Feature request: Parametrize agents to allow for extra workflow steps / override steps

[Razkaroth]

Hey! I have been using this and found it really useful. However, one thing I found myself doing frequently is taking the results of octto and have the agents process it in other ways.

An example of this would be taking the final plan and having the agent process it into beads (an ai issue tracker). It would be nice to be able to have some way of extending/modifying agent prompts.

I think the config file could have something like this:

{
  "agents": {
      "octto": { 
         "model": "github-copilot/claude-sonnet-4.5"
         "workflow-steps" :  { 
                "5": "An aditional step for this agent after the default"
            }
         },
   } 
}

Thanks for your time reading this.

[vtemian]

Implemented in v0.3.0 via PR #4.

Features:

• Two-layer config: global (~/.config/opencode/octto.json) + project (.octto/fragments.json)
• Fragments merged and prepended to agent prompts as <user-instructions> XML blocks
• Typo detection with Levenshtein distance suggests similar agent names

Usage:

{
  "fragments": {
    "octto": ["Always suggest 3 implementation approaches"],
    "probe": ["Include emoji in every question"]
  }
}

See README for full documentation.

[Razkaroth]

Awesome!

Thank you for taking the time!