ai-agent-builder
AI tooling: write/merge/audit skills and plugins, package safety, MCP servers, ChatGPT apps, model choice, context budget. Skill yaz, MCP yap, hangi model daha iyi.
SKILL.mdskills/claude/ai-agent-builder/SKILL.md
GitHubAI Agent Builder
Build and maintain AI tooling: skills, plugins, MCP servers, ChatGPT apps, agent configuration and model choice.
Choose the module
| Need | Module |
|---|---|
| Write, edit, split, merge or port a skill | writing-skills |
| Compare, repair or improve existing skill variants without regressions | safe-skill-updater |
| Trust and security check of a downloaded skill, plugin, ZIP or repo before installing | skill-package-security-audit |
| Build an MCP server | mcp-builder |
| ChatGPT Apps SDK app (MCP server + widget) | chatgpt-apps |
| Convert a Claude Code agent kit to Codex (or back) | codex-agent-dev-kit |
| Context, skill or tool budget audit | context-budget |
| Which AI model, provider or subscription for a task | ai-model-route-evaluator |
Host facts (verify against current docs before relying on them)
- Codex and the ChatGPT desktop app read personal skills from
~/.agents/skills(and~/.codex/skills); per-skill enable/disable lives in~/.codex/config.tomlunder[[skills.config]]. - ChatGPT web and mobile use skills bundled in plugins, or Plugins → Skills → Create → Upload on eligible plans. Desktop and web skills do not sync.
- Claude Code reads
~/.claude/skills; claude.ai uploads one ZIP per skill (the skill folder at the ZIP root, at most 200 files, description at most 200 characters in the uploader). - Skill names: at most 64 characters, lowercase letters, digits and hyphens; never "claude" or "anthropic". Descriptions: third person, front-loaded with triggers, no XML tags.
Yiğit's skill workflow (single source of truth)
- Master copy:
<SKILL_MASTER>/.skills/holds every skill with its long bilingual description;variants/codex|claude|cloud/hold platform-specific copies (e.g. the Weizhenaresearchfamily, the <=200-file cloud subset ofai-research-skills);claude-descriptions.jsonholds <=200-char descriptions for claude.ai;targets.jsonlists ChatGPT-only, local-only and install-location overrides;externals/holds agent files (web-search agents). - Two kinds of skills: (1) Yiğit's own hubs — personal skills merged as
modules/<name>/MODULE.md; (2) upstream skills he supplied from the web (ui-ux-pro-max family, caveman, Weizhena research, primary-source-research, ai-research-skills, gemini-deep-research, parallel-web, graphify) — kept SEPARATE, complete and current, never shortened; his customized copy lives inside the same skill underyigit/. Never duplicate a skill in two places. - To update an upstream skill: pull the new upstream version into
skills/<name>(keepyigit/and the appended "Yiğit" sections), keep the original description and append Turkish triggers. - Never edit installed copies directly. Edit the master, then run
python <SKILL_MASTER>/sync_skills.py(--check= drift report only). It deploys to Codex/ChatGPT desktop (~/.agents/skills, Weizhena to~/.codex/skills), Claude Code (~/.claude/skills), agent files, and rebuilds the Desktop upload ZIPs; OpenCode follows automatically. Archive retention followstargets.json→keep_backups; Yiğit disabled it, so do not create archive/backup copies. - After a sync, cloud copies (ChatGPT web/mobile, claude.ai) remain stale until the current
Desktop/Yigit-Skill-Paketleri/ChatGPT.zipplugin or the individual skill ZIPs insideClaude.zipare re-uploaded. This Desktop folder contains exactly these two files; both have the same personal skill list. Literature examples remain separate. Do not recreate the old detailed upload folders or archives.
Rules
- Respect the user's archive-retention preference. Validate source and targets before deployment; do not create backups when
keep_backupsis false. - Never execute downloaded scripts merely to inspect them.
- Test triggering with a few realistic prompts after changing a description.
Module map
Open only the module(s) the request needs and read the module file completely before acting. Several modules may combine in one task.
| Module | Use when | File |
|---|---|---|
writing-skills |
Create, edit, split, merge, port, or validate agent skills and SKILL.md packages. | MODULE.md |
safe-skill-updater |
Audit, compare, merge, repair, or improve user-owned Codex and ChatGPT skill variants while preserving proven behavior. | MODULE.md |
skill-package-security-audit |
Perform a static, read-only security and trust preflight on a downloaded or third-party skill, plugin, ZIP, repository, installer, or update. | MODULE.md |
mcp-builder |
Guide for creating high-quality MCP (Model Context Protocol) servers that enable LLMs to interact with external services through well-designed tools. | MODULE.md |
chatgpt-apps |
Build, scaffold, refactor, and troubleshoot ChatGPT Apps SDK applications