cc-viewer 1.7.22 → 1.8.0

package/concepts/en/Tool-REPL.md

@@ -0,0 +1,47 @@
+# REPL
+
+Executes JavaScript in a persistent Node.js vm context inside the session. Top-level `await` is supported, and variables/functions defined in one call remain available in later calls.
+
+## When to Use
+
+- Quick computation, data transformation, or JSON wrangling that is easier in code than in shell one-liners.
+- Multi-step scripting where intermediate state should persist between calls (counters, accumulated results).
+- Probing an API or library behavior interactively before writing it into a file.
+
+## Activation
+
+- Off by default — set `CLAUDE_CODE_REPL=true` to enable it.
+- In terminal (`cli`) and claude.ai (`remote`) sessions, a server-side feature flag may also enable it.
+- When off, REPL is hidden from the model's tool list. When on, `Read`, `Glob`, `Grep`, `Bash`, `PowerShell`, and `NotebookEdit` are replaced by REPL shorthands.
+
+## Parameters
+
+- `code` (string, required): JavaScript code to execute. Supports top-level await. State persists across calls.
+- `description` (string, optional): Clear, concise description of what this script does in active voice (5–10 words), e.g. "Trace upgrade message to its GrowthBook flag".
+- `timeout` (number, optional): Timeout in milliseconds. Defaults to 30000; maximum 600000.
+
+## Examples
+
+### Example 1: Compute and reuse state
+
+```
+REPL(code="const counts = new Map(); ['a','b','a'].forEach(k => counts.set(k, (counts.get(k)||0)+1)); counts.get('a')")
+```
+
+Returns `2`; `counts` stays defined for subsequent REPL calls in the same session.
+
+### Example 2: Top-level await with a longer timeout
+
+```
+REPL(
+  code="const res = await fetch('https://example.com/api'); await res.json()",
+  description="Fetch example API and parse JSON",
+  timeout=60000
+)
+```
+
+## Notes
+
+- State is per-session: restarting the session clears all definitions.
+- This is a JavaScript (Node) environment — use Bash for shell commands, filesystem-heavy work, or non-JS runtimes.
+- Long-running code should set an explicit `timeout`; the default 30s kills anything slower.

package/concepts/en/Tool-ReadMcpResource.md

@@ -0,0 +1,32 @@
+# ReadMcpResource
+
+Reads a single resource exposed by a connected MCP (Model Context Protocol) server, addressed by its URI.
+
+## When to Use
+
+- An MCP server advertises a resource (file, record, document) whose content you need in context.
+- You have a concrete resource URI — from `ListMcpResources`, from the server's documentation, or from a previous tool result.
+
+## Activation
+
+- Always enabled, but not exposed to the model's tool list — intended for thin-client / sidecar use.
+
+## Parameters
+
+- `server` (string, required): The MCP server name.
+- `uri` (string, required): The resource URI to read.
+
+## Examples
+
+### Example 1: Read a server resource by URI
+
+```
+ReadMcpResource(server="github", uri="file:///repo/docs/architecture.md")
+```
+
+Returns the resource content as provided by the `github` MCP server.
+
+## Notes
+
+- Use `ListMcpResources` first if you don't know which resources a server exposes; use `ReadMcpResourceDir` for directory-style listings.
+- The URI scheme is server-specific (`file://`, `https://`, custom schemes) — check what the target server advertises.

package/concepts/en/Tool-ReadMcpResourceDir.md

@@ -0,0 +1,32 @@
+# ReadMcpResourceDir
+
+Lists the entries of a directory-style resource exposed by a connected MCP server, addressed by its URI.
+
+## When to Use
+
+- An MCP server organizes resources hierarchically and you need to enumerate one level of that hierarchy.
+- You want to browse before reading individual resources with `ReadMcpResource`.
+
+## Activation
+
+- Always enabled, but not exposed to the model's tool list — intended for thin-client / sidecar use.
+
+## Parameters
+
+- `server` (string, required): The MCP server name.
+- `uri` (string, required): The directory resource URI to list.
+
+## Examples
+
+### Example 1: List a resource directory
+
+```
+ReadMcpResourceDir(server="filesystem", uri="file:///project/src/")
+```
+
+Returns the child entries the server exposes under that directory URI.
+
+## Notes
+
+- Only servers that model their resources as directories support this; flat servers will return an error or an empty listing — fall back to `ListMcpResources`.
+- Combine with `ReadMcpResource` to drill into the entries that look relevant.

