Claude Code Dynamic Workflows: Fan-Out, Priced
Three agents, three files, 71,835 tokens. The session that did the work itself was faster, four times cheaper, and the only fan-out arm that got every number right.
Sep 1312 min read
CATEGORY · 65 ARTICLES
AI Coding Assistants
Articles
AI Coding Assistants
Claude Code Nested Subagents: The Tool That Vanishes
The documented depth-limit refusal never fired. At the cap, the Agent tool is absent from the tool list and the agent says so in its own words.
Sep 1311 min read
Claude Code Background Tasks: Two Surfaces, Measured
One backgrounds the agent, one backgrounds a command. A 15-second job inside a 6.6-second session came back killed — twice, with a control that passed.
Sep 1211 min read
Claude Code GitHub Integration: What It Actually Buys
With no GitHub integration installed, a repository question took one tool call. A pull request question took three, crossed two shells, and returned NOT_AVAILABLE.
Sep 1211 min read
Claude Code Output Styles: The 750-Token Switch
A six-word style worked in all three arms. The material difference was whether Claude Code retained 750 tokens of coding instructions.
Sep 1010 min read
Claude Code Plugins: 2,369 Tokens Before Use
One Shopify plugin, 22 skills, two hooks, and a 2,369-token charge before invocation. The inventory is the installation decision.
Sep 1010 min read
Claude Code Statusline: 68 ms, No jq, No Tokens
A dependency-free Node status line that survives startup nulls, malformed JSON, and Windows. Twenty-five cold starts measured 67.6 ms median.
Sep 1010 min read
Claude Code Explore Subagent: Searching Without Your Rules
Explore is not measurably faster or cheaper than the general-purpose agent on a plain codebase search. It is the agent working without your project instructions, and it is not read-only.
Sep 911 min read