpers/analyze_text.ts to
not mention ".csv" when suggesting the text tool; ensure CSV handling remains
mapped to the excel tool (leave TOOL_FILE_MAP as-is) and adjust phrasing so the
agent will route CSVs to the excel tool rather than the text tool.
In `@services/platform/convex/lib/agent_response/types.ts`:
- Around line 140-150: Reintroduce the deprecated alias subAgentUsage into the
public response type so the contract stays backward-compatible: add an optional
subAgentUsage?: Array<{ toolName: string; model?: string; provider?: string;
inputTokens?: number; outputTokens?: number; totalTokens?: number; durationMs?:
number; input?: string; output?: string; }>; alongside toolsUsage in
services/platform/convex/lib/agent_response/types.ts, and update the runtime
validator in services/platform/convex/lib/agent_response/validators.ts (the
validator that validates toolsUsage) to accept and normalize the alias (i.e.,
validate the same shape for subAgentUsage as for toolsUsage and continue to
support either field at runtime).
In `@services/platform/convex/message_metadata/internal_mutations.ts`:
- Line 27: The mutation stopped populating the deprecated alias field
subAgentUsage even though the table schema still exposes it; update the insert
and patch handlers in internal_mutations.ts that write to messageMetadataTable
so they persist subAgentUsage alongside toolsUsage (map the same array/value
used for toolsUsage into subAgentUsage) whenever toolsUsage is present, and do
the same mapping in any validators or transformation steps referenced (see the
toolsUsage: v.optional(...) validator and the mutations around lines referenced
61-62 and 82-82) so older readers continue to see fresh data.
---
Outside diff comments:
In `@services/crawler/app/services/file_parser_service.py`:
- Around line 54-74: The docstrings for parse_pdf_with_vision,
parse_docx_with_vision, parse_pptx_with_vision, and parse_file_with_vision are
missing documentation for the new model parameter; update each method's
docstring to add a brief "model" entry in the Args section (e.g., model:
Optional[str] = None — name of the vision/LLM model to use for
extraction/analysis, defaults to the service-configured model), keeping wording
consistent with existing parameter docs and noting default/behavior when None.
In `@services/crawler/app/services/vision/openai_client.py`:
- Around line 289-294: process_pages_with_llm currently creates a new
AsyncOpenAI client on each call (AsyncOpenAI instantiation with timeout=180.0)
which duplicates connection overhead instead of reusing the VisionClient
singleton; refactor to reuse a shared client by either (a) adding a
parameterized client getter on VisionClient (e.g.,
VisionClient._get_client(timeout=180)) that returns a cached AsyncOpenAI
instance per timeout or (b) extracting a shared helper factory used by
process_pages_with_llm to obtain the client; update process_pages_with_llm to
call that helper instead of instantiating AsyncOpenAI directly and preserve the
intended resolved_model logic (resolved_model = model or
settings.get_fast_model()) and the different timeout behavior.
In `@services/platform/app/features/chat/components/message-info-dialog.tsx`:
- Around line 223-234: The UI is using the old formatter formatAgentName in
message-info-dialog.tsx which doesn't handle the renamed tool ids (e.g.,
file_assistant), so create or expose a single shared formatter (e.g.,
formatToolName or export the formatter used by ToolDetailsDialog) and replace
formatAgentName usages with that shared function; update message-info-dialog.tsx
(the render mapping that calls formatAgentName) to import and call the shared
formatter so both the list and ToolDetailsDialog produce identical
human-friendly titles for tools like document_assistant and file_assistant.
In `@services/platform/convex/agent_tools/files/helpers/analyze_text.ts`:
- Around line 1-6: Update the top-of-file documentation to remove the hardcoded
"fast model" reference and document the new configurable model parameter (e.g.,
model) accepted by the analyzeText function; mention that the caller supplies
the model and that the helper will use that model for LLM analysis, and keep
existing notes about encoding detection, chunking, ctx.storage.get(), and
saveMessages: 'none' intact.
In `@services/platform/convex/lib/agent_chat/internal_actions.ts`:
- Around line 574-595: The handler currently drops attachments by not reading
args.attachments and always returning promptContent: undefined; fix it by
destructuring attachments from args alongside
threadId/promptMessage/contextMessagesTokens, then restore the
attachment-derived prompt content before return (e.g., if attachments exist,
produce promptContent from args.attachments using the same conversion logic used
elsewhere / by buildHooksFromConfig and return that value instead of undefined),
keeping the contextExceedsBudget logging via beforeGenerateDebugLog and
preserving existing variables like threadId and contextMessagesTokens.
