SShortSingh.
Back to feed

Three Ways to Safely Run CI Workflows on Forked Pull Requests in GitHub Actions

0
·11 views

GitHub repository maintainers face a security challenge when running secret-dependent workflows on pull requests submitted by external, untrusted contributors. By default, GitHub restricts forked PRs triggered via the pull_request event to a read-only GITHUB_TOKEN with no access to repository secrets, preventing exposure of sensitive data. GitHub introduced the pull_request_target event in 2021 to elevate token permissions, but using it to execute untrusted forked code directly is considered dangerous without additional safeguards. Security experts, including a GitHub researcher, warn that any automated processing of external fork PRs should be treated as untrusted input due to risks like code injection and race condition exploits. The article outlines three solutions of varying security levels for maintainers needing to run tests or other secret-dependent workflows against community-contributed code changes.

Read the full story at DEV Community

This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)

Log in to join the discussion and vote.

Log in

Related stories

0
ProgrammingDEV Community ·

PATH SSC: Developer Builds Frontend-Only Prototype of Government Exam Prep App

A developer has built PATH SSC, a government exam preparation companion aimed at Indian aspirants preparing for SSC and Railway NTPC/Group D exams. The app is a React 18 and Vite 5 frontend prototype with 22 screens, a design system, and a PWA manifest. The developer states it has no backend yet, no real notifications, and job data that is hardcoded sample content rather than live government feeds. Features such as the AI Teacher use static replies, the notes library lacks real PDFs, and the leaderboard and user accounts are mock. The developer plans to add a Node.js and Express backend, SQLite storage, and secure PDF delivery in future upgrades.

0
ProgrammingDEV Community ·

Video guide breaks down 12 core building blocks of modern AI architecture

A new video explains how modern AI systems work internally, covering twelve foundational components. It uses everyday analogies such as LEGO bricks, USB-C cables, and open-book exams to make the concepts accessible. The material includes Python SDK code examples and targets software engineers, developers, product managers, and AI enthusiasts. Topics range from tokens and system prompts to RAG, vector databases, AI agents, and the Model Context Protocol (MCP).

0
ProgrammingDEV Community ·

Building a Verified n8n Node Exposed as an AI Agent Tool

The author published n8n-nodes-vatnode, a verified n8n community node for EU VAT number validation. Setting the usableAsTool flag causes n8n to generate an agent-facing Tool variant, where the node's descriptions are read by a language model rather than a person. Evidence for how this flag behaves comes from n8n's type definitions, its node-creation template guide, and its source code, since the flag is not covered in the docs the author checked. The author notes that the Tool variant's behavior with agents was not tested and that availability may depend on instance configuration. To guide model choice, the node's operations state their cost directly in their descriptions, with the offline Check Format operation distinguished from the quota-spending Validate operation.

0
ProgrammingDEV Community ·

Study Compares Four Methods for Detecting Forged Documents

A DEV Community article compares document forgery detection techniques, which it divides into four classes. It uses a hypothetical loan applicant who edited a payslip's figures and re-saved it, showing how the altered file can pass visual and arithmetic checks. The piece explains that manual review relies on human judgement but cannot see file-level edits and does not scale well. Content-based checks such as OCR, consistency testing, visual analysis and cross-checks against outside data are presented as catching errors that structural methods miss. The article argues that no single method is sufficient and that the approaches complement one another.