---
title: "Capabilities"
canonical: "https://canmyagentuse.com/features"
contentKind: "page"
locale: "en"
description: "Vendor-neutral capabilities with exact harness assertions and public evidence."
llmSummary: "Vendor-neutral capabilities with exact harness assertions and public evidence."
publishedAt: null
updatedAt: null
verifiedAt: null
tags: []
---

# Capabilities

Vendor-neutral capabilities with exact harness assertions and public evidence.

- HTML: https://canmyagentuse.com/features
- JSON: https://canmyagentuse.com/api/v1/features.json
- Markdown: https://canmyagentuse.com/features.md

- [Agent Plugins failure isolation](/features/agent-plugin-failure-isolation.md) — Agent Plugins failure isolation is atomic and does not follow automatically from broader plugins support.
- [Agent Plugins packaged MCP](/features/agent-plugin-mcp.md) — Agent Plugins packaged MCP is atomic and does not follow automatically from broader plugins support.
- [Agent Plugins packaged skills](/features/agent-plugin-skills.md) — Agent Plugins packaged skills is atomic and does not follow automatically from broader plugins support.
- [Agent Plugins portable manifest](/features/agent-plugin-core.md) — Agent Plugins portable manifest is atomic and does not follow automatically from broader plugins support.
- [Agent Skills](/features/agent-skills.md) — Agent Skills support is derived from atomic discovery, progressive loading, component, invocation, and scope claims.
- [Agent Skills core and discovery](/features/skills.md) — This atomic row covers recognizing, discovering, and loading standard SKILL.md packages. Scripts, assets, invocation, and scopes have separate sibling rows.
- [Agent Skills manual invocation](/features/skill-manual-invocation.md) — Agent Skills manual invocation is atomic and does not follow automatically from broader agent-skills support.
- [Agent Skills progressive loading](/features/skill-progressive-loading.md) — Agent Skills progressive loading is atomic and does not follow automatically from broader agent-skills support.
- [Agent Skills references and assets](/features/skill-references-assets.md) — Agent Skills references and assets is atomic and does not follow automatically from broader agent-skills support.
- [Agent Skills scripts](/features/skill-scripts.md) — Agent Skills scripts is atomic and does not follow automatically from broader agent-skills support.
- [AGENTS.md root instructions](/features/agents-md.md) — AGENTS.md root instructions is atomic and does not follow automatically from broader instructions support.
- [Artifact export](/features/artifact-export.md) — Artifact export means a product lets the user download generated files or other documented outputs beyond the visible chat text.
- [Audio file input](/features/audio-file-input.md) — Audio file input covers recorded uploads and is separate from realtime voice. Transcription, timestamps, speaker labels, and acoustic analysis are recorded as qualifiers only when documented.
- [Audit logs](/features/audit-logs.md) — Audit logs are product-provided administrative records of security-relevant activity; ordinary conversation history is not automatically an audit log.
- [Automatic context compaction](/features/automatic-context-compaction.md) — Automatic context compaction lets a harness reduce older context to continue a run; it must be distinguished from silent truncation and from user-triggered compaction.
- [Background agents](/features/background-agents.md) — Background agents continue running after the operator leaves the active session. Hosting, subscription, duration, and notification behavior are qualifiers.
- [Browser automation](/features/browser-automation.md) — Browser automation controls a web browser for navigation and page interaction. Browser-only automation is distinct from general operating-system control.
- [Code execution](/features/code-execution.md) — Code execution runs generated code in a product-provided environment. Sandbox isolation, workspace access, permissions, language support, and persistence are qualifiers.
- [Collaboration](/features/collaboration-and-portability.md) — Collaboration is an internal catalog grouping for conversation sharing, shared workspaces, role-based access control, conversation export, and artifact export.
- [Computer use](/features/computer-use.md) — Computer use is screenshot-informed interaction with a graphical interface through input actions. Browser-only and general desktop control are recorded separately.
- [Concurrent subagents](/features/subagent-concurrency.md) — Concurrent subagents means multiple child-agent tasks can overlap in time; sequential delegation, tool-call parallelism, and background shell processes are separate behaviors.
- [Connectors](/features/connectors.md) — Connectors provide documented access to external services authorized by the user or organization. Service coverage, authentication, read/write scope, and policy are qualifiers.