package/concepts/en/Tool-ReadNotifications.md

@@ -0,0 +1,27 @@
+# ReadNotifications
+
+Reads notifications queued for the assistant in the current session — GitHub activity on subscribed PRs (`github_webhook`), scheduled-trigger fires (`trigger_fire`), and messages arriving from other Claude sessions (`mcp_send_message`).
+
+## When to Use
+
+- You were notified that something happened — a subscribed PR updated, a scheduled trigger fired, another session messaged you — and need the actual payload.
+- Draining a backlog: large batches are returned in parts, so keep calling until the result reports 0 `remaining`.
+
+## Parameters
+
+This tool takes no parameters.
+
+## Examples
+
+### Example 1: Drain pending notifications
+
+```
+ReadNotifications()
+```
+
+Returns queued notifications oldest first. The result includes a `remaining` count of notifications still queued after this drain — call the tool again to read them.
+
+## Notes
+
+- Drains are size-budgeted: a follow-up call returns the rest of the SAME queue (plus anything newly arrived), not only new arrivals. Loop until `remaining` is 0.
+- Notifications originate from GitHub webhooks on subscribed PRs, scheduled triggers, and messages from other Claude sessions; there is no filtering parameter in the current version.

package/concepts/en/Tool-RemoteTrigger.md

@@ -10,6 +10,13 @@ Call the claude.ai remote-trigger API to manage scheduled and on-demand trigger
 - Listing or auditing all current triggers to review their configuration and status
 - Updating trigger settings such as schedule, payload, or description without recreating the trigger
 
+## Activation
+
+- Requires a claude.ai Pro, Max, Team, or Enterprise plan.
+- Not available on Amazon Bedrock, Claude Platform on AWS, Google Cloud, or Microsoft Foundry.
+- Requires server-side feature flags and the `allow_remote_sessions` / `allow_routines` policy settings.
+- Not for remote sessions themselves.
+
 ## Parameters
 
 - `action` (string, required): the operation to perform — one of `list`, `get`, `create`, `update`, or `run`

package/concepts/en/Tool-ScheduleWakeup.md

@@ -10,6 +10,11 @@ Schedule when to resume work in `/loop` dynamic mode. The tool lets Claude self-
 - Running an autonomous loop with no user prompt — pass the literal sentinel `<<autonomous-loop-dynamic>>` as `prompt`
 - Polling a process that is about to change state soon (stay cache-warm with a short delay)
 
+## Activation
+
+- Not available on Amazon Bedrock, Claude Platform on AWS, Google Cloud Agent Platform, or Microsoft Foundry.
+- Also off when feature-flag fetching is disabled (telemetry/traffic env vars).
+
 ## Parameters
 
 - `delaySeconds` (number, required): Seconds from now to resume. The runtime clamps the value to `[60, 3600]`, so you do not need to clamp it yourself.

package/concepts/en/Tool-SearchMcpRegistry.md

@@ -0,0 +1,31 @@
+# SearchMcpRegistry
+
+Searches the MCP connector registry by keyword to discover connectors that might help complete the task.
+
+## When to Use
+
+- The task would benefit from an external service (a database, an issue tracker, a SaaS API) and you want to check whether an MCP connector exists for it.
+- The user names a product and asks to connect it — search the registry for a matching connector.
+
+## Activation
+
+- Only available in remote (claude.ai) sessions on the first-party API.
+
+## Parameters
+
+- `keywords` (array of strings, required): Keyword phrases describing the user's intent or a named product. 1–8 items, each 1–64 characters.
+
+## Examples
+
+### Example 1: Find a connector for a named product
+
+```
+SearchMcpRegistry(keywords=["linear", "issue tracker"])
+```
+
+Returns registry entries whose connectors match the keywords. Resolve full connector details with `SuggestConnectors`.
+
+## Notes
+
+- Read-only and concurrency-safe; results are capped in size.
+- Searching installs nothing — it is purely discovery.

