Imported from paniciu/next-tutorial (AGENTS.md). Install upstream with npx skills add paniciu/next-tutorial. Copyright stays with the author.

AGENTS.md

Shared Agent Rules

• docs/requirements.md is the canonical product requirements file.
• Update docs/requirements.md first when scope or terminology changes.
• README.md must stay short and act as a navigation document.
• Requirements are written in Romanian.
• Agent instructions are written in English.
• SkillForge is an agent-first product with server-side LLM access, streaming, persistent profile context, and memory across sessions.
• Never commit or document real secrets.
• All provider and credential access must remain server-side.
• Chat message state belongs to useChat in UI components; global store keeps only shell state (profile, settings, conversation summaries).
• LLM provider calls are allowed only from server routes/actions, never directly from browser components.
• System prompt persona/guardrails must be composed only on the server, only in src/lib/system-prompt.ts; duplicating persona composition elsewhere is an error.
• Any data received from the browser must be normalized server-side before it is interpolated into prompts.
• Message transformations must live in src/lib/message-utils.ts as pure functions (no store access, no DOM APIs, no network calls).
• Text extraction from chat messages must exist in one single project location and be reused by UI + exports.
• Product UI rule: never introduce a global action bar above the conversation area; message actions stay on message hover, export stays in header menu, new chat stays in sidebar.
• Every external integration must include docs/<integration>/README.md with manual setup, environment variables, dashboard configuration, pricing, and official links.
• Every external integration must also update docs/README.md in the same commit with the integration name, the course step, and the new documentation link.

Shared instructions for all coding agents working in this repository.

Skills convention

• Skill source of truth: .claude/skills/
• Copilot mirror: .github/skills/
• Keep mirrors synchronized with sh scripts/sync-skills.sh.
• Validate synchronization with sh scripts/sync-skills.sh --check.

Definition of done before publish

• Run skill pre-deploy before each publish.
• Run sh scripts/sync-agent-instructions.sh after agent-instruction edits.

Agent-specific notes

• Use small, explainable modules.
• Update docs together with behavior changes.
• Do not introduce undocumented external dependencies.
• Review integration docs whenever a new external service appears.