In `@services/platform/convex/lib/attachments/process_attachments.ts`:
- Around line 60-65: ProcessAttachmentsConfig currently declares model as
optional but the exported function expects a guaranteed model; change the
interface ProcessAttachmentsConfig to require model: string (remove the optional
mark) and update any related type intersections (e.g., the exported function
signature that used ProcessAttachmentsConfig & { model: string }) to just use
ProcessAttachmentsConfig. Ensure callers and any defaulting logic (referenced in
the file around the exported function and related helper types) are updated to
pass a model string or handle validation so the new required property is
satisfied.
🪄 Autofix (Beta)
Fix all unresolved CodeRabbit comments on this PR:
- Push a commit to this branch (recommended)
- Create a new PR with the fixes
ℹ️ Review info
⚙️ Run configuration
Configuration used: Organization UI
Review profile: ASSERTIVE
Plan: Pro
Run ID: 89d75a5e-f413-4551-b2bc-4f331953b88a
⛔ Files ignored due to path filters (1)
services/platform/convex/_generated/api.d.tsis excluded by!**/_generated/**
📒 Files selected for processing (36)
services/crawler/app/routers/docx.pyservices/crawler/app/routers/pdf.pyservices/crawler/app/routers/pptx.pyservices/crawler/app/services/file_parser_service.pyservices/crawler/app/services/vision/openai_client.pyservices/platform/app/features/chat/components/message-info-dialog.tsxservices/platform/app/features/chat/components/tool-details-dialog.tsxservices/platform/app/features/chat/hooks/queries.tsservices/platform/app/features/custom-agents/components/tool-selector.tsxservices/platform/convex/agent_tools/delegation/create_delegation_tool.tsservices/platform/convex/agent_tools/files/docx_tool.tsservices/platform/convex/agent_tools/files/helpers/analyze_text.tsservices/platform/convex/agent_tools/files/helpers/get_agent_model.tsservices/platform/convex/agent_tools/files/helpers/parse_file.tsservices/platform/convex/agent_tools/files/internal_actions.tsservices/platform/convex/agent_tools/files/pdf_tool.tsservices/platform/convex/agent_tools/files/pptx_tool.tsservices/platform/convex/agent_tools/files/text_tool.tsservices/platform/convex/agent_tools/tool_names.tsservices/platform/convex/agent_tools/tool_registry.tsservices/platform/convex/agents/chat/agent.tsservices/platform/convex/agents/file/agent.tsservices/platform/convex/custom_agents/system_defaults.tsservices/platform/convex/lib/agent_chat/internal_actions.tsservices/platform/convex/lib/agent_completion/on_agent_complete.tsservices/platform/convex/lib/agent_response/generate_response.tsservices/platform/convex/lib/agent_response/types.tsservices/platform/convex/lib/agent_response/validators.tsservices/platform/convex/lib/attachments/process_attachments.tsservices/platform/convex/message_metadata/internal_mutations.tsservices/platform/convex/streaming/schema.tsservices/platform/convex/streaming/validators.tsservices/platform/lib/shared/__tests__/file-types-tool-mapping.test.tsservices/platform/lib/shared/file-types.tsservices/platform/messages/en.jsonservices/platform/public/openapi.json
Comment on lines
+289
to
+292
| <ToolDetailsDialog | ||
| isOpen={selectedTool !== null} | ||
| onOpenChange={(open) => !open && setSelectedTool(null)} | ||
| usage={selectedTool} |
Contributor
There was a problem hiding this comment.
Reset the nested tool dialog when the parent closes.
selectedTool outlives isOpen, so the child dialog can survive the parent close or reopen immediately on the next open. Clear the selection from the parent close path instead of only when the child closes itself.
🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed.
In `@services/platform/app/features/chat/components/message-info-dialog.tsx`
around lines 289 - 292, The nested ToolDetailsDialog can remain open because
selectedTool outlives the parent dialog; update the parent dialog's close path
to clear selection so the child cannot survive a parent close — specifically,
ensure the parent's onOpenChange handler (the prop controlling isOpen) calls
setSelectedTool(null) when open becomes false (in addition to or instead of only
relying on ToolDetailsDialog's onOpenChange), so that ToolDetailsDialog receives
isOpen={selectedTool !== null} and will always reset when the parent closes.
Comment on lines
+85
to
+91
| systemAgentSlug: 'file', | ||
| name: 'file-assistant', | ||
| displayName: 'File Assistant', | ||
| description: | ||
| 'Reads and creates documents (PDF, Word, Excel, etc.). Can also create visual images by rendering HTML/CSS/SVG — use it for illustrations, charts, infographics, diagrams, and web page screenshots.', | ||
| systemInstructions: DOCUMENT_AGENT_INSTRUCTIONS, | ||
| toolNames: ['pdf', 'image', 'docx', 'pptx', 'txt', 'excel'], | ||
| 'Reads and creates files (PDF, Word, Excel, etc.). Can also create visual images by rendering HTML/CSS/SVG — use it for illustrations, charts, infographics, diagrams, and web page screenshots.', | ||
| systemInstructions: FILE_AGENT_INSTRUCTIONS, | ||
| toolNames: ['pdf', 'image', 'docx', 'pptx', 'text', 'excel'], |
Contributor
There was a problem hiding this comment.