- [Context compaction controls](/features/context-compaction-controls.md) — Compaction controls cover operator actions such as manual compaction, threshold configuration, summary inspection, retained-item pinning, and disabling automatic compaction.
- [Context usage](/features/context-usage-visibility.md) — Context usage means the product reports current context consumption or remaining capacity; exact counters, estimates, source breakdowns, and warnings are qualifiers.
- [Context window](/features/context-window.md) — Context-window records distinguish published model capacity from the limit documented for a particular product and mode. Usable input can be lower because of output reservations, instructions, tools, files, and product caps.
- [Conversation export](/features/conversation-export.md) — Conversation export means a product provides a documented way to export chat or conversation history; exported fields and formats are recorded as qualifiers.
- [Conversation sharing](/features/conversation-sharing.md) — Conversation sharing means a product provides a documented link or invitation for another person to view or use a conversation; permissions and revocation are qualifiers.
- [Custom instructions](/features/custom-instructions.md) — Custom instructions persist user- or project-scoped guidance. Storage format, discovery, scope, and precedence are qualifiers.
- [Data residency](/features/data-residency.md) — Data residency is the documented geographic location of covered storage or processing; storage residency does not automatically establish regional inference.
- [Data retention controls](/features/data-retention-controls.md) — Retention controls cover prompts, files, outputs, tool data, memory, logs, feedback, and backups; deleting a visible conversation alone is a narrower claim.
- [Diffs and patches](/features/diffs-and-patches.md) — Diffs and patches represent proposed file changes in a reviewable patch form. Generation, application, approval, and source-control integration are qualifiers.
- [Document input](/features/office-document-input.md) — Document input means uploaded text, word-processing, presentation, or spreadsheet content is available as model input. Supported formats and extraction fidelity are qualifiers.
- [Encryption key management](/features/encryption-key-controls.md) — Encryption key management covers documented provider-managed or customer-managed keys; coverage, rotation, and revocation are recorded as qualifiers.
- [File and media inputs](/features/file-inputs.md) — File and media inputs is an internal catalog grouping for image, PDF, office-document, audio, video, and upload-limit questions.
- [File upload limits](/features/upload-limits.md) — File upload limits record documented size, count, page, duration, frequency, and storage limits with their scope. Accepting one test file does not establish every limit.
- [Human approval](/features/human-approval.md) — Human approval pauses an action until a person confirms or denies it. Action scope, remembered decisions, and policy behavior are qualifiers.
- [Image input](/features/image-input.md) — Image input means a product accepts attached, pasted, or selected images as model input. Supported methods, formats, limits, and surfaces are recorded as qualifiers.
- [Local models](/features/local-models.md) — Local model support lets a product use a model hosted on the operator's machine. Model servers, formats, authentication, and feature limitations are qualifiers.
- [Long-term memory](/features/long-term-memory.md) — Long-term memory persists documented preferences or working context across sessions. Scope, user controls, shared storage, deletion, and security boundaries are qualifiers.
- [MCP Apps](/features/mcp-apps.md) — MCP Apps let tools return interactive interfaces that render inside a compatible MCP host. Selected current-track cells are sourced; all others remain unknown.
- [MCP cancellation](/features/mcp-cancellation.md) — MCP cancellation is an atomic MCP capability. Support for another MCP primitive does not imply this capability.
- [MCP client role](/features/mcp-client-role.md) — MCP client role is an atomic MCP capability. Support for another MCP primitive does not imply this capability.
- [MCP elicitation](/features/mcp-elicitation.md) — MCP elicitation is an atomic MCP capability. Support for another MCP primitive does not imply this capability.
- [MCP legacy HTTP and SSE](/features/mcp-legacy-sse.md) — MCP legacy HTTP and SSE is an atomic MCP capability. Support for another MCP primitive does not imply this capability.
- [MCP list-changed notifications](/features/mcp-list-changed.md) — MCP list-changed notifications is an atomic MCP capability. Support for another MCP primitive does not imply this capability.
- [MCP logging](/features/mcp-logging.md) — MCP logging is an atomic MCP capability. Support for another MCP primitive does not imply this capability.
- [MCP OAuth authorization](/features/mcp-oauth.md) — MCP OAuth authorization is an atomic MCP capability. Support for another MCP primitive does not imply this capability.
- [MCP progress notifications](/features/mcp-progress.md) — MCP progress notifications is an atomic MCP capability. Support for another MCP primitive does not imply this capability.
