file-reader

name	file-reader
description	Read text content from files the current user can access.
when_to_use	Use this skill when the user asks to inspect, quote, summarize, compare, translate, or answer from an uploaded file, historical file reference, or a file shown in the console files page.
provider_type	builtin
provider_id	files
runtime_type	tool
tools	["list_visible_files","read_file"]
max_calls_per_turn	10
timeout_seconds	120
tool_governance	{"list_visible_files":{"tool_id":"file.list_visible","skill_id":"file-reader","domain":"files","effect":"read","asset_type":"file","risk_level":"low","requires_asset_resolution":false,"reversible":false,"bulk_sensitive":false,"external_side_effect":false,"permission_scopes":["file:read"],"default_approval_policy":"auto_by_permission_tier","allowed_permission_tiers":["basic","advanced","full"],"audit_required":true,"idempotency_required":false},"read_file":{"tool_id":"file.read","skill_id":"file-reader","domain":"files","effect":"read","asset_type":"file","risk_level":"low","requires_asset_resolution":true,"reversible":false,"bulk_sensitive":false,"external_side_effect":false,"permission_scopes":["file:read"],"default_approval_policy":"auto_by_permission_tier","allowed_permission_tiers":["basic","advanced","full"],"audit_required":true,"idempotency_required":false}}
display	{"icon":"file-text","category":"document_processing","scenarios":["document_handling","office_collaboration","knowledge_research"],"label":{"en_US":"File Reader","zh_Hans":"文件读取"},"description":{"en_US":"Designed for understanding uploaded, historical, or console files; locates accessible files and reads their text for answering, summarizing, comparing, or translating.","zh_Hans":"适用于理解上传文件、历史文件或控制台文件，可定位并读取有权限访问的文本，用于问答、总结、对比或翻译。"},"when_to_use":{"en_US":"Use when an answer needs content from a specific file available to the current user."},"tags":{"en_US":["File","Reading"]}}
supported_callers	["aichat"]

File Reader Skill

Use this skill to list visible file context or read content from a file that has already been resolved for the current assistant turn.

Workflow

1. Use this skill only when the user is asking about a specific file, uploaded attachment, historical file reference, or current console file context.
2. Do not invent file IDs. Use a file ID supplied by resolved page context, attachment context, or governed asset resolution.
3. For listing requests such as "what files do I have", "which files are visible", "current files", or "selected files", use the current backend-backed page context when it is present and fresh. Call list_visible_files when that context is missing, ambiguous, stale, or needs an authoritative refresh. Do not use read_file for a list-only request unless the user asks for file contents.
4. For read, summary, translation, extraction, comparison, or question answering, call read_file with file_id. Set max_chars only when you need more or less returned content.
5. Inspect content_status before answering:
   • If content_status is extracted, answer from content.
   • If content_status is empty, say the file has no extractable text content.
   • If content_status is error, explain that the file could not be read and include the short error reason when useful.
6. The raw content returned by read_file is not durable across history reloads, approvals, navigation, refresh, or later tool phases. If later steps will use the file body, a summary, theme, topic, quote, title, prompt, config value, generated asset, or final answer derived from this file, call submit_turn_state before leaving the file-reading phase. Record a concise reusable fact with kind=working_fact, visibility=model_only, source=file-reader/read_file, and a meaningful key such as source_file_summary, source_file_theme, or exact_short_text. When the summary is useful for the user to verify, use kind=user_deliverable, visibility=user_visible, and a short title/content instead.
7. If content_truncated is true and the missing tail matters, ask for a narrower question or retry with a higher up to the tool limit.