🧩 Analysis chain
🏁 Script executed:
#!/bin/bash
# Check if FILE_AGENT_INSTRUCTIONS mentions .csv for text tool
rg -n '\.csv' services/platform/convex/agents/file/agent.tsRepository: tale-project/tale
Length of output: 179
🏁 Script executed:
find . -name "file-types.ts" -type fRepository: tale-project/tale
Length of output: 105
🏁 Script executed:
rg -n "\.csv|TOOL_FILE_MAP" services/platform/convex -A 3 -B 3 | head -100Repository: tale-project/tale
Length of output: 3072
🏁 Script executed:
cat services/platform/lib/shared/file-types.tsRepository: tale-project/tale
Length of output: 13439
Update agent instructions and tool descriptions to remove .csv from text tool references.
The .csv extension was correctly removed from the text tool mapping in TOOL_FILE_MAP (now exclusively mapped to excel tool), but the agent instructions still mention .csv for the text tool:
FILE_AGENT_INSTRUCTIONSinservices/platform/convex/agents/file/agent.ts:52textTooldescription inservices/platform/convex/agent_tools/files/text_tool.ts:4, :86- Error message in
services/platform/convex/agent_tools/files/helpers/analyze_text.ts:338
Remove .csv from all these text tool references to prevent the AI agent from incorrectly attempting CSV parsing with the text tool.
🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed.
In `@services/platform/convex/custom_agents/system_defaults.ts` around lines 85 -
91, Update the agent instructions and tool descriptions to remove any reference
to the ".csv" extension for the text tool: edit FILE_AGENT_INSTRUCTIONS in
services/platform/convex/agents/file/agent.ts to stop instructing the agent to
treat .csv via the text tool, update the textTool description in
services/platform/convex/agent_tools/files/text_tool.ts (both the header and the
detailed description) to list only plain text-like extensions (e.g., .txt, .md)
or generic "text" instead of ".csv", and change the error message in
services/platform/convex/agent_tools/files/helpers/analyze_text.ts to not
mention ".csv" when suggesting the text tool; ensure CSV handling remains mapped
to the excel tool (leave TOOL_FILE_MAP as-is) and adjust phrasing so the agent
will route CSVs to the excel tool rather than the text tool.
Comment on lines
+140
to
150
| toolsUsage?: Array<{ | ||
| toolName: string; | ||
| model?: string; | ||
| provider?: string; | ||
| inputTokens?: number; | ||
| outputTokens?: number; | ||
| totalTokens?: number; | ||
| durationMs?: number; | ||
| input?: string; | ||
| output?: string; | ||
| }>; |
Contributor
There was a problem hiding this comment.
Preserve the deprecated subAgentUsage field in the public response type.
Removing it here makes the shared response contract backward-incompatible, even though the PR objective says older clients should keep receiving the alias during the transition. The runtime validator in services/platform/convex/lib/agent_response/validators.ts should mirror the same alias.
Proposed compatibility fix
export interface GenerateResponseResult {
threadId: string;
text: string;
usage?: {
inputTokens?: number;
outputTokens?: number;
totalTokens?: number;
reasoningTokens?: number;
cachedInputTokens?: number;
};
finishReason?: string;
durationMs: number;
timeToFirstTokenMs?: number;
toolCalls?: Array<{ toolName: string; status: string }>;
+ /** `@deprecated` Use toolsUsage instead. */
+ subAgentUsage?: Array<{
+ toolName: string;
+ model?: string;
+ provider?: string;
+ inputTokens?: number;
+ outputTokens?: number;
+ totalTokens?: number;
+ durationMs?: number;
+ input?: string;
+ output?: string;
+ }>;
toolsUsage?: Array<{
toolName: string;
model?: string;
provider?: string;
inputTokens?: number;
outputTokens?: number;
totalTokens?: number;
durationMs?: number;
input?: string;
output?: string;
}>;📝 Committable suggestion
‼️ IMPORTANT
Carefully review the code before committing. Ensure that it accurately replaces the highlighted code, contains no missing lines, and has no issues with indentation. Thoroughly test & benchmark the code to ensure it meets the requirements.
