Intelligent GitHub repository automation, with your own agents.

Copilot, Claude Code, Codex or Gemini triage issues, investigate failures and open pull requests on your repo. Safely, with strong guardrails, in GitHub Actions.

Create a workflowCopy installation promptPaste into your favourite coding agentInitialize this repository for GitHub Agentic Workflows using https://raw.githubusercontent.com/github/gh-aw/main/install.md

Prefer a guided form? Try the workflow wizard ↗

You

Create a daily agentic workflow that gives me an issue summary

Added daily-issue-summary.md and opened a pull request.

.github/workflows/daily-issue-summary.md

---
on: daily
safe-outputs:
  create-issue:
---
Summarise yesterday's issues. Flag
anything that needs a reply.

Next morning, 08:00

Issue summary — Oct 2 #241Open

• 9 new issues, 6 closed
• 2 waiting on a maintainer reply
• #482 blocked on design review

Start every day with a better repository, improved while you sleep.

The coding agents you know and love triage issues, investigate failures and open pull requests overnight. Under your rules, in GitHub Actions.

Create a workflowCopy installation promptPaste into your favourite coding agentInitialize this repository for GitHub Agentic Workflows using https://raw.githubusercontent.com/github/gh-aw/main/install.md

Prefer a guided form? Try the workflow wizard ↗

While you sleptocto-org/app

Good morning — 3 updates awaiting your review.

1. 02:10

   02:10CI Doctor opened an issue

   Flaky test: db timeout in integration suite #519Root cause: shared fixture reused across workers
2. 04:30

   04:30Test Improver opened a pull request

   Add tests for parser edge cases #231Coverage +3.1%
3. 07:00

   07:00Repo Status posted the daily report

   Daily status — Oct 2 #23312 issues triaged · 4 PRs merged

Continuous AI for every repository, always running.

A third pillar beside CI and CD: coding agents that keep improving your repository, triaging, documenting, testing and fixing in GitHub Actions, under guardrails you control.

Create a workflowCopy installation promptPaste into your favourite coding agentInitialize this repository for GitHub Agentic Workflows using https://raw.githubusercontent.com/github/gh-aw/main/install.md

Prefer a guided form? Try the workflow wizard ↗

CIBuilds it7b21e0c04d9a19ef3b

CDShips itv2.3.1v2.4.0

Continuous AIImproving your repo, around the clockAround the clock

1. TriageLabelled and asked for repro steps #482now
2. DocsSynced CLI flags with --help #2141h
3. TestsCoverage up 3.1% in the parser #2313h
4. FixesDiagnosed a flaky db timeout #5195h

Workflows that already work

Ready-made workflows, running in real repositories today. Paste one prompt into your coding agent to add it to your repo.

Repo AssistTwice a dayCI DoctorWhen CI failsRepo StatusDailyTest ImproverDaily

Repo Assist

Twice a day

A repository assistant that triages issues, fixes bugs and proposes improvements.

Each run it picks tasks to suit the state of your repository: labelling and investigating issues, fixing bugs, adding tests and engineering improvements. It keeps a monthly activity summary so maintainers can see what it did.

Running in 13 open-source repos, Repo Assist has closed 578 issues, with issues closing a median 8X faster.

Copy setup promptPaste into your favourite coding agentCreate a workflow for GitHub Agentic Workflows using https://raw.githubusercontent.com/github/gh-aw/main/create.md The purpose of the workflow is to act as a repository assistant: twice a day, triage new issues, fix bugs, and open small pull requests with improvements.Read more →

[Repo Assist] Monthly activity, October #300

Open

github-actions commented · this week

• Labelled 14 new issues
• Opened #318 Fix trailing comment handling
• #309 Speed up config loading was merged
• Closed #297 as a duplicate of #284
• Opened #322 Add tests for the config loader

CI Doctor

When CI fails

Investigates failing CI and opens an issue with the likely cause.

It reads the failed job’s logs alongside the commit and pull request that triggered it, then files one issue with the root cause and next steps. Repeat failures update that issue instead of adding noise.

Copy setup promptPaste into your favourite coding agentCreate a workflow for GitHub Agentic Workflows using https://raw.githubusercontent.com/github/gh-aw/main/create.md The purpose of the workflow is to investigate failing CI runs: read the logs, find the likely cause, and open an issue with the diagnosis and next steps.Read more →

Some checks were not successful
1 failing and 1 successful check

CI / testFailing after 2m

CI / lintSuccessful in 12s

github-actions opened test job times out on db setup #519

github-actions commented

Likely cause: DB_HOST is unset since the runner image update in #512. Next step: set the service hostname in ci.yml.

