Devin Action

A reusable GitHub action which calls out to Devin.ai, creating a new Devin session with a given prompt or playbook. Designed to be used directly, or in slash commands. When invoked via slash commands, Devin can:

post a response back as a comment
update the current PR
open an new PR if needed

Features

✅ Creates Devin.ai sessions with custom prompts
✅ Supports triggering via slash commands in PR and issue comments
✅ Automatically gathers context from GitHub issues and comments
✅ Supports playbook macro integration
✅ Posts status updates and session links as comments
✅ Flexible input handling - use any combination of inputs

Inputs

Name	Description	Required	Default
`comment-id`	Comment ID for context and reply chaining	false
`issue-number`	Issue number for context gathering	false
`playbook-macro`	Playbook macro for structured workflows - should start with '!' (e.g., !my_playbook)	false
`prompt-text`	Additional custom prompt text. Values over ~128 KiB exceed Linux's per-environment-variable limit; use `prompt-file` for larger prompts.	false
`prompt-file`	Path to a file containing the prompt text. Use this for prompts over ~128 KiB, which exceed Linux's per-environment-variable limit for `prompt-text`.	false
`devin-token`	Devin API Token (required for authentication)	true
`github-token`	GitHub Token (required for posting comments and accessing repo context)	false
`start-message`	Custom message for the start comment	false	🤖 Starting Devin AI session...
`tags`	Additional tags to apply to the Devin session (supports CSV or line-delimited format). Automatic tags are always added: `gh-actions-trigger` and `playbook-{macro-name}` if playbook-macro is provided. Cannot be used with `reuse-session`.	false
`reuse-session`	Existing Devin v3 session ID or URL to inject a message into. Accepts either a session ID or a full URL (e.g., `https://app.devin.ai/sessions/abc123`). When provided, sends a message to an existing session instead of creating a new one. Requires `org-id` and is mutually exclusive with `tags`.	false
`wait-for-stopped-status`	If `true`, polls until `status_detail` is any non-`working` state.	false	`false`
`wait-minutes-max`	Maximum minutes to poll before timing out (only used when `wait-for-stopped-status` is enabled)	false	`20`
`api-version`	Devin API version. Only `v3` is supported; any other value fails fast.	false	`v3`
`advanced-mode`	V3 API advanced mode. Option: `analyze`. Requires `org-id`. See Advanced Mode below.	false
`session-links`	Session URLs or IDs to analyze (CSV or line-delimited). Required for `analyze` mode. When provided without `advanced-mode`, defaults to `analyze`.	false
`org-id`	Devin organization ID. Required for v3 session creation, session reuse, and polling.	false
`max-acu-limit`	Maximum ACU limit for the session.	false
`playbook-id`	Playbook ID to use for the session.	false
`child-playbook-id`	Playbook ID for child sessions.	false
`bypass-approval`	If `true`, bypass approval for session creation. Requires `UseDevinExpert` permission.	false	`false`
`structured-output-schema`	JSON Schema describing the structured output Devin should produce. Accepts YAML or JSON (YAML 1.2 is a superset of JSON, so plain JSON also works). Defaults `structured_output_required` to `true`. Ignored when using `reuse-session`. See Structured Output below.	false
`structured-output-required`	Flag passed as `structured_output_required` on session creation. Defaults to `true` when `structured-output-schema` is provided. Set to `false` to make structured output available but optional. Ignored when using `reuse-session`.	false

Session Tagging

This action automatically adds tags to Devin sessions for better monitoring and searching:

Automatic Tags

gh-actions-trigger - Always added to identify sessions triggered from GitHub Actions
playbook-{macro-name} - Added when playbook-macro is provided (e.g., playbook-issue-ask for !issue_ask)

Additional Tags

You can provide additional tags using the tags input parameter, which supports both formats:

CSV format:

tags: 'priority-high,bug-fix,frontend'

Line-delimited format:

tags: |
  priority-high
  bug-fix
  frontend

Mixed format (CSV per line):

tags: |
  priority-high,urgent
  bug-fix,frontend
  team-alpha

Usage

Basic Example

- name: Create Devin Session
  uses: airbytehq/devin-action@v1
  with:
    prompt-text: "Please review this code and suggest improvements"
    devin-token: ${{ secrets.DEVIN_TOKEN }}
    org-id: ${{ vars.DEVIN_ORG_ID }}
    github-token: ${{ secrets.GITHUB_TOKEN }}
    tags: 'code-review,improvement'

Slash Command Example

This action is designed to work with slash commands in issue and PR comments. The action will automatically gather context from the comment and/or issue.

