Open framework uses versioned ADRs in Git to record AI-assisted design decisions and oversight

·1 views

A developer has shared ADRs4AI, a framework for documenting human-AI collaboration in software projects, posted to Hacker News as a Show HN. It addresses oversight of AI-assisted programming through the ORRCF format, in which models must present options, a recommendation, rationale, confidence, and a falsifier for human approval or override. It also uses Architectural Deliberation Records, modeled on Michael Nygard's decision records, to capture the reasoning behind design choices and the conditions under which alternative decisions would apply. The author has developed the approach over more than a year and has recently released a plugin, now submitted to Anthropic's marketplace, to ease onboarding. The post requested feedback on the onboarding materials and the clarity of the online documentation.

Read the full story at Hacker News

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

Discussion (0)

Seven Self-Hosted Jira Alternatives Compared on Migration, Deployment and Workflow Fit

Replacing Jira can affect more than ticket tracking, since projects often rely on long-standing history, workflows, and reporting routines. This guide evaluates seven self-hosted options using four criteria: deployment fit, project and agile coverage, workflow and reporting flexibility, and migration effort. A comparison table lists each tool's best use case, deployment options, AI agent readiness, pricing, and free plan availability. ONES.com is highlighted for teams that need a structured Jira data migration path, with documented mapping of issue types, fields, and users. Other options named include Jira Data Center, GitLab, Redmine, Taiga, Tuleap, and Trac, each suited to different team needs.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

cAdvisor Monitor Found on 13,948 Internet-Facing Hosts on Port 8080

A search-index analysis of cAdvisor, a container resource monitor usually installed alongside Kubernetes, counted 134,520 pages with a loose title match and 13,948 that also answered on port 8080. The researchers found that strict title and application queries returned no results, because the tool has no standard product fingerprint for classification. cAdvisor exposes container names, image names, volume paths, labels, and process details, along with a debug profiling endpoint enabled by default unless disabled. The findings indicate how common the deployment is but do not establish whether these listeners are reachable beyond the scanner's vantage point. The analysis recommends reviewing bind addresses, authentication, and profiling settings for such deployments.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Radar directory aims to help users discover websites through category browsing

The author, who runs the web directory Radar, argues that search engines work well only when users already know what they are looking for. When people want to explore unfamiliar tools, communities, or services, they often cannot form a useful query at all. To address this, Radar separates Search for finding known sites from Discovery for uncovering new options. The platform organizes websites by category, letting visitors pick a field, compare alternatives, and then visit the site they need. The author stresses that browsing complements rather than competes with search, and that it helps users see choices beyond the same few sites that keep appearing in results.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Owe: Gemma 4 drafts replies to avoided messages, with a local check on sending

Owe is an open-source tool that takes a message the user has been avoiding and uses Gemma 4 through the Gemini API to draft a reply of at most 60 words. The model also names the one fact it must not invent and indicates whether the reply can be sent. A local script then checks the draft for word count and for any numbers, emails, or names absent from the original message or user note, and can block sending. Without a user decision, the tool produced a short holding reply, and with a note it produced a reply matching that decision. The message text goes to the Gemini API, so the tool does not work offline, though the final send decision stays on the user's machine.

0 comments Read more at DEV Community