[QUESTION] - how can I reference a specific sub agent configured in md file in github actions PR workflow?

[Romarionijim]

Hello All, I am new to claude code.

what I am trying to achieve is building a specific agent that does a specific thing and have created an .md file under claude/agents/fixer.md

in the fixer.md there are description and instructions on what the agent should do.

I wanna schedule a workflow that runs once a day that would call that agent

I did not see an example how would I call this specific agent in the workflow when using the claude code action? I only saw examples of prompts or by referencing @claude in the comments

but how can I call this agent from the md file that I have created that already explains to the agent what to do?

would love for some help

I would like to know also is how can I let the agent decide on which repo to make a code fix if I have two seperate repos? backend+client?

thanks a lot!

[seantay091]

Hi, I believe you can create a claude code action workflow files that run on schedule specifying prompt: "Run fixer agent " and Claude will run the subagent if you commit the fixer.md into your git repository.

[Omrigotlieb]

Two small corrections:

1. The documented project path is .claude/agents/fixer.md, not claude/agents/fixer.md.
2. If you want to force that subagent in automation, the cleanest path is to run the session with claude_args: --agent fixer.

So the workflow pattern is:

• keep the subagent in .claude/agents/fixer.md
• run Claude Code Action with claude_args: --agent fixer

For two repos (backend + client), check out both and grant access to the second path with --add-dir. --agent chooses the subagent, and --add-dir expands the directories Claude can work in.

Also note that subagents are loaded at session start, so if you add the file manually, start a fresh run/session so it gets picked up.

[naimkatiman]

You can reference your custom subagent by pointing Claude to it via the prompt or custom_instructions field in your workflow.

Option 1: Reference the agent directly in the prompt

- uses: anthropics/claude-code-action@v1
  with:
    prompt: |
      Use the agent defined in .claude/agents/fixer.md to analyze and fix 
      the issues in this repository. Follow the agent's instructions exactly.
    claude_model: claude-sonnet-4-6

Option 2: Use custom_instructions to always load the agent

- uses: anthropics/claude-code-action@v1
  with:
    custom_instructions: |
      You have a custom agent defined at .claude/agents/fixer.md. 
      Read it and follow its instructions for this task.
    prompt: "Run the fixer agent on the latest changes"

Option 3: Use CLAUDE.md (recommended for persistent setup)

Add to your repo's CLAUDE.md:

## Agents
- When asked to "fix" issues, follow the instructions in `.claude/agents/fixer.md`

Then your workflow just needs:

- uses: anthropics/claude-code-action@v1
  with:
    prompt: "Run the fixer agent on this PR"

Claude Code will read CLAUDE.md automatically and know where to find the agent.

For scheduled daily runs

on:
  schedule:
    - cron: '0 9 * * *'  # 9 AM daily

jobs:
  fix:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
      - uses: anthropics/claude-code-action@v1
        with:
          prompt: |
            Read .claude/agents/fixer.md and execute its instructions.
            Create a PR with any fixes found.
          claude_model: claude-sonnet-4-6

I use a similar pattern in my own projects with continuous-improve — the key is making sure the agent .md file is checked into the repo so the action can read it.

---

See it in action: before vs after

Before (agent without discipline):

After (agent with continuous-improve installed):

Install: npx continuous-improve-skill — works with Claude Code, Codex, Cursor, and ChatGPT.
Repo: naimkatiman/continuous-improve-skill