package/concepts/en/Tool-SendFeedback.md

@@ -0,0 +1,38 @@
+# SendFeedback
+
+Sends structured feedback about Claude Code to Anthropic — bug reports, feature ideas, or missing capabilities — without leaving the session.
+
+## When to Use
+
+- The user asks to report a bug or send feedback about Claude Code itself.
+- You hit a clear product defect (broken command, wrong behavior, crash) worth reporting.
+- The user describes a feature they wish existed (an idea or missing capability).
+
+## Parameters
+
+- `type` (string, required): One of `bug`, `idea`, `missing_capability`.
+- `title` (string, required): Short, specific one-line summary of the issue.
+- `details` (string, required): Labeled bullets, in order: **What happened:** (observed vs. expected, exact error text if short); **What the user said:** (quoted, or "User didn't comment; observed by the model."); **Repro:** (minimal steps); **Evidence:** (request IDs, timestamps, paths, versions — omit if none); optionally a final **Cause:** only if verified in-session. One to three lines per bullet; no narrative paragraphs, no speculation, no secrets.
+- `area` (string, optional): Short tag naming the part of Claude Code this is about (e.g. "hooks config", "/help", "file editing"). Leave blank if unclear.
+- `failure_mode` (string, optional): For model-behavior reports, the closest failure mode (e.g. `instruction_following`, `repetition_and_looping`, `context_and_memory`, `stopping_short`, or `other`). Omit only when the report is a pure product/tool bug.
+- `task_category` (string, optional): What the session was doing when the issue occurred: `code_edit`, `debug`, `explain`, `plan`, `shell`, `search`, `review`, or `other`.
+
+## Examples
+
+### Example 1: Report a product bug
+
+```
+SendFeedback(
+  type="bug",
+  title="/export truncates the last message",
+  details="**What happened:** exported transcript is missing the final assistant message.\n**What the user said:** \"the last reply never shows up in the file\".\n**Repro:** run /export after any multi-turn session.\n**Evidence:** v2.1.233, macOS.",
+  area="/export",
+  task_category="other"
+)
+```
+
+## Notes
+
+- Never include secrets, tokens, or private user data in `details`.
+- Quote the user's words when available; otherwise state that the model observed the issue.
+- Keep the report factual — speculation about root cause belongs in `**Cause:**` only when verified in-session.

package/concepts/en/Tool-SendFile.md

@@ -0,0 +1,36 @@
+# SendFile
+
+Sends one or more files to another Claude Code session — a peer listed by `ListAgents`, or an explicit session address.
+
+## When to Use
+
+- A peer session needs a file from your working directory (a report, a patch, a fixture) to continue its own task.
+- You are coordinating work across sessions and want to hand off artifacts, not just text (use `SendMessage` for text).
+
+## Activation
+
+- Cross-session file transfer must be available in the session; when it is not, validation fails with "Cross-session file transfer is not available in this session."
+- Gated by the same cross-session messaging conditions as `ListAgents` (server-side feature flags, off by default).
+
+## Parameters
+
+- `to` (string, required): Recipient — a peer session name from `ListAgents`, or an explicit `uds:<socket>` / `bridge:<session id>` address.
+- `files` (array of strings, required): File paths (absolute or relative to cwd) to send. Always pass an array, even for a single file. 1–16 files, at most 30 MiB each.
+- `message` (string, optional): Short message delivered alongside the files.
+
+## Examples
+
+### Example 1: Send a report to a peer session
+
+```
+SendFile(
+  to="teammate-a",
+  files=["./dist/report.html"],
+  message="The analysis you asked for"
+)
+```
+
+## Notes
+
+- Transfers to remote machines may require additional approval.
+- Reading the file contents is part of the send — denied if file reads are disabled by permission rules.

package/concepts/en/Tool-SendMessage.md

@@ -10,6 +10,11 @@ Delivers a message from one team member to another within an active team, or bro
 - Replying to a protocol prompt such as a shutdown request or a plan-approval request from the team leader.
 - Closing the loop at the end of a delegated task so the leader can mark the item complete.
 
+## Activation
+
+- Gated by the same cross-session messaging conditions as `ListAgents` (server-side feature flags, off by default).
+- Teammates additionally require experimental agent teams to be enabled.
+
 ## Parameters
 
 - `to` (string, required): The target teammate's `name` as registered in the team, or `*` to broadcast to all teammates at once.