Repo Status

Daily

A daily report of what happened in your repository, posted as an issue.

It gathers pull requests, issues and workflow results, then writes a short status report. Yesterday’s report closes when today’s arrives.

Copy setup promptPaste into your favourite coding agentCreate a workflow for GitHub Agentic Workflows using https://raw.githubusercontent.com/github/gh-aw/main/create.md The purpose of the workflow is a daily report on recent activity in the repository, delivered as an issue.Read more →

Daily status, 2 October #507

Open

github-actions opened this issue

A busy day: two long-running PRs landed.

6PRs merged4issues opened9issues closed

• #309 made config loading 3X faster
• CI is green on main after #521
• #301 needs a maintainer decision

Test Improver

Daily

Opens pull requests that improve test coverage where it matters.

It works out how your project builds and measures coverage, finds high-value gaps, and opens pull requests with the measured improvement. It keeps its own pull requests up to date.

Copy setup promptPaste into your favourite coding agentCreate a workflow for GitHub Agentic Workflows using https://raw.githubusercontent.com/github/gh-aw/main/create.md The purpose of the workflow is to run daily and improve test coverage: find untested code and open pull requests that add focused tests.Read more →

Add tests for the config loader #322

Open

test/config.test.ts+64−0

test/fixtures/app.yml+12−0

Coverage for src/config.ts41%88%

All checks have passed212 tests

Show 12 more workflowsShow fewer

Maintain

• Issue TriageLabels new issues and pull requests.
• AI ModeratorDetects and moderates spam and link spam.
• Daily Documentation UpdaterUpdates docs from recent code changes.
• Link CheckerFinds and fixes broken links in your docs.

Improve code

• Code SimplifierSimplifies recently modified code.
• Duplicate Code DetectorSpots duplication and suggests refactors.
• Daily Perf ImproverBenchmarks and optimises slow code.
• PR FixFixes failing checks on a pull request.

Report and plan

• Weekly Issue ActivityWeekly issue trends and recommendations.
• Daily Team StatusAn upbeat summary of team activity.
• Daily PlanKeeps planning issues up to date.
• Grumpy ReviewerA thorough, opinionated code review on demand.

Workflow patterns in the docs →

Watch it run

Create and run your first workflow in under two minutes.

From the terminalTerminal1:16From github.comgithub.com1:41

Install the extension, add a sample workflow and trigger the first run. Follow the quick start →

Ask Copilot in your repository's Agents tab to write the workflow for you. Create a workflow →

Safe by default

Agents read by default. Every write is checked before it lands.

An agent becomes dangerous when it has private data, untrusted content and outbound access all at once. Security researchers call it the lethal trifecta. gh-aw defends in depth through six security layers that prevent agents from leaking private data.

Private dataSecrets and credentialsIsolated sandboxUntrusted contentIssues and PR commentsIntegrity filteringOutbound accessWeb pages, APIs, writesAgent firewallIsolated sandboxIntegrity filteringAgent firewallthe lethal trifectaAgent

Diagram of the lethal trifecta: three capabilities that are dangerous when an agent holds all of them at once. Private data means secrets and credentials. Untrusted content means issues and pull request comments. Outbound access means web pages, APIs and writes. Each one would reach the agent at the centre, but gh-aw cuts every leg: the isolated sandbox keeps private data away from the agent, integrity filtering screens untrusted content, and the agent firewall restricts outbound access.

Step through the layers.

The agent

1. Compile-time validation

   Compile-time validation

   Schema validation, expression allowlisting and pinned actions are checked before a workflow can ever run.

   Learn more
2. Sandboxed agent

   Sandboxed agent

   The agent runs read-only by default, with optional Cloud Hypervisor microVMs for stronger isolation.

   Learn more
3. Credential isolation

   Credential isolation

   An API proxy holds the tokens, so the agent never sees them and cannot leak them.

   Learn more
4. Integrity filtering

   Integrity filtering

   Untrusted GitHub content is filtered before the agent sees it.

   Learn more
5. Threat detection

   Threat detection

   A separate job scans proposed outputs and blocks suspicious ones.

   Learn more
6. Safe outputs

   Safe outputs

   Only the writes you declared are applied, by a separate job with its own permissions.

   Learn more

Your repository

Try it safely: staged: true previews every output without writing anything.

