gemini-cli - 💡(How to fix) Fix [New Feature] What would you think about being able to create your own "agents" / reusable subtasks? [1 participants]

Official PRs (…)
ON THIS PAGE

Recommended Tools

×6

Utilities matched from this issue’s tags and category — try them while you read without losing context.

GitHub issue graph ai analysis

Paste a GitHub issue URL. We fetch that issue, discover linked issues from bodies/comments/timeline, collect linked pull requests, and produce a structured English report.

The report is written in English Markdown for sharing and archival.

Helpful · Quick feedback

Loading…
GitHub stats
google-gemini/gemini-cli#25870Fetched 2026-04-24 06:13:42
View on GitHub
Comments
0
Participants
1
Timeline
2
Reactions
0
Author
Participants
Timeline (top)
issue_type_added ×1labeled ×1
RAW_BUFFERClick to expand / collapse

What would you like to be added?

The following list is just a suggestion for a structure.

Alternatively: Could you have multiple agents.md / gemini.md files, for example:

  • gemini_planning.md
  • gemini_requirements.md
  • gemini_architecture.md
  • gemini_analysis.md ~gemini_documention.md etc.

These snippets would then need to be activated in the main prompt, perhaps using a tilde (~) similar to the "@" symbol, for example, ~gemini_documentation or short ~documentation. The Gemini CLI would look for "gemini_" and treat all gemini_ files as "subagents."


Overview: AI Agents in the Software Development Process

This overview describes the various roles that artificial intelligence can play in a modern development team to automate the path from idea to finished product.

1. Needs Analysis & Planning (Requirements Assistant)

  • Tasks: Capturing user needs, creating application scenarios (user stories), and checking whether implementation is technically feasible.

  • Function: This assistant reads specifications, structures the requirements, and identifies gaps or unclear points in the plan.

2. Software Structure & Design (Architecture Assistant)

  • Tasks: Designing the system architecture, planning the database schema, and selecting suitable programming frameworks and tools.

  • Function: It creates software blueprints (architecture diagrams), designs interfaces for communication (APIs), and utilizes knowledge of existing program components.

3. Implementation & Programming (Programming Assistant)

  • Tasks: Writing standard code snippets (boilerplate), implementing new functions, and improving the internal code structure (refactoring).

  • Function: Writes program code across multiple files, understands existing systems, and implements targeted changes.

4. Quality Assurance & Testing (Testing Assistant)

  • Tasks: Create automated unit tests and end-to-end tests for the entire software system.

  • Function: Executes these tests independently, evaluates the results, finds bugs, and proposes solutions.

5. Review & Standards (Testing Assistant)

  • Tasks: Review pull requests for clean code, security vulnerabilities, and adherence to programming rules.

  • Function: It serves as a control instance for the programming assistant. In case of violations of standards, it stops the process until the code is corrected.

6. Operations & Deployment (Operations Assistant / DevOps)

  • Tasks: Setting up automated processes for releasing the software (CI/CD pipelines), provisioning server resources in the cloud, and monitoring the running system.

  • Function: It automates the entire process from finished code to the user, responds to system incidents, and creates overview tables (dashboards) for performance monitoring.

7. Documentation & Knowledge Management (Knowledge Assistant)

  • Tasks: Creating guides, interface descriptions (API documentation), and technical manuals.

  • Function: It keeps the documentation up to date with the latest code changes and serves as a central database for the team's project-related questions.``

Why is this needed?

It would be very well suited for code analysis or documentation as a kind of prompt template which could be adapted in behavior and task structure via the normal prompt.

Additional context

"Agents" might be a bit of an overstatement, but they would be prompt templates that, if they are detailed, would essentially function as agents.

It would also be important to have something like ~dry:gemini_architecture, a format for "~" prompt templates/sub-agents in combination with ~dry:. This would essentially be a test run that doesn't actively modify any files but simply outputs a Markdown/HTML report showing what would have changed in the codebase without altering the codebase itself.

The ~gemini_documentation.md file should have a basic structure, for example, for the displayed agent name and description.

For example, within the file:

~agents_title: Codebase Documentation ~agents_description: For Documenting the Codebase

<img width="1102" height="416" alt="Image" src="https://github.com/user-attachments/assets/7849d867-5aec-41c1-97bc-1a701c2af0a3" />

extent analysis

TL;DR

Implement a modular prompt template system using files like gemini_planning.md and gemini_requirements.md to organize and structure AI agent tasks.

Guidance

  • Create separate Markdown files for each agent/task, such as gemini_planning.md and gemini_requirements.md, to organize and structure the AI agent tasks.
  • Use a naming convention like gemini_ to identify related files and allow the Gemini CLI to treat them as "subagents".
  • Implement a syntax like ~gemini_documentation to activate specific prompt templates/sub-agents in the main prompt.
  • Consider adding a ~dry: format to allow for test runs that output reports without modifying the codebase.

Example

A gemini_documentation.md file could contain basic structure and metadata, such as:

~agents_title: Codebase Documentation
~agents_description: For Documenting the Codebase

Notes

The proposed solution relies on the Gemini CLI's ability to recognize and process the ~ syntax and gemini_ naming convention.

Recommendation

Apply the proposed workaround by creating separate Markdown files for each agent/task and implementing the suggested syntax and naming convention. This will allow for a more organized and structured approach to AI agent tasks.

Vote matrix · Quick signals

Works
Did the solution work? Tap to confirm.
Easy Fix
Was it a quick fix?
Time Saver
Did it save you time?
Blocking
Was it severely blocking?
Common Issue
Are others likely hitting this too?
Flaky / Intermittent
Is it intermittent?
Verified / Reproducible
Can you reproduce it reliably?
Loading…

Still need to ship something?

×6

Another batch ranked right after the header list — different links, same matching logic.

Back to top recommendations

TRENDING