package/concepts/en/Tool-SendUserFile.md

@@ -0,0 +1,38 @@
+# SendUserFile
+
+Sends one or more files to the user — generated artifacts, screenshots, reports — with control over how the client presents them.
+
+## When to Use
+
+- You produced a file the user needs (a report, an image, an HTML page) and want to surface it, not just mention its path.
+- Replying with an attachment (`status="normal"`), or proactively surfacing something the user hasn't asked for but needs to see now (`status="proactive"`).
+
+## Activation
+
+- Only available when a Remote Control client is connected, or the session runs in a managed cloud environment (e.g. Claude Code on the web).
+- Not available on Amazon Bedrock, Google Cloud, or Microsoft Foundry.
+- Requires the session to allow sending files (a settings/feature-gated capability); not offered in brief mode.
+
+## Parameters
+
+- `files` (array of strings, required): File paths (absolute or relative to cwd) to send to the user. Always pass an array, even for a single file.
+- `caption` (string, optional): Short caption for the file(s).
+- `status` (string, required): `proactive` when surfacing a file the user hasn't asked for and needs to see now — a generated artifact, a completed report; `normal` when replying to something the user just said.
+- `display` (string, optional): `render` opens the file inline in the side panel (HTML, SVG, Mermaid, images, PDFs); `attach` shows a download card only (deliverables the user will save and open elsewhere). Omit to let the client decide by file type.
+
+## Examples
+
+### Example 1: Deliver a generated report
+
+```
+SendUserFile(
+  files=["./out/weekly-report.html"],
+  caption="Weekly usage report",
+  status="proactive",
+  display="render"
+)
+```
+
+## Notes
+
+- Choose `display="attach"` for files the user saves and opens in another app; `render` for anything they should look at immediately.

package/concepts/en/Tool-SendUserMessage.md

@@ -0,0 +1,38 @@
+# SendUserMessage
+
+Sends a message to the user — the primary visible output channel in brief-style sessions. Also known by its legacy alias `Brief`.
+
+## When to Use
+
+- Replying to something the user just said (`status="normal"`).
+- Proactively surfacing something the user hasn't asked for and needs to see now — a task completing while they're away, a blocker you hit, an unsolicited status update (`status="proactive"`).
+
+## Activation
+
+- Hidden in interactive sessions by default; most interactive CLI sessions talk to the user directly instead.
+- Enabled in brief mode or via server-side feature flags.
+
+## Parameters
+
+In brief mode:
+
+- `message` (string, required): The message for the user. Supports markdown formatting.
+- `attachments` (array, optional): Attachments shown alongside the message. Each entry is either a file path (absolute or relative to cwd) for a locally readable file, or a pre-resolved `{file_uuid, file_name, size, is_image}` object obtained from a device tool such as `attach_file`.
+- `status` (string, required): `proactive` for unsolicited updates the user needs now; `normal` when replying to the user.
+
+In non-brief builds only `message` is available.
+
+## Examples
+
+### Example 1: Proactive completion notice
+
+```
+SendUserMessage(
+  message="The migration finished — 42 files updated, tests green.",
+  status="proactive"
+)
+```
+
+## Notes
+
+- Use `proactive` sparingly — it is meant for things that genuinely need the user's attention now.

package/concepts/en/Tool-SuggestConnectors.md

@@ -0,0 +1,28 @@
+# SuggestConnectors
+
+Resolves full connector payloads for `directoryUuid` values returned by `SearchMcpRegistry`, so the user can be offered concrete connectors to enable.
+
+## When to Use
+
+- After `SearchMcpRegistry` returns candidate connectors, to fetch their full details for presentation.
+
+## Activation
+
+- Only available in remote (claude.ai) sessions on the first-party API.
+
+## Parameters
+
+- `uuids` (array of strings, required): `directoryUuid` or `server_id` values to resolve. 1–32 items, each 1–64 characters.
+
+## Examples
+
+### Example 1: Resolve two registry hits
+
+```
+SuggestConnectors(uuids=["d290f1ee-6c54-4b01-90e6-d701748f0851", "a1b2c3d4-0000-4000-8000-abcdefabcdef"])
+```
+
+## Notes
+
+- Never guess UUIDs — only resolve identifiers that came back from `SearchMcpRegistry`.
+- The tool connects nothing itself; enabling a connector happens out of band.