- name: Run Devin from Comment
  uses: airbytehq/devin-action@v1
  with:
    comment-id: ${{ github.event.comment.id }}
    issue-number: ${{ github.event.issue.number }}
    devin-token: ${{ secrets.DEVIN_TOKEN }}
    org-id: ${{ vars.DEVIN_ORG_ID }}
    github-token: ${{ secrets.GITHUB_TOKEN }}

With Playbook Macro Integration

- name: Run Devin with Playbook Macro
  uses: airbytehq/devin-action@v1
  with:
    playbook-macro: "!fix-ci-failures"
    issue-number: ${{ github.event.issue.number }}
    prompt-text: "Additional context about the specific failure"
    devin-token: ${{ secrets.DEVIN_TOKEN }}
    org-id: ${{ vars.DEVIN_ORG_ID }}
    github-token: ${{ secrets.GITHUB_TOKEN }}
    tags: |
      ci-failure
      urgent

Tags applied: gh-actions-trigger, playbook-fix-ci-failures, ci-failure, urgent

Injecting a Message into an Existing Session

Instead of creating a new session, you can inject a message into an existing Devin session using the reuse-session input. This is useful for long-running workflows that need to send multiple messages to the same session.

You can provide either a session ID or a full session URL:

# Using a session ID
- name: Send Message to Existing Devin Session
  uses: airbytehq/devin-action@v1
  with:
    reuse-session: 'c002a79b24b74f5b918ebc7dc6c5205b'
    prompt-text: |
      New task triggered at ${{ github.event.repository.updated_at }}
      Please process the following scope: all certified connectors
    devin-token: ${{ secrets.DEVIN_AI_API_KEY }}
    org-id: ${{ vars.DEVIN_ORG_ID }}

# Using a session URL (session ID is automatically extracted)
- name: Send Message to Existing Devin Session
  uses: airbytehq/devin-action@v1
  with:
    reuse-session: 'https://app.devin.ai/sessions/c002a79b24b74f5b918ebc7dc6c5205b'
    prompt-text: |
      New task triggered at ${{ github.event.repository.updated_at }}
      Please process the following scope: all certified connectors
    devin-token: ${{ secrets.DEVIN_AI_API_KEY }}
    org-id: ${{ vars.DEVIN_ORG_ID }}

Note: The reuse-session input is mutually exclusive with tags. When injecting a message into an existing session, tags cannot be specified since they are only applicable when creating new sessions.

Waiting for Session Completion

Use wait-for-stopped-status to poll the Devin session until it reaches any non-working state. This is useful for workflows that need the session's output before proceeding (e.g., posting a summary, gating subsequent steps).

Wait during session creation (single step):

- uses: airbytehq/devin-action@v1
  id: devin
  with:
    devin-token: ${{ secrets.DEVIN_AI_API_KEY }}
    org-id: ${{ vars.DEVIN_ORG_ID }}
    prompt-text: "Analyze commits and write a summary..."
    wait-for-stopped-status: true

Wait on an existing session (multi-step):

- uses: airbytehq/devin-action@v1
  id: create
  with:
    devin-token: ${{ secrets.DEVIN_AI_API_KEY }}
    org-id: ${{ vars.DEVIN_ORG_ID }}
    prompt-text: "Do something..."

# ... other steps ...

- uses: airbytehq/devin-action@v1
  id: wait
  with:
    devin-token: ${{ secrets.DEVIN_AI_API_KEY }}
    org-id: ${{ vars.DEVIN_ORG_ID }}
    reuse-session: ${{ steps.create.outputs.session-id }}
    wait-for-stopped-status: true

When polling completes, the following additional outputs are available:

Output	Description
`status`	The terminal `status_detail` value when polling completes
`summary`	The last message from the session
`structured-output`	The session `structured_output` as compact JSON, when present

The action waits for any status_detail value other than working. See the Devin API docs for the full list of possible status values.

Advanced Mode (v3 API)

The action supports the Devin v3 API's advanced mode, which enables specialized session behaviors for automation workflows.

API Version Selection

The action only supports the Devin v3 API. The api-version input defaults to v3; setting it to anything else fails before any Devin API call is made. All create, message, and polling calls require org-id.

Prerequisites

The v3 API requires:

A v3-compatible API token — must be a service user credential (not a PAT). Requires ManageOrgSessions permission. Advanced mode additionally requires UseDevinExpert.
Organization ID via the org-id input
Team or Enterprise plan — the v3 API and Advanced Mode are available on Team and Enterprise plans.

