✨
Échantillon de contenu généré par l'IA de BlogFlow Secteur : Logiciel / développement · Conçu pour générer du trafic qualifié sur Google
Obtenir ces articles sur mon site →
Logiciel / développement • • 572 mots

GitHub Launches Agentic Workflows with Enhanced Security and Cost Controls

GitHub introduces Agentic Workflows with isolation and budget controls, offering developers new tools for automation and security.

A New Era in GitHub Automation

GitHub has recently made its Agentic Workflows generally available, marking a significant step forward in how developers automate tasks within their repositories. This new feature allows coding agents to perform complex tasks such as issue triage and continuous integration (CI) failure analysis directly within GitHub Actions. These agents operate with read-only permissions by default, ensuring that any changes they propose are handled by separate jobs that have their own write tokens. This separation helps maintain security and control over repository modifications.

Enhanced Security Through Isolation

One of the key aspects of GitHub's Agentic Workflows is the isolation of agents within a sandboxed container. This container is protected by the Agent Workflow Firewall, which adds an extra layer of security. Additionally, a threat-detection job scans any proposed changes before they are applied, reducing the risk of unintended or malicious modifications. However, it's important to note that these protections do not automatically extend to custom jobs or direct write permissions, meaning the safety of a workflow depends heavily on its configuration.

Customization and Configuration

Developers can define their workflows using natural-language Markdown, with YAML frontmatter for configuration. This approach makes it easier to create and manage workflows without needing to write complex code. The project compiles these definitions into standard Actions workflows, reusing existing runner groups and policy constraints. Teams can also use GitHub Copilot or bring in other coding agents to assist with development tasks.

Cost Management and Budget Controls

GitHub has implemented default spending caps to help manage costs associated with AI usage. Each AI Credit (AIC) is priced at $0.01, with a default cap of 1,000 AIC ($10) per run. For enterprise users, additional controls include a daily limit of 5,000 AIC, a cap of 400 AIC per threat-detection run, and a 20-minute timeout on the agent step. These settings can be customized by organizations to fit their specific needs.

Real-World Impact and User Feedback

Early adopters of the Agentic Workflows have reported significant time savings. According to a statement from Accenture, one of their leads saved an estimated 29 hours a week on development tasks alone. However, some engineers have taken a more cautious approach, adjusting their workflows to ensure they remain within acceptable limits. For example, one developer limited his runs to 100 AIC, 15 turns, and 10 minutes to avoid unnecessary costs and ensure efficiency.

Comparison with Competitors

While GitHub's approach focuses on isolation and budget controls, other platforms like GitLab take a different route with its Duo Agent Platform. GitLab's flows run on runners under a composite identity that pairs a service account with a human user, ensuring that no agent can start another flow without human intervention. This approach prevents agents from chaining work across flows without oversight.

Choosing the Right Tool for the Job

GitHub's FAQ emphasizes that conventional Actions are still the best choice for reproducible builds and tests, while agent workflows are better suited for tasks that require interpretation or generation. This distinction helps developers choose the right tool for their specific needs, whether they're looking for automation, security, or cost control.

Availability and Licensing

The Agentic Workflows project is available on GitHub Free, Team, and Enterprise on github.com. However, the default Copilot CLI engine requires a Copilot license, which may be an additional consideration for some users.

{{appel_action}}

{{lien:1}}


Source : InfoQ

Mots-clés : #GitHub #Agentic Workflows #CI/CD #AI Automation #Software Development
Auteur / Émetteur

Découvrez comment BlogFlow peut automatiser la rédaction de votre blog d'entreprise.

Essai 7 jours gratuit