Follow Discord
Sweep 25 Sep 2026 · 19:33Z Build v2.1.283 504 read Stable v2.1.274 Latest v2.1.283 Next v2.1.283 Feeds RSS JSON llms.txt llms-full.txt Unofficial

Claude Code v2.1.281 ·

Self-hosted runner: server-composed agent files delivered via --agents file; system prompts delivered by file

The self-hosted runner now checks agent_files, passes agents via --agents <file>, and passes system prompts as files

Entry
JSON All of v2.1.281
EntryKind: in v2.1.281,
ChangesSection of the release
What

The remote or self-hosted runner is the program that starts Claude Code on a machine you host. It now accepts server-supplied subagent definitions in agent_files. A subagent is a helper agent that Claude Code can hand work to. The runner checks these definitions:

  • no more than 64 entries
  • each name matches ^[a-z][a-z0-9-]{0,63}$: it starts with a lowercase letter and uses only lowercase letters, digits and hyphens, up to 64 characters
  • each file's content is at most 1 MiB
  • no two entries share a name

The runner combines these with the server's agents argument. It writes the result to remote-agents.json, which only the file's owner can read or write (mode 0600), and starts Claude Code with the --agents flag pointing at that file. The system prompt, append-system-prompt and append-subagent-system-prompt values are also written to files and passed as --<key>-file. A system prompt is the standing instructions the model receives before your messages.

The runner also sorts out errors that happen when Claude Code fails to start, such as E2BIG (the arguments were too large). It records the length of the longest argument and the longest environment entry. It now drops environment entries that are malformed or that are variants of names the runner itself controls.

Why

Large agent definitions and prompts now travel in files, not on the command line. This avoids start-up failures when the command line gets too long. When start-up does fail, the records say more about why.

What the documentation says
--agents Create custom subagents modified, high confidence The `--agents` flag accepts JSON with a `prompt` field plus these [frontmatter](#supported-frontmatter-fields) fields: `description`, `tools`, `disallowedTools`, `model`, `permissionMode`, `mcpServers`, `hooks`, `maxTurns`, `skills`, `init… see the edit
How sure we are
Two sources agreeTwo things we can check say the same as this entry.
Anthropic's documentation agrees--agents on Create custom subagents
Anthropic's release notes agreeImproved --agents to accept the path to a JSON file (with -p) as well as inline JSON, and to allow an empty prompt

See this entry in the whole of v2.1.281 →

Feedback