github / github/copilot-cli

Support bundled reference resources for plugin-installed custom agents in GitHub Copilot CLI

Abierto
#3,131 0 comentarios 1 reacción 0 asignados Ver en GitHub
area:agents area:plugins
Lenguaje dominante
Shell
Estrellas
11.2k
Forks
1.9k
Merge medio
14 h 16 min
PR fusionados (30 d)
6

Descripción

### Describe the feature or problem you'd like to solve

GitHub Copilot CLI plugins can include both custom agents and skills, but their resource-loading behavior appears inconsistent. Skills are documented as directory-based bundles. The official **Adding agent skills for GitHub Copilot CLI** documentation says that agent skills are “folders of instructions, scripts, and resources,” and that when a skill is invoked, Copilot “automatically discovers all of the files in the skill's directory and makes them available alongside the skill's instructions.”
Docs:
- https://docs.github.com/en/copilot/how-tos/copilot-cli/customize-copilot/add-skills
- https://docs.github.com/en/copilot/reference/copilot-cli-reference/cli-plugin-reference
- https://docs.github.com/en/copilot/how-tos/copilot-cli/customize-copilot/plugins-creating
Custom agents, however, are documented primarily as `.agent.md` profile files with YAML frontmatter and prompt content. Plugin docs show agents under an `agents/` directory, but do not describe a comparable resource-discovery model for files such as `agents/references/*.md`.
Docs:
- https://docs.github.com/en/copilot/how-tos/use-copilot-agents/cloud-agent/create-custom-agents
- https://docs.github.com/en/copilot/reference/custom-agents-configuration
- https://docs.github.com/en/copilot/how-tos/copilot-cli/customize-copilot/plugins-creating
In practice, this means a plugin-installed skill can reliably reference bundled files with paths such as:
```text
references\template.md
scripts\helper.ps1
```
because those files live under the skill directory:
```text
skills/my-skill/
├── SKILL.md
├── references/template.md
└── scripts/helper.ps1
```
But a plugin-installed agent cannot reliably reference files under: `agents/references/template.md` using: `references\template.md` relative path from its `.agent.md` prompt. This creates friction for plugin authors who want to keep agent prompts concise and maintain reusable supporting materials, templates, checklists, examples, or routing guidance outside the main `.agent.md` file.

### Proposed solution

Add support for bundled resources for plugin-installed custom agents, matching the skill resource model as closely as possible.

For example, allow plugin agents to resolve relative paths from the containing agent directory:

```text
my-plugin/
├── plugin.json
└── agents/
├── my-agent.agent.md
└── references/
└── checklist.md
```

Then `my-agent.agent.md` could safely reference:

```text
references/checklist.md
```

Alternatively, document and support an explicit agent resource directory convention, such as:

```text
agents/my-agent/
├── AGENT.md
├── references/checklist.md
└── scripts/helper.ps1
```

If implementing bundled agent resources is not currently planned, please update the plugin and custom agent documentation to explicitly state that:

1. Plugin-installed agents are single-file runtime artifacts.
2. `agents/references/*` is not automatically discovered or made available.
3. Authors should inline required agent instructions or move reusable resources into skills.
4. Skills are the recommended mechanism for bundled scripts, templates, examples, and reference files.

### Example prompts or workflows

_No response_

### Additional context

_No response_

Guía de contribución

Abrir la guía de contribución

Línea de trabajo

Empieza leyendo la documentación sobre plugins y agentes personalizados enlazada en el issue; después, inspecciona plugin.json y la estructura de agents/ para determinar cómo se resuelven actualmente los recursos. Se considera terminado cuando las referencias incluidas se resuelvan de forma relativa al agente y este comportamiento esté verificado, o cuando la limitación y el flujo de trabajo recomendado con skills estén documentados explícitamente.

Escrito por el modelo de indexación a partir del texto del issue.

Evaluación

Stack tecnológico
github
Área
cli, tooling
Tipo de issue
Nueva funcionalidad
Dificultad
4/5
Tiempo estimado
3-5 días
Estado de actividad
Tranquilo
Claridad
Bastante claro
Aptitud para principiantes
45/100

Recibe los nuevos issues en tu correo

Un resumen breve de issues de GitHub para principiantes.