Read the security architecture →

How it works

Prompt your coding agent to create a workflow. GitHub compiles it into an Actions workflow and runs it.

1. GitHub Copilot

   /agentic-workflows

   Create a daily status workflow.

   /agentic-workflow optimize it

   Prompt

   Ask your coding agent for a workflow in plain language, then refine it with /agentic-workflow optimize it.

2. Terminal

   $ gh aw compile

   triage.lock.yml

   Pinned, validated GitHub Actions YAML

   Compile

   gh aw compile turns it into a hardened, pinned Actions workflow.

3. Actions

   triage · agent48s

   Applied labeltype: bug

   Run

   Your chosen agent runs in GitHub Actions; approved outputs are applied.

Runs the coding agent you choose

• GitHub CopilotGitHub
• Claude CodeAnthropic
• CodexOpenAI
• Gemini CLIGoogle
• PiMulti-provider

Bring your own engine →Import OpenCode, Cursor, Kiro, Aider, Crush or your own definition.

Cost under control

Set a budget per run and see where every credit goes.

issue-triageCopilot

24h7 days30 days

AI credits

256.9 AIC≈ $2.57

Runs

61 stopped at cap

Tokens

2.0M334K per run

Avg turns

17.3max 44

AI credits per run

0255075

max-ai-credits 100

1. Run 18409322715, Sep 25: 28.1 AIC, 11 turns, 2.6m28.1#1840932271528.1 AIC · 11 turns · 2.6mSep 25
2. Run 18409951280, Sep 26: 33.2 AIC, 13 turns, 3.1m33.2#1840995128033.2 AIC · 13 turns · 3.1mSep 26
3. Run 18410763311, Sep 28: 24.9 AIC, 9 turns, 2.2m24.9#1841076331124.9 AIC · 9 turns · 2.2mSep 28
4. Run 18411520968, Sep 29: 100.6 AIC, 44 turns, 9.8m, stopped at cap100.6#18411520968100.6 AIC · 44 turns · 9.8mSep 29
5. Run 18411877042, Oct 1: 38.7 AIC, 15 turns, 3.4m38.7#1841187704238.7 AIC · 15 turns · 3.4mOct 1
6. Run 18412093317, Oct 1: 31.4 AIC, 12 turns, 2.9m31.4#1841209331731.4 AIC · 12 turns · 2.9mOct 1

Today

70.1 / 5,000 AIC

max-daily-ai-credits

Most expensive runs

1. #18411520968Stopped at cap100.6
2. #1841187704238.7
3. #1840995128033.2

• Cap every run

  max-ai-credits and max-turns stop a run that runs away. On by default: 1,000 AI Credits per run.

  Set a budget →

• Cap every day

  max-daily-ai-credits limits a workflow across 24 hours, 5,000 AIC by default. Over budget, the agent is skipped.

  Daily limits →

• Find the expensive runs

  gh aw logs lists tokens, AIC and turns per run. gh aw audit breaks one down. 1 AIC is $0.01.

  Monitor cost →

Send traces to any OTLP backend with OpenTelemetry, and compare cost with outcomes.

Ready for organizations

Run on your own infrastructure, across every repository.

.github/workflows/triage.md

org/acme16 repositories

• api
• web
• docs
• infra
• mobile
• billing
• auth
• search
• cli
• payments
• analytics
• sdk-js
• ios
• android
• platform
• support

Self-hosted runners

• arc-linux-01
• arc-linux-02
• arc-linux-03

• Your own runners

  Run on Linux self-hosted runners, including Actions Runner Controller with Docker-in-Docker.

  Self-hosted runners →

• Every repository

  Roll workflows out across an organization or enterprise and manage them centrally.

  Using at scale →

Keep going

• Hands-on workshop (opens in a new tab)Guided exercises in the terminal, the browser or Copilot.
• Peli's Agent FactoryA tour of the agents the team runs on its own repositories.
• BlogRelease notes, patterns and stories from the team.
• DocumentationConcepts, guides and the full reference.
• Slide deck (PDF, 11 MB)The technical preview deck: concepts, security model and examples.
• Agentic workflows on GitHub (opens in a new tab)Source, issues and discussions.

Create your first workflow

Create a workflowCopy installation promptPaste into your favourite coding agentInitialize this repository for GitHub Agentic Workflows using https://raw.githubusercontent.com/github/gh-aw/main/install.md

Prefer a guided form? Try the workflow wizard ↗