Path-specific Copilot custom instructions: how to scope rules to files in a repo

·3 views

Teams often put every Copilot rule into one repository file, so unrelated guidance applies everywhere and the output gets noisy. The article proposes splitting rules into small files under .github/instructions/, each scoped to matching files via glob patterns like applyTo. It explains the three kinds of repository instructions: repository-wide, path-specific, and agent files such as AGENTS.md. Support differs by environment, since path-specific files are not read by every Copilot surface, such as VS Code code review or the JetBrains chat. The author demonstrates a setup for a fictional TypeScript bike-rental repo and includes a script that detects broken glob patterns.

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)

How Developers Should Use AI in Software Development in 2026

A guide published on DEV Community argues that the central question for software engineering in 2026 is whether teams can manage the large volume of code AI now produces, rather than whether to use AI at all. It traces the evolution of AI tooling from basic autocomplete to multi-file copilots and now to agentic systems linked through the Model Context Protocol. The author notes that while engineering speed has risen, technical debt has grown too, shifting the bottleneck toward system design, validation, and security. The piece describes three waves of AI-assisted development and the distinct failure modes each one carries, from context gaps to autonomous agents introducing unmaintainable code. It recommends constrained agent workflows, sandboxed execution, and automated static analysis and testing before AI-generated changes reach human review.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Criminal Networks Adapt Tactics as India's Digital Economy Expands

India's growth in online banking, digital payments and e-commerce has made daily activities easier but has also opened new avenues for criminal groups. Offenders are reportedly shifting to digital tools that let them coordinate across locations with little direct contact. Operations are often split among loosely linked individuals handling communication, money movement and account access, which complicates attribution. Investigators therefore have to trace relationships among digital identities, financial flows and participants rather than single transactions. The article notes that such shifts pose fresh challenges for Indian law enforcement while requiring safeguards for privacy and lawful evidence handling.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

NVIDIA's auto-ontology: A Chinese translation of its context layer for database agents

A Chinese translation of the documentation for NVIDIA's auto-ontology project was published on November 5, 2026, on GitHub. The Python, Apache-2.0 tool compiles database metadata into an ontology that maps business terms to columns, stores reviewed metric SQL, and defines table relationships. Agents query through this layer instead of raw schema, which the project's documentation says reduces wrong answers and lets agents flag questions they cannot answer. The README also includes a Docker Compose quickstart that uses the Wide World Importers sample database. The original repository had about 4,400 stars at the time of the report.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

TrueNorth open-source tool flags wrong figures before forms are submitted

A developer built TrueNorth, an open-source verification engine that checks whether a value is plausible in context and matches its supporting document. The idea followed an April 2024 incident in which a Citigroup employee entered an $81 trillion payment instead of $280, caught only after about 90 minutes. The tool runs as a Next.js web workspace for reviewing documents and as a Chromium extension that holds back suspicious values in live forms. Verification combines deterministic rules, statistical outlier checks, and Gemma 4 reasoning for cross-field validity, with human review retaining final approval. The project was built in one day at Hacktoberfest '26 at NIC Faisalabad, where it won first place, Best Open-Source AI Project, and a special Best Use of Gemma 4 award.

0 comments Read more at DEV Community