axiopera: run other coding agents (claude, codex, grok) as your sub-agents.
Use axiopera for a different agent or model, or for work that must continue after your
session ends. For same-agent delegation that needs your full context, use your built-in
sub-agents: axiopera sub-agents see only the task you write.
Dispatch
axiopera spawn codex "Fix the failing test in tests/parser.rs; run cargo test before finishing."
axiopera spawn claude --tier read --name review -f task.md
Tiers: read (look, don't change; MCP: read-only tools only, none on Grok), work (default: the agent's own auto mode), full (the agent's full-access mode).
Options: --tier, --model, --effort, --name (use it instead of the ID), --cwd, --timeout 30m,
--worktree (an isolated git worktree).
Get results
axiopera wait blocks until one of your sub-tasks finishes a turn or hits a problem,
then prints it. Each result is delivered once.
- Claude Code: run `axiopera wait` as a background command; you are woken when it exits.
- Codex: if spawn says to use wait, poll with `axiopera wait --timeout 60s`.
- Grok: run it in the background, or use `axiopera watch` with your monitor tool.
Exit codes: 0 success, 1 command error, 3 timed out, 4 nothing left to wait for,
5 something failed, 6 something was cancelled.
Talk to a running sub-agent
axiopera send <task> "Also update the docs." steers the running turn, or starts a new turn
in the same session if it already finished.
Results note when they may predate your message; late messages are handled in the next turn.
Answer questions
Questions arrive like results; sub-agents can keep working until you answer.
axiopera reply <question> "Use port 8080." reaches it mid-turn, or resumes it.
Approval requests pause the sub-agent until you decide; unanswered requests are denied
after the configured approval timeout.
axiopera approve <request> axiopera deny <request> --reason "Use a dry run instead."
For long messages, use `send <task> -f msg.md` or `reply <question> -f answer.md`; use `-` to read stdin.
Sub-agents can talk to each other: give them each other's task IDs, and they can run
`axiopera tell <task> "..."` (Codex sub-agents have it as the `tell` tool). You get a copy of
every note.
Inspect and stop
axiopera status your sub-tasks axiopera log <task> recent events
axiopera result <task> latest turn result axiopera cancel <task> stop it and its children
axiopera detach <task> hand it to the user: prints the native command to continue it