rust-tokio

Files (1)

• SKILL.md

SKILL.md (first lines)

# Rust Tokio

## When to use

Select for asynchronous process ownership, transport, cancellation and timers.

## Procedure

1. Read root AGENTS.md, prompt 04 and the scoped Tokio instructions.
2. Use the existing current-thread runtime; never create an executor per
   request, tool call or seat. Keep synchronous leaves executor-independent.
3. Register notifications before checking state. Recheck cancellation after
   cleanup and before effects, errors or success; first cancellation reason wins.
4. Own every child/task through cleanup. Close stdin, bound both output
   streams, kill and await owned children on interruption, and report unknown
   cleanup without claiming release. Child exit is not descendant containment.
5. Do not detach writes, model calls, process drains or cleanup. Join bounded
   work; propagate failures and preserve observed failure evidence.
6. Keep locks out of await points. No unsafe Rust, explanatory comments,
   speculative concurrency, background swarms or extra task boards.
7. Run real cancellation/overflow/late-result controls through native Cargo
   tests. Consumer Node checks cannot substitute for Rust verification.

## Learnings

Wrong: release the writer because the primary transport reports no uncertainty.
Right: any trusted `ROSTER_PROCESS_CLEANUP`, including metadata/origin Git,
keeps ownership quarantined. Pair a real-lock retention control with settled
failure release, and label injected errors as boundary tests, not OS-failure proof.

## Stop condition

Stop on failed ownership, unknown cleanup, policy denial or missing checks.
This packet grants no model endpoint, network, credential or publishing access.

From R4yHarris/github-agent-roster, MIT.

More from R4yHarris/github-agent-roster