lkn
/
ollie-9p
Archived
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Add Sub-Agents section to Multi-Agent Workflows 

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
This commit is contained in:
Levi Neely 2026-04-10 02:59:45 +02:00
parent a42a7b7624
commit 272e095bc5
1 changed files with 6 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

6
README.md
View File

@ -196,6 +196,12 @@ done
Each agent only sees what the script explicitly sends it. Each can use a different backend, model, or tool configuration. A human can intervene at any point by writing directly to a session's `prompt`. The boundary between fully automated and human-in-the-loop is just whether the script pauses to ask. Each agent only sees what the script explicitly sends it. Each can use a different backend, model, or tool configuration. A human can intervene at any point by writing directly to a session's `prompt`. The boundary between fully automated and human-in-the-loop is just whether the script pauses to ask.
### Sub-Agents
A session can spawn ephemeral child sessions to handle a focused task, then tear them down when done. The parent uses `execute_code` to write `new` to `ctl`, send an initial context and task to the child's `prompt`, poll until `reply` is ready, read the result, and write `kill <id>` to `ctl`. From the child's perspective it is just a normal session; it has no knowledge of its own ephemerality.
**Sub-agents as function calls** — the parent primes the child with exactly the context it needs (excerpts from its own `chat`, relevant beads, local file contents) and receives a single focused reply. Failures are isolated; a child that errors or stalls can be killed and retried without affecting the parent. The pattern composes naturally: a sub-agent can itself spawn sub-agents, building a call tree bounded only by the number of sessions open at once.
### Self-Generating Workflows ### Self-Generating Workflows
Since agents have access to `execute_code`, a session can write and execute a workflow script without any human involvement. Given a task and knowledge of the filesystem layout, an agent can decompose the work, spawn sessions, write the coordination script, and run it — all in a single turn. The README you are reading is essentially its system prompt. Conductor functionality, for free. Since agents have access to `execute_code`, a session can write and execute a workflow script without any human involvement. Given a task and knowledge of the filesystem layout, an agent can decompose the work, spawn sessions, write the coordination script, and run it — all in a single turn. The README you are reading is essentially its system prompt. Conductor functionality, for free.