Suggested change
| toolsUsage?: Array<{ | |
| toolName: string; | |
| model?: string; | |
| provider?: string; | |
| inputTokens?: number; | |
| outputTokens?: number; | |
| totalTokens?: number; | |
| durationMs?: number; | |
| input?: string; | |
| output?: string; | |
| }>; | |
| /** `@deprecated` Use toolsUsage instead. */ | |
| subAgentUsage?: Array<{ | |
| toolName: string; | |
| model?: string; | |
| provider?: string; | |
| inputTokens?: number; | |
| outputTokens?: number; | |
| totalTokens?: number; | |
| durationMs?: number; | |
| input?: string; | |
| output?: string; | |
| }>; | |
| toolsUsage?: Array<{ | |
| toolName: string; | |
| model?: string; | |
| provider?: string; | |
| inputTokens?: number; | |
| outputTokens?: number; | |
| totalTokens?: number; | |
| durationMs?: number; | |
| input?: string; | |
| output?: string; | |
| }>; |
🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed.
In `@services/platform/convex/lib/agent_response/types.ts` around lines 140 - 150,
Reintroduce the deprecated alias subAgentUsage into the public response type so
the contract stays backward-compatible: add an optional subAgentUsage?: Array<{
toolName: string; model?: string; provider?: string; inputTokens?: number;
outputTokens?: number; totalTokens?: number; durationMs?: number; input?:
string; output?: string; }>; alongside toolsUsage in
services/platform/convex/lib/agent_response/types.ts, and update the runtime
validator in services/platform/convex/lib/agent_response/validators.ts (the
validator that validates toolsUsage) to accept and normalize the alias (i.e.,
validate the same shape for subAgentUsage as for toolsUsage and continue to
support either field at runtime).
| durationMs: v.optional(v.number()), | ||
| timeToFirstTokenMs: v.optional(v.number()), | ||
| subAgentUsage: v.optional(v.array(subAgentUsageItemValidator)), | ||
| toolsUsage: v.optional(v.array(toolUsageItemValidator)), |
Contributor
There was a problem hiding this comment.
Persist the deprecated alias while the schema still exposes it.
messageMetadataTable still defines subAgentUsage, but this mutation no longer writes it on insert or patch. Older readers of that field will stop seeing fresh data even though the schema advertises compatibility.
Proposed compatibility fix
args: {
@@
timeToFirstTokenMs: v.optional(v.number()),
toolsUsage: v.optional(v.array(toolUsageItemValidator)),
contextWindow: v.optional(v.string()),
contextStats: v.optional(contextStatsValidator),
},
@@
durationMs: args.durationMs ?? existing.durationMs,
timeToFirstTokenMs:
args.timeToFirstTokenMs ?? existing.timeToFirstTokenMs,
toolsUsage: args.toolsUsage ?? existing.toolsUsage,
+ subAgentUsage: args.toolsUsage ?? existing.subAgentUsage,
contextWindow: contextWindow ?? existing.contextWindow,
contextStats: args.contextStats ?? existing.contextStats,
});
@@
reasoning: args.reasoning,
providerMetadata: args.providerMetadata,
durationMs: args.durationMs,
timeToFirstTokenMs: args.timeToFirstTokenMs,
toolsUsage: args.toolsUsage,
+ subAgentUsage: args.toolsUsage,
contextWindow,
contextStats: args.contextStats,
});Also applies to: 61-62, 82-82
🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed.
In `@services/platform/convex/message_metadata/internal_mutations.ts` at line 27,
The mutation stopped populating the deprecated alias field subAgentUsage even
though the table schema still exposes it; update the insert and patch handlers
in internal_mutations.ts that write to messageMetadataTable so they persist
subAgentUsage alongside toolsUsage (map the same array/value used for toolsUsage
into subAgentUsage) whenever toolsUsage is present, and do the same mapping in
any validators or transformation steps referenced (see the toolsUsage:
v.optional(...) validator and the mutations around lines referenced 61-62 and
82-82) so older readers continue to see fresh data.
- Rename document→file in AGENT_CONTEXT_CONFIGS key - Update txt→text in document_find and document_retrieve tool descriptions - Fix formatAgentName mapping: document_assistant→file_assistant - Harden getAgentModelId to handle both modelId and model properties - Guard aggregateChunkResults against empty chunkResults array
Add UsageAccumulator to collect input/output tokens and duration across all OpenAI Vision API calls (OCR, image description, LLM processing) for PDF, DOCX, and PPTX extractors. Surface usage data through the crawler response and remap to camelCase in the platform parse_file helper.
This was referenced Mar 24, 2026
yannickmonney
pushed a commit
that referenced
this pull request
Apr 8, 2026
6 tasks
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
txttool totextand thedocumentagent tofilefor claritysubAgentUsagetotoolsUsageand track all tool calls (not just delegation tools), with backwards-compatible deprecation aliases in the APITest plan
texttool handles all previously supported file typesfileagent delegation works end-to-endtoolsUsageis populated for all tool calls in message metadatasubAgentUsagefield still returns data for existing clients🤖 Generated with Claude Code
Summary by CodeRabbit
Release Notes
New Features
Improvements