- [MCP prompt-list notifications](/features/mcp-prompts-list-changed.md) — MCP prompt-list notifications is an atomic MCP capability. Support for another MCP primitive does not imply this capability.
- [MCP prompts](/features/mcp-prompts.md) — MCP prompts: Invoke reusable MCP prompt templates. Unreviewed harness cells remain unknown until dated public evidence exists.
- [MCP Registry metadata](/features/mcp-registry.md) — MCP Registry metadata is an atomic MCP capability. Support for another MCP primitive does not imply this capability.
- [MCP resource subscriptions](/features/mcp-resource-subscriptions.md) — MCP resource subscriptions is an atomic MCP capability. Support for another MCP primitive does not imply this capability.
- [MCP resources](/features/mcp-resources.md) — MCP resources: Read MCP resources as live context. Unreviewed harness cells remain unknown until dated public evidence exists.
- [MCP revision compatibility](/features/mcp-revision.md) — MCP revision compatibility is an atomic MCP capability. Support for another MCP primitive does not imply this capability.
- [MCP roots](/features/mcp-roots.md) — MCP roots is an atomic MCP capability. Support for another MCP primitive does not imply this capability.
- [MCP sampling](/features/mcp-sampling.md) — MCP sampling is an atomic MCP capability. Support for another MCP primitive does not imply this capability.
- [MCP server instructions](/features/mcp-server-instructions.md) — MCP server instructions is an atomic MCP capability. Support for another MCP primitive does not imply this capability.
- [MCP server role](/features/mcp-server-role.md) — MCP server role is an atomic MCP capability. Support for another MCP primitive does not imply this capability.
- [MCP stdio transport](/features/mcp-stdio.md) — MCP stdio transport is an atomic MCP capability. Support for another MCP primitive does not imply this capability.
- [MCP Streamable HTTP](/features/mcp-streamable-http.md) — MCP Streamable HTTP is an atomic MCP capability. Support for another MCP primitive does not imply this capability.
- [MCP task listing](/features/mcp-task-listing.md) — MCP task listing is an atomic MCP capability. Support for another MCP primitive does not imply this capability.
- [MCP tasks](/features/mcp-tasks.md) — MCP tasks is an atomic MCP capability. Support for another MCP primitive does not imply this capability.
- [MCP tools](/features/mcp-tools.md) — MCP tools let a harness invoke operations exposed by a Model Context Protocol server. Selected current-track cells are sourced; all others remain unknown.
- [Model Context Protocol](/features/mcp.md) — MCP is a capability family. Support is rolled up from atomic child capabilities so tool support never implies prompt, resource, or app support.
- [Model fallback](/features/model-fallback.md) — Model fallback is documented switching to another model after a capacity, policy, or service failure. Disclosure, configuration, and opt-out behavior are recorded as qualifiers.
- [Model selection](/features/model-selection.md) — Model selection means the harness exposes a meaningful model choice for the run; a generic speed or quality label is partial unless it maps to documented behavior.
- [Models and context](/features/models-and-context.md) — Models and context is a capability family covering model choice, reasoning controls, usable token budgets, context visibility, compaction, fallback, local inference, and prompt-cache behavior.
- [Multi-file edit](/features/multi-file-edit.md) — Multi-file editing changes more than one file during a task. Review, approval, and source-control behavior are qualifiers rather than requirements of the term.
- [Native plugin system](/features/native-plugins.md) — A native plugin system is product-specific and does not establish support for the portable Agent Plugins manifest.
- [Nested AGENTS.md instructions](/features/nested-agents-md.md) — Nested AGENTS.md instructions is atomic and does not follow automatically from broader instructions support.
- [Nested subagents](/features/nested-subagents.md) — Nested subagents is atomic and does not follow automatically from broader subagents support.
- [Offline operation](/features/local-only-mode.md) — Offline operation means a documented product workflow can run without a required network connection; a desktop or CLI interface alone does not establish offline support.
- [Organization policy controls](/features/admin-policy-controls.md) — Organization policy controls are centrally enforced product settings; user preferences and prompt instructions are not organization policy.
- [Output token limit](/features/output-token-limit.md) — Output-token limits are recorded separately from total context because harnesses may reserve, cap, or truncate generated text and artifacts independently.
- [PDF input](/features/pdf-documents.md) — PDF input makes an uploaded PDF available as model input. Text extraction, visual page analysis, file limits, and attachment limits are qualifiers.
