diff --git a/docs/claude-code/context.md b/docs/claude-code/context.md
index c3683e5..9aa7636 100644
--- a/docs/claude-code/context.md
+++ b/docs/claude-code/context.md
@@ -35,7 +35,7 @@ Claude Code enables extended thinking by default, giving Claude space to reason
 
 ### Effort levels
 
-On supported models (Sonnet 4.6, Sonnet 5, Opus 4.6, Opus 4.7, Opus 4.8, and Fable 5), thinking uses **adaptive reasoning**: the model dynamically allocates thinking tokens based on your effort level and the task at hand. This means Claude responds faster to routine prompts and reserves deeper thinking for steps that benefit from it.
+On supported models (Sonnet 4.6, Sonnet 5, Opus 4.6, Opus 4.7, Opus 4.8, Opus 5, and Fable 5), thinking uses **adaptive reasoning**: the model dynamically allocates thinking tokens based on your effort level and the task at hand. This means Claude responds faster to routine prompts and reserves deeper thinking for steps that benefit from it.
 
 | Way to adjust | How |
 |---------------|-----|
diff --git a/docs/claude-code/flags.md b/docs/claude-code/flags.md
index d9f98d6..954e5da 100644
--- a/docs/claude-code/flags.md
+++ b/docs/claude-code/flags.md
@@ -123,5 +123,6 @@ keywords: [Claude Code CLI flags, command line options, --continue, --resume, --
 | `CLAUDE_CODE_DISABLE_AUTO_MEMORY` | Set to `1` to disable auto memory globally |
 | `CLAUDE_CODE_MAX_WEB_SEARCHES_PER_SESSION` | Tune the per-session cap on WebSearch calls (default `200`) that stops runaway loops |
 | `CLAUDE_CODE_MAX_SUBAGENTS_PER_SESSION` | Tune the per-session cap on subagent spawns (default `200`) that stops runaway loops |
+| `CLAUDE_CODE_MAX_SUBAGENT_SPAWN_DEPTH` | Set how many layers of nested subagents can spawn below the main conversation (default `3` since v2.1.219) |
 | `CLAUDE_CODE_MCP_AUTO_BACKGROUND_MS` | Tune or disable the threshold after which a long-running MCP tool call moves to the background automatically (default two minutes) so the session stays usable |
 | `ANTHROPIC_MODEL` | Override the default model for all sessions |
diff --git a/docs/claude-code/subagents.md b/docs/claude-code/subagents.md
index 1e6f596..c8a933b 100644
--- a/docs/claude-code/subagents.md
+++ b/docs/claude-code/subagents.md
@@ -734,7 +734,7 @@ When a background subagent needs a permission decision, the prompt surfaces in y
 
 Since v2.1.172, a subagent can spawn its own subagents. This helps when a delegated task itself splits into parallel subtasks, such as a reviewer subagent that dispatches a verifier per finding: the intermediate output never reaches your main conversation, and only the top-level subagent's summary returns to you. The subagent panel below the prompt shows the full tree, with a descendant count and a path back to `main` on each row.
 
-Subagent chains are capped at **five levels deep** to prevent runaway concurrent trees. A subagent at depth five does not receive the Agent tool and cannot spawn further. The limit is fixed and not configurable.
+By default, a subagent can spawn its own subagents up to **three layers below the main conversation**. At the depth limit, a subagent does not receive the Agent tool and cannot spawn further, so it does its delegated work itself and returns one summary. To change the limit, set the `CLAUDE_CODE_MAX_SUBAGENT_SPAWN_DEPTH` environment variable (v2.1.217+) to the number of layers you want below the main conversation. The default was one from v2.1.217 through v2.1.218, and v2.1.219 raised it to three.
 
 ### Parallel subagents