package/concepts/en/Tool-SuggestPluginInstall.md

@@ -0,0 +1,39 @@
+# SuggestPluginInstall
+
+Renders an inline plugin install card from `SearchPlugins` results, tying plugin suggestions to the user's request.
+
+## When to Use
+
+- A plugin search surfaced plugins that match what the user is trying to do, and you want to offer them for installation.
+
+## Activation
+
+- Only when a Remote Control client is connected, or the session runs in a managed cloud environment.
+- Disabled under HIPAA enterprise configurations.
+- Not in brief mode.
+
+## Parameters
+
+- `contextLabel` (string, required): Short header tying the suggestion to the user request (max 128 characters).
+- `plugins` (array, required): Plugins sourced from `SearchPlugins` results — 1–16 entries, each with:
+  - `pluginId` (string, required)
+  - `pluginName` (string, required)
+  - `description` (string, required)
+  - `skills` (array, optional): Up to 32 `{name, description?}` entries describing the plugin's skills.
+
+## Examples
+
+### Example 1: Offer a matching plugin
+
+```
+SuggestPluginInstall(
+  contextLabel="For reviewing pull requests",
+  plugins=[{pluginId="pr-toolkit", pluginName="PR Toolkit", description="Review helpers"}]
+)
+```
+
+The card is rendered for the user; enabling the plugin happens out of band. Call `ListPlugins` on follow-up to discover what was actually installed.
+
+## Notes
+
+- Only include plugins that came from search results — never invent plugin entries.

package/concepts/en/Tool-SuggestSkills.md

@@ -0,0 +1,33 @@
+# SuggestSkills
+
+Renders a card of standalone skills the user can add (skills that are not yet enabled), based on topic keywords.
+
+## When to Use
+
+- The user's request matches skills they don't have enabled (`trigger="user_asked"` when they asked, `trigger="proactive"` when you suggest unprompted).
+
+## Activation
+
+- Only when a Remote Control client is connected, or the session runs in a managed cloud environment.
+- Disabled under HIPAA enterprise configurations.
+- Not in brief mode.
+
+## Parameters
+
+- `keywords` (array of strings, required): Topic keywords from the user's request. 1–8 items, each 1–64 characters.
+- `contextLabel` (string, optional): Short label tying the suggestion to the request (max 128 characters).
+- `trigger` (string, optional): How this suggestion started — `user_asked` or `proactive`.
+
+## Examples
+
+### Example 1: Suggest skills by topic
+
+```
+SuggestSkills(keywords=["data visualization", "charts"], contextLabel="For building the dashboard", trigger="user_asked")
+```
+
+Already-enabled skills are filtered out of the result.
+
+## Notes
+
+- Renders a suggestion card only — adding a skill happens out of band; call `ListSkills` afterwards to confirm.

package/concepts/en/Tool-TaskCreate.md

@@ -12,6 +12,12 @@ Creates a new task in the current team's task list (or the session's task list w
 
 Skip `TaskCreate` for trivial one-shot actions, pure conversation, or anything completable in two or three direct tool calls.
 
+## Activation
+
+- Available by default on most models.
+- Not available on Opus 4.8 / Sonnet 5 / Fable 5 / Mythos 5 and later families (v2.1.233+) unless opted in via `CLAUDE_CODE_ENABLE_TODO_TOOLS=1`, `--allowedTools`, or `--tools`.
+- The whole task system is disabled when `CLAUDE_CODE_ENABLE_TASKS=false`.
+
 ## Parameters
 
 - `subject` (string, required): Short imperative title, e.g. `Fix login redirect on Safari`. Keep it under roughly eighty characters.

package/concepts/en/Tool-TaskGet.md

@@ -12,6 +12,12 @@ Fetches the full record for a single task by ID, including its description, curr
 
 Prefer `TaskList` when you only need a high-level scan; reserve `TaskGet` for the specific record you intend to read carefully or modify.
 
+## Activation
+
+- Available by default on most models.
+- Not available on Opus 4.8 / Sonnet 5 / Fable 5 / Mythos 5 and later families (v2.1.233+) unless opted in via `CLAUDE_CODE_ENABLE_TODO_TOOLS=1`, `--allowedTools`, or `--tools`.
+- The whole task system is disabled when `CLAUDE_CODE_ENABLE_TASKS=false`.
+
 ## Parameters
 
 - `taskId` (string, required): The task identifier returned by `TaskCreate` or `TaskList`. IDs are stable for the life of the task.

package/concepts/en/Tool-TaskList.md

@@ -12,6 +12,12 @@ Returns every task in the current team (or session) in summarised form. Use it t
 
 `TaskList` is read-only and cheap; call it freely whenever you need an overview.
 
