# Long Horizon Composite Execution

> Use this skill when the user wants long chains of simple actions that only become hard because they accumulate over time. Trigger it for requests like “compose several easy steps into one hard task,” “make it remember earlier work,” “test long multi-step desktop workflows,” or “build a task that only looks easy when broken into pieces.” This skill is for GUI tasks where subtask sequencing, memory, and dependency management drive difficulty.

- Skill: `dingxingdi/long-horizon-composite-execution-2` (Agent Skill, multi-file: 5 files)
- Install (CLI): `npx skillmds@latest add dingxingdi/long-horizon-composite-execution-2`
- Raw SKILL.md: https://api.skillmd.com/api/skills/dingxingdi/long-horizon-composite-execution-2/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Productivity
- Author: dingxingdi (https://skillmd.com/u/dingxingdi)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/dingxingdi/long-horizon-composite-execution-2

---


# Skill: long-horizon-composite-execution
## 1. Capability Definition & Real Case
* **Professional Definition**: The capability to maintain coherent progress across a long sequence of dependent GUI subtasks and to complete the resulting composite objective without dropping state, forgetting prior outputs, or prematurely stopping after partial completion. The agent must preserve memory over extended horizons while integrating earlier artifacts into later operations.
* **Dimension Hierarchy**: Goal-Directed GUI Workflow Execution->Verified Task Completion->long-horizon-composite-execution
### Real Case
**[Case 1]**
* **Initial Environment**: A desktop environment includes a browser for fact gathering and a spreadsheet application for structured entry and visualization. The agent must progressively collect statistics across several decades, maintain the accumulated evidence, enter it into a table, and finish with a correctly labeled chart.
* **Real Question**: Collect recidivism rate statistics in the United States for the 1980s, 1990s, 2000s, and 2010s using online sources, enter this data into a LibreOffice Calc spreadsheet with columns labeled 'Period' and 'Recidivism Rate', and create a clearly-labeled line chart titled 'US Recidivism Rates by Decade' with the X axis labeled 'Decade' and the Y axis labeled 'Recidivism Rate (%)' to visually compare changes in recidivism rates across these decades.
* **Real Trajectory**: Gather one decade at a time from credible sources, maintain a running memory of the collected values, enter the data into the spreadsheet with the required schema, and create the requested visualization only after the full table is complete.
* **Real Answer**: A spreadsheet containing the decade-level recidivism data and a correctly labeled line chart is produced.
* **Why this demonstrates the capability**: Each individual step is manageable, but the task becomes hard because later operations depend on accurate outputs from earlier ones. The agent must sustain memory, avoid partial completion, and keep the evolving workspace coherent over many steps. That is the hallmark of long-horizon composite execution.

## Pipeline Execution Instructions
To synthesize data for this capability, you must strictly follow a 3-phase pipeline. **Do not hallucinate steps.** Read the corresponding reference file for each phase sequentially:

1. **Phase 1: Environment Exploration**
   Read the exploration guidelines to discover raw knowledge seeds:
   `references/EXPLORATION.md`

2. **Phase 2: Trajectory Selection**
   Once Phase 1 is complete, read the selection criteria to evaluate the trajectory:
   `references/SELECTION.md`

3. **Phase 3: Data Synthesis**
   Once a trajectory passes Phase 2, read the synthesis instructions to generate the final data:
   `references/SYNTHESIS.md`