Update Ai Config

Manage AI Config Templates

IMPORTANT: This project uses universal-ai-config (uac). All AI configuration lives in the / directory as universal templates. NEVER edit generated target-specific files (e.g. .claude/, .github/copilot-instructions.md, .cursor/, .codex/, AGENTS.md at root, .agents/skills/) directly — those are overwritten on every uac generate run. Always modify the source templates in /.

When the user wants to add or change AI configuration for this project, follow these steps:

1. Understand the Request

Read the template guide at `` to understand the available template types and when to use each.

2. Determine the Template Type

Based on the user's request, decide which type of template to work with:

• Persistent context, rules, or guidelines → Instruction template
• Repeatable task, workflow, or slash command → Skill template
• Specialized AI persona with restricted tools → Agent template
• Automatic lifecycle automation (on events) → Hook template
• MCP server configuration → MCP template

Examples:

• "Add a rule about error handling" → Instruction
• "Create a deploy workflow" → Skill
• "Set up a code reviewer" → Agent
• "Run linting after file edits" → Hook
• "Add an MCP server for GitHub" → MCP

If the intent is ambiguous, ask the user to clarify before proceeding.

3. Delegate

Read the detailed guide for the chosen template type and follow its instructions:

• Instructions → ``
• Skills → ``
• Agents → ``
• Hooks → ``
• MCP → ``

These skills primarily create or modify files inside /. They may also modify files in directories listed under additionalTemplateDirs in the config — but only with explicit user confirmation, since those are shared templates that may affect other projects. Never edit generated output files in .claude/, .cursor/, .github/, .codex/, AGENTS.md at root, .agents/skills/, or similar target-specific directories — those are regenerated from templates and any direct changes will be lost.

Additional Template Directories

This project may have additional template directories configured via additionalTemplateDirs. To find them, search the project root for all config files matching universal-ai-config.* (e.g. universal-ai-config.config.ts, universal-ai-config.overrides.config.ts, and any other variants) and read the additionalTemplateDirs field from each. If the user asks to update a template that doesn't exist in the main templates directory, or explicitly refers to shared/global/external templates:

1. Read all universal-ai-config.* config files in the project root to find additionalTemplateDirs paths
2. Search those directories for the relevant template
3. IMPORTANT: Before editing any file outside the main / directory, ask the user for explicit confirmation — these are shared templates that may affect other projects

Writing Style

Every token in a template costs model attention — ~60% of instruction tokens are typically removable without degrading output quality. When writing or editing any template, apply these rules:

• Use imperative sentences. "Use strict mode" not "You should try to ensure that strict mode is used."
• Cut filler phrases. Drop "please note that", "it is important to", "keep in mind that", "remember to always" — just state the rule.
• Remove hedging on directives. "Be accurate" not "You should try to ensure responses are accurate."
• Deduplicate. If two rules express the same constraint in different words, keep only the most specific version. "Be professional" + "Maintain professional tone" + "Always respond professionally" → "Maintain professional tone."
• Drop self-referential framing. "The following rules govern your behavior:" adds nothing — just list the rules.
• Strip unnecessary articles in imperatives. "Write response" not "Write the response."
• Flatten verbose conditionals. "If user asks X, check Y, and if Y then Z" → "X + Y → Z"
• Minimize examples. 1–2 examples covering distinct cases beat 5 illustrating the same pattern. If a rule is clear without examples, omit them — but note that cutting examples carries the most risk since few-shot examples significantly reduce prompt sensitivity.

After Changes

Run uac generate to regenerate target-specific config files and verify the output.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: fabis94
• Source: fabis94/universal-ai-config
• License: MIT

Install and usage instructions live in the source repository linked above.