+## Activation
+
+- Available by default on most models.
+- Not available on Opus 4.8 / Sonnet 5 / Fable 5 / Mythos 5 and later families (v2.1.233+) unless opted in via `CLAUDE_CODE_ENABLE_TODO_TOOLS=1`, `--allowedTools`, or `--tools`.
+- The whole task system is disabled when `CLAUDE_CODE_ENABLE_TASKS=false`.
+
 ## Parameters
 
 `TaskList` takes no parameters. It always returns the full task set for the active context.

package/concepts/en/Tool-TaskUpdate.md

@@ -10,6 +10,12 @@ Modifies an existing task — its status, content, ownership, metadata, or depen
 - Recording newly discovered dependencies with `addBlocks` / `addBlockedBy`.
 - Attaching structured `metadata` such as external ticket IDs or priority hints.
 
+## Activation
+
+- Available by default on most models.
+- Not available on Opus 4.8 / Sonnet 5 / Fable 5 / Mythos 5 and later families (v2.1.233+) unless opted in via `CLAUDE_CODE_ENABLE_TODO_TOOLS=1`, `--allowedTools`, or `--tools`.
+- The whole task system is disabled when `CLAUDE_CODE_ENABLE_TASKS=false`.
+
 ## Parameters
 
 - `taskId` (string, required): The task to modify. Obtain from `TaskList` or `TaskCreate`.

package/concepts/en/Tool-TodoWrite.md

@@ -0,0 +1,43 @@
+# TodoWrite
+
+Writes a structured todo list for the current session, replacing the previous list. Each item carries its text, a status, and a present-continuous form shown in progress indicators.
+
+## When to Use
+
+- A task has several distinct steps and tracking them helps you (and the user) see progress.
+- The user explicitly asks for a todo list.
+- You want to mark exactly one item as in progress while the rest stay pending or completed.
+
+## Activation
+
+- Legacy tool: disabled by default in sessions that offer the Task tools (`TaskCreate`, `TaskUpdate`, `TaskList`).
+- Re-enable it with `CLAUDE_CODE_ENABLE_TASKS=0`.
+
+## Parameters
+
+- `todos` (array, required): The complete updated todo list. Each entry has:
+  - `content` (string): The task description.
+  - `status` (string): One of `pending`, `in_progress`, `completed`.
+  - `activeForm` (string): Present-continuous text shown while the item is in progress (e.g. "Running tests").
+
+## Examples
+
+### Example 1: Track a three-step change
+
+```
+TodoWrite(
+  todos=[
+    {content="Update the parser", status="in_progress", activeForm="Updating the parser"},
+    {content="Add unit tests", status="pending", activeForm="Adding unit tests"},
+    {content="Run the full test suite", status="pending", activeForm="Running the full test suite"}
+  ]
+)
+```
+
+The whole list is rewritten on every call — always include all items, not just the ones that changed.
+
+## Notes
+
+- The list is replaced wholesale on each call; to update one item, resubmit every item with the new status.
+- Keep exactly one item `in_progress` at a time.
+- In sessions where the structured task tools (`TaskCreate`/`TaskUpdate`/`TaskList`) are enabled, the harness may offer those instead of `TodoWrite` — prefer whichever tool set is advertised.

package/concepts/en/Tool-ToolSearch.md

@@ -10,6 +10,11 @@ Fetches the full schema definitions for "deferred tools" on demand so they becom
 
 If a tool's schema is already in context, do not search again — just call it.
 
+## Activation
+
+- On by default.
+- Off when `ANTHROPIC_BASE_URL` points at a non-Anthropic endpoint (unless `ENABLE_TOOL_SEARCH` is set), when `CLAUDE_CODE_DISABLE_EXPERIMENTAL_BETAS` is set, when the model lacks tool-reference support (Vertex AI models before Claude 4.5), or when denied via `"deny": ["ToolSearch"]`.
+
 ## Parameters
 
 - `query` (string, required): The query used to locate deferred tools. Three forms are supported:

package/concepts/en/Tool-WebSearch.md

@@ -10,6 +10,12 @@ Performs a live web search and returns ranked results that the assistant uses to
 - Verifying a fact that may have changed since the model was trained.
 - Discovering multiple perspectives on a topic before fetching any single page with `WebFetch`.
 
+## Activation
+
+- Availability is provider- and model-dependent: available on the Anthropic API and Claude Platform on AWS; on Microsoft Foundry it requires an Anthropic-hosted deployment; on Google Cloud it works with Claude 4+ models.
+- Not available on Amazon Bedrock.
+- Cap 200 calls per session with `CLAUDE_CODE_MAX_WEB_SEARCHES_PER_SESSION`.
+
 ## Parameters
 
 - `query` (string, required): The search query. Minimum length 2 characters. Include the current year when asking about "latest" or "recent" information so results are fresh.