Analyze Sessions Example

This is particularly useful for automated triage workflows where one Devin session needs to inspect the conversation history of another session:

- name: Analyze Devin Session
  uses: airbytehq/devin-action@v1
  with:
    prompt-text: "Triage the linked session and identify the root cause of the reported issue."
    devin-token: ${{ secrets.DEVIN_TOKEN }}
    github-token: ${{ secrets.GITHUB_TOKEN }}
    org-id: ${{ secrets.DEVIN_ORG_ID }}
    advanced-mode: analyze
    session-links: |
      https://app.devin.ai/sessions/abc123
      https://app.devin.ai/sessions/def456
    tags: 'session-triage,automated'

Auto-detect Analyze Mode

When session-links is provided without advanced-mode, the action automatically defaults to analyze mode:

- name: Analyze Session (auto-detect)
  uses: airbytehq/devin-action@v1
  with:
    prompt-text: "Review this session and summarize what happened."
    devin-token: ${{ secrets.DEVIN_TOKEN }}
    org-id: ${{ secrets.DEVIN_ORG_ID }}
    session-links: 'https://app.devin.ai/sessions/abc123'

ACU Budget Control

Use max-acu-limit to cap the compute budget for v3 sessions:

- name: Budget-limited Session
  uses: airbytehq/devin-action@v1
  with:
    prompt-text: "Quick analysis of this session."
    devin-token: ${{ secrets.DEVIN_TOKEN }}
    org-id: ${{ secrets.DEVIN_ORG_ID }}
    advanced-mode: analyze
    session-links: 'https://app.devin.ai/sessions/abc123'
    max-acu-limit: 5

Structured Output

Use structured-output-schema to request that Devin produce structured output matching a JSON Schema. The schema is passed through as the structured_output_schema field on the v3 session creation request; it is ignored when using reuse-session.

When structured-output-schema is provided, the action defaults structured_output_required to true. This means Devin must call provide_structured_output with is_final=true before its turn ends. Set structured-output-required: "false" to make the tool available but optional; in that mode, it is not guaranteed to be called in a given turn.

The input accepts either YAML or JSON. YAML 1.2 is a strict superset of JSON, so existing JSON schemas are still valid. The input is validated up front: anything that doesn't parse to a top-level object fails the action before calling the API.

Once the session is running, the schema tells Devin what shape its "notepad" should take. If wait-for-stopped-status is also true, the action emits the session's final structured_output as the structured-output action output when polling completes. See the Devin structured output docs for more details.

Inline Schema (YAML)

YAML is usually easier to read inside a workflow since you can skip the quotes and braces:

- name: Run Devin with Structured Output
  id: devin
  uses: airbytehq/devin-action@v1
  with:
    devin-token: ${{ secrets.DEVIN_AI_API_KEY }}
    org-id: ${{ vars.DEVIN_ORG_ID }}
    prompt-text: |
      Review this PR and keep the structured output up to date with any issues
      you find, suggestions you have, and your current approval status.
    structured-output-schema: |
      type: object
      required: [approved]
      properties:
        issues:
          type: array
          items:
            type: object
            properties:
              file: { type: string }
              line: { type: integer }
              type: { type: string }
              description: { type: string }
        suggestions:
          type: array
          items: { type: string }
        approved:
          type: boolean
    wait-for-stopped-status: true

The structured output can then be consumed by later workflow steps:

- name: Use structured output
  run: |
    echo '${{ steps.devin.outputs.structured-output }}' | jq .

Inline Schema (JSON)

Plain JSON also works unchanged:

- name: Run Devin with Structured Output
  uses: airbytehq/devin-action@v1
  with:
    devin-token: ${{ secrets.DEVIN_AI_API_KEY }}
    prompt-text: "Review this PR and keep the structured output up to date."
    org-id: ${{ vars.DEVIN_ORG_ID }}
    structured-output-schema: |
      {
        "type": "object",
        "properties": {
          "approved": { "type": "boolean" },
          "suggestions": { "type": "array", "items": { "type": "string" } }
        },
        "required": ["approved"]
      }

Schema From a File

For longer schemas, store them in a .yaml (or .json) file and pass the contents through a prior step output:

- id: load-schema
  shell: bash
  run: |
    {
      echo "schema<<EOF"
      cat .github/schemas/pr-review.yaml
      echo "EOF"
    } >> "$GITHUB_OUTPUT"

