# Ordered Streaming Tool Execution

> Buffer and execute streaming tool calls with concurrency awareness so results appear in request order even when multiple runners overlap.

- Skill: `vishuwa2004/ordered-streaming-tool-execution` (Agent Skill, multi-file: 2 files)
- Install (CLI): `npx skillmds@latest add vishuwa2004/ordered-streaming-tool-execution`
- Raw SKILL.md: https://api.skillmd.com/api/skills/vishuwa2004/ordered-streaming-tool-execution/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools
- Author: vishuwa2004 (https://skillmd.com/u/vishuwa2004)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/vishuwa2004/ordered-streaming-tool-execution

---


# SKILL: Ordered Streaming Tool Execution
**Domain:** tool-orchestration  
**Trigger:** Streaming toolUse input where progress must be shown in the original queue order, even if some tools run concurrently.  
**Source Pattern:** Distilled from reviewed tool execution, streaming, persistence, and output-budget implementations.

## Core Method
Maintain an ordered queue of tracked tools whose statuses include `queued`, `executing`, and `completed`. When a tool becomes eligible, start it only if no incompatible executor is running (non-concurrency-safe tools block the queue) and mark it `executing`. Buffer results and progress messages per tool and never yield them until all previous tools have finished; only then append their buffered messages to the output stream in the original order. This preserves a deterministic user view while still allowing safe tools to run in parallel.

## Key Rules
- Represent each tool use with a tracked descriptor holding queue position, assistant metadata, context modifiers, pending progress, and results.
- Only start tools when can execute tool permits; concurrent-safe ones may run alongside others but non-safe ones must wait for the queue drain.
- Buffer output (results/progress/heuristics) and flush it only after every earlier tool has resolved, so ordering cannot be disrupted by race conditions.

## Example Application
Implement a streaming tool runner for a conversational agent by streaming helper tasks into the executor, letting concurrency-safe tools run in parallel, and deferring emission until all prior toolUse entries have completed so the transcript matches the request order.

## Anti-Patterns (What NOT to do)
- Emit tool results as soon as they arrive in the background, which scrambles the perceived order and confuses the user.
- Ignore is concurrency safe and start mutually exclusive tools simultaneously, risking inconsistent shared resources.
- Fail to keep pending progress separate and merge them out of order, which causes partial updates to interleave incorrectly.