Merge remote-tracking branch 'origin/main' into feature/agent-avatar-support

docs/gateway/configuration.md

@@ -1280,6 +1280,18 @@ Default: `~/clawd`.
 If `agents.defaults.sandbox` is enabled, non-main sessions can override this with their
 own per-scope workspaces under `agents.defaults.sandbox.workspaceRoot`.
 
+### `agents.defaults.repoRoot`
+
+Optional repository root to show in the system prompt’s Runtime line. If unset, Clawdbot
+tries to detect a `.git` directory by walking upward from the workspace (and current
+working directory). The path must exist to be used.
+
+```json5
+{
+  agents: { defaults: { repoRoot: "~/Projects/clawdbot" } }
+}
+```
+
 ### `agents.defaults.skipBootstrap`
 
 Disables automatic creation of the workspace bootstrap files (`AGENTS.md`, `SOUL.md`, `TOOLS.md`, `IDENTITY.md`, `USER.md`, and `BOOTSTRAP.md`).
@@ -1983,7 +1995,7 @@ Per-agent override (further restrict):
 
 Notes:
 - `tools.elevated` is the global baseline. `agents.list[].tools.elevated` can only further restrict (both must allow).
-- `/elevated on|off` stores state per session key; inline directives apply to a single message.
+- `/elevated on|off|ask|full` stores state per session key; inline directives apply to a single message.
 - Elevated `exec` runs on the host and bypasses sandboxing.
 - Tool policy still applies; if `exec` is denied, elevated cannot be used.
 

docs/gateway/sandbox-vs-tool-policy-vs-elevated.md

@@ -91,7 +91,8 @@ Available groups:
 ## Elevated: exec-only “run on host”
 
 Elevated does **not** grant extra tools; it only affects `exec`.
-- If you’re sandboxed, `/elevated on` (or `exec` with `elevated: true`) runs on the host.
+- If you’re sandboxed, `/elevated on` (or `exec` with `elevated: true`) runs on the host (approvals may still apply).
+- Use `/elevated full` to skip exec approvals for the session.
 - If you’re already running direct, elevated is effectively a no-op (still gated).
 - Elevated is **not** skill-scoped and does **not** override tool allow/deny.
 

docs/gateway/security.md

@@ -178,6 +178,20 @@ Even with strong system prompts, **prompt injection is not solved**. What helps
 - Run sensitive tool execution in a sandbox; keep secrets out of the agent’s reachable filesystem.
 - **Model choice matters:** older/legacy models can be less robust against prompt injection and tool misuse. Prefer modern, instruction-hardened models for any bot with tools. We recommend Anthropic Opus 4.5 because it’s quite good at recognizing prompt injections (see [“A step forward on safety”](https://www.anthropic.com/news/claude-opus-4-5)).
 
+### Prompt injection does not require public DMs
+
+Even if **only you** can message the bot, prompt injection can still happen via
+any **untrusted content** the bot reads (web search/fetch results, browser pages,
+emails, docs, attachments, pasted logs/code). In other words: the sender is not
+the only threat surface; the **content itself** can carry adversarial instructions.
+
+When tools are enabled, the typical risk is exfiltrating context or triggering
+tool calls. Reduce the blast radius by:
+- Using a read-only or tool-disabled **reader agent** to summarize untrusted content,
+  then pass the summary to your main agent.
+- Keeping `web_search` / `web_fetch` / `browser` off for tool-enabled agents unless needed.
+- Enabling sandboxing and strict tool allowlists for any agent that touches untrusted input.
+
 ### Model strength (security note)
 
 Prompt injection resistance is **not** uniform across model tiers. Smaller/cheaper models are generally more susceptible to tool misuse and instruction hijacking, especially under adversarial prompts.
@@ -187,6 +201,7 @@ Recommendations:
 - **Avoid weaker tiers** (for example, Sonnet or Haiku) for tool-enabled agents or untrusted inboxes.
 - If you must use a smaller model, **reduce blast radius** (read-only tools, strong sandboxing, minimal filesystem access, strict allowlists).
 - When running small models, **enable sandboxing for all sessions** and **disable web_search/web_fetch/browser** unless inputs are tightly controlled.
+ - For chat-only personal assistants with trusted input and no tools, smaller models are usually fine.
 
 ## Reasoning & verbose output in groups