- [Per-subagent model selection](/features/subagent-model-selection.md) — Per-subagent model selection means a child can use a documented model choice or constraint distinct from its parent; a global conversation model picker is insufficient.
- [Persistent instructions](/features/instructions.md) — Persistent instructions are a capability family covering AGENTS.md, nested scopes, and native rule systems.
- [Plans and pricing](/features/plan-pricing-disclosure.md) — Plans and pricing records documented subscription and seat prices, billing periods, included usage, credits or overages, API-key billing, taxes, and custom enterprise terms without treating a model API rate as the price of a hosted product.
- [Plugin marketplace](/features/plugin-marketplace.md) — A plugin marketplace is a documented registry or marketplace for discovering and installing extensions. Update, removal, private-extension, and organization-policy behavior are qualifiers.
- [Plugin systems](/features/plugins.md) — Plugin support is a family of independent claims covering portable manifests, native extensions, packaged skills, packaged MCP, distribution, isolation, and vendor extensions.
- [Plugin vendor extensions](/features/plugin-vendor-extensions.md) — Plugin vendor extensions is atomic and does not follow automatically from broader plugins support.
- [Pricing, usage, and limits](/features/usage-and-reliability.md) — Pricing, usage, and limits is an internal catalog grouping for plan disclosure, usage metering, rate limits, quotas, and resumable runs.
- [Project rules](/features/native-rules.md) — Product-native rules is atomic and does not follow automatically from broader instructions support.
- [Project-scoped Agent Skills](/features/skill-project-scope.md) — Project-scoped Agent Skills is atomic and does not follow automatically from broader agent-skills support.
- [Prompt cache controls](/features/prompt-cache-controls.md) — Prompt-cache controls let an operator or configuration mark reusable segments or manage cache lifetime; automatic caching alone does not establish control.
- [Prompt cache telemetry](/features/prompt-cache-telemetry.md) — Prompt-cache telemetry exposes whether reuse occurred and, ideally, cached read/write tokens, cost, latency, and cache scope; documented caching without visibility is a separate claim.
- [Prompt caching](/features/prompt-cache-reuse.md) — Prompt caching reuses eligible repeated prompt content under documented latency or billing behavior. Browser, HTTP, file, and retrieval caching do not count.
- [Pull request integration](/features/git-pull-requests.md) — Pull request integration means a product documents opening or reviewing pull requests; unreviewed harness cells remain unknown.
- [Rate limits and quotas](/features/rate-limit-disclosure.md) — Rate limits and quotas are documented product usage boundaries; their scope and reset period are recorded as qualifiers.
- [Realtime voice](/features/realtime-voice.md) — Realtime voice supports live spoken input and audio responses. Turn-taking, interruption, latency, models, and session limits are qualifiers.
- [Reasoning effort controls](/features/reasoning-effort-controls.md) — Reasoning effort controls let an operator choose a documented reasoning mode or budget; model selection alone does not prove this capability.
- [Role-based access control (RBAC)](/features/role-based-access.md) — Role-based access control assigns permissions to users through roles; possession of a shared link alone is not RBAC.
- [Sandbox network access](/features/sandbox-network.md) — Sandbox network access means a product documents allowing, denying, or restricting outbound access from its execution sandbox.
- [Scheduled runs](/features/scheduled-runs.md) — Scheduled runs start saved tasks according to a schedule without a new human prompt. Hosting, limits, retention, and unattended-run behavior are qualifiers.
- [Screenshots](/features/screenshots.md) — Screenshot support captures a screen, window, or browser page for use as model input. Browser-only and arbitrary desktop capture are distinct qualifiers.
- [Secrets management](/features/secrets-management.md) — Secrets management is a documented product facility for storing credentials and providing them to tools or execution environments; plain-text configuration is not a secrets manager.
- [Security and privacy](/features/data-security-controls.md) — Security and privacy is an internal catalog grouping for retention, training-data use, data residency, key management, secrets, offline operation, organization policy, and audit logs.
- [Session resume](/features/resumable-runs.md) — Session resume means a product documents continuing a saved session; preserved state and cross-device behavior are qualifiers.
- [Shared projects](/features/shared-projects.md) — Shared projects are product-defined workspaces that multiple people can access; included context and permissions vary by product.
- [Streaming output](/features/streaming-output.md) — Streaming output displays text or tool events before a run completes. Event types, ordering, transport, and structured output are qualifiers.
- [Structured outputs](/features/structured-output.md) — Structured outputs are documented JSON or schema-constrained product outputs; unreviewed harness cells remain unknown.
- [Subagent context](/features/subagent-context-isolation.md) — Subagent context describes what a child receives from its parent. Fresh context, copied history, selected files, shared memory, and task-only context are recorded as distinct qualifiers.
- [Subagent delegation](/features/subagent-delegation.md) — Basic delegation establishes one child-agent boundary but not recursion, MCP access, or write access.
- [Subagent management](/features/subagent-lifecycle-controls.md) — Subagent management covers visibility into active children and documented cancel, pause, resume, retry, or limit behavior. Each operation is recorded separately as a qualifier.
- [Subagent MCP access](/features/subagent-mcp-access.md) — Subagent MCP access is atomic and does not follow automatically from broader subagents support.
- [Subagent permissions](/features/subagent-approval-boundaries.md) — Subagent permissions describe which tools and grants a child receives and how child actions obtain approval. Inheritance, restrictions, and escalation behavior are qualifiers.
- [Subagent results](/features/subagent-result-handoff.md) — Subagent results describe what a child returns to its parent. Text summaries, files, citations, diffs, status, errors, and structured result objects are recorded as separate qualifiers.
- [Subagent write access](/features/subagent-write-access.md) — Subagent write access is atomic and does not follow automatically from broader subagents support.
- [Subagents](/features/subagents.md) — Subagents are a family of separate claims covering delegation, nesting, concurrency, context isolation, model choice, tool and write permissions, approvals, lifecycle control, and result handoff.
- [Terminal commands](/features/terminal.md) — Terminal command support lets a product propose or execute shell commands in a project environment. Approval, shell, sandbox, working directory, and persistence are qualifiers.
- [Training data controls](/features/training-data-controls.md) — Training-data controls distinguish model training, service improvement, evaluation, safety review, and user feedback; consumer and business defaults may differ.
- [Usage metering](/features/usage-metering.md) — Usage metering reports documented consumption such as requests, tokens, or cost; available dimensions and reporting delay are qualifiers.
- [User-scoped Agent Skills](/features/skill-user-scope.md) — User-scoped Agent Skills is atomic and does not follow automatically from broader agent-skills support.
- [Video input](/features/video-input.md) — Video input means an uploaded video contributes model context. Frame, audio, transcript, timing, and format behavior are recorded as qualifiers only when documented.
- [Web fetch](/features/web-fetch.md) — Web fetch retrieves current web pages or API responses during a run. Authentication, network policy, rendering, and source attribution are qualifiers.
- [WebMCP](/features/webmcp.md) — WebMCP is a draft capability family covering imperative and declarative tools, discovery and execution, permissions, cancellation, and security annotations.
- [WebMCP declarative form tools](/features/webmcp-declarative-tools.md) — WebMCP declarative form tools is an atomic WebMCP capability and does not imply every other part of the draft.
- [WebMCP execution cancellation](/features/webmcp-cancellation.md) — WebMCP execution cancellation is an atomic WebMCP capability and does not imply every other part of the draft.
- [WebMCP imperative tools](/features/webmcp-imperative-tools.md) — WebMCP imperative tools is an atomic WebMCP capability and does not imply every other part of the draft.
- [WebMCP permissions policy](/features/webmcp-permissions-policy.md) — WebMCP permissions policy is an atomic WebMCP capability and does not imply every other part of the draft.
- [WebMCP security annotations](/features/webmcp-security-annotations.md) — WebMCP security annotations is an atomic WebMCP capability and does not imply every other part of the draft.
- [WebMCP tool discovery](/features/webmcp-tool-discovery.md) — WebMCP tool discovery is an atomic WebMCP capability and does not imply every other part of the draft.
- [WebMCP tool execution](/features/webmcp-tool-execution.md) — WebMCP tool execution is an atomic WebMCP capability and does not imply every other part of the draft.
- [Workspace files](/features/workspace-files.md) — Workspace file access means reading or changing files inside a user-selected project. Local or hosted scope, permissions, sandboxing, and persistence are qualifiers.
- [Workspace search](/features/workspace-search.md) — Workspace search locates files, text, or symbols across a project. Query types, indexing, scope, and semantic retrieval are qualifiers.