- name: Run Devin with Structured Output
  uses: airbytehq/devin-action@v1
  with:
    devin-token: ${{ secrets.DEVIN_AI_API_KEY }}
    org-id: ${{ vars.DEVIN_ORG_ID }}
    prompt-file: .github/prompts/pr-review.md
    structured-output-schema: ${{ steps.load-schema.outputs.schema }}

Runner Requirements

Parsing tries the following in order, so any one of these on the runner is enough:

yq (mikefarah's — preinstalled on GitHub-hosted ubuntu-latest).
python3 with PyYAML (also preinstalled on GitHub-hosted ubuntu-latest).
jq (JSON-only fallback — jq is already a hard dependency of this action).

On minimal self-hosted runners without yq or PyYAML, YAML input will be rejected and only JSON will parse.

Context Gathering

The action intelligently builds prompts by combining available context:

Comment Context: If comment-id is provided, includes the comment body and author
Issue Context: If issue-number is provided, includes issue title, description, and author
Playbook Macro Reference: If playbook-macro is provided, instructs Devin to use the specified playbook macro
Custom Prompt: If prompt-text is provided, includes additional custom instructions

All provided inputs are concatenated to create a comprehensive prompt for Devin.

Available Slash Commands

When using the slash command dispatch workflow here in this repo, the following commands are available:

/devin [prompt] - Creates a general Devin session with optional custom prompt
/ai-fix - Creates a Devin session focused on analyzing and fixing issues
/ai-ask - Creates a Devin session focused on providing help and guidance

Example Executions

For execution examples, check the pinned issues here in this repo.

Example Workflow [Slash Commands]

.github/workflows/ai-help-command.yml

name: AI Help Command

on:
  repository_dispatch:
    types: [ai-help-command]
  workflow_dispatch:
    inputs:
      issue-number:
        description: 'Issue number for context'
        required: true
        type: string
      comment-id:
        description: 'Comment ID that triggered the command (optional)'
        required: false
        type: string

permissions:
  contents: read
  issues: write
  pull-requests: read

jobs:
  ai-repro:
    runs-on: ubuntu-latest
    steps:
      - name: Run AI Help
        uses: airbytehq/devin-action@v1
        with:
          comment-id: ${{ github.event.client_payload.slash_command.args.named.comment-id || inputs.comment-id }}
          issue-number: ${{ github.event.client_payload.slash_command.args.named.issue || inputs.issue-number }}
          playbook-macro: '!issue_help'
          devin-token: ${{ secrets.DEVIN_AI_API_KEY }}
          org-id: ${{ vars.DEVIN_ORG_ID }}
          github-token: ${{ secrets.MY_COMMENTS_PAT }}
          start-message: '🤖 **AI Help session starting...**'

License

This project is licensed under the terms of the MIT License.

Name		Name	Last commit message	Last commit date
.github/workflows		.github/workflows
CONTRIBUTING.md		CONTRIBUTING.md
LICENSE		LICENSE
README.md		README.md
action.yml		action.yml

Navigation Menu

Repository navigation

About

Repository files navigation

Devin Action

Features

Inputs

Session Tagging

Automatic Tags

Additional Tags

Usage

Basic Example

Slash Command Example

With Playbook Macro Integration

Injecting a Message into an Existing Session

Waiting for Session Completion

Advanced Mode (v3 API)

API Version Selection

Prerequisites

Analyze Sessions Example

Auto-detect Analyze Mode

ACU Budget Control

Structured Output

Inline Schema (YAML)

Inline Schema (JSON)

Schema From a File

Runner Requirements

Context Gathering

Available Slash Commands

Example Executions

Example Workflow [Slash Commands]

License

About

Releases

Packages

Used by

Contributors

Latest commit

History

Folders and files

Repository files navigation

Devin Action

Features

Inputs

Session Tagging

Automatic Tags

Additional Tags

Usage

Basic Example

Slash Command Example

With Playbook Macro Integration

Injecting a Message into an Existing Session

Waiting for Session Completion

Advanced Mode (v3 API)

API Version Selection

Prerequisites

Analyze Sessions Example

Auto-detect Analyze Mode

ACU Budget Control

Structured Output

Inline Schema (YAML)

Inline Schema (JSON)

Schema From a File

Runner Requirements

Context Gathering

Available Slash Commands

Example Executions

Example Workflow [Slash Commands]

License

About

Resources

Contributing

Security policy

Stars

Watchers

Forks

Releases

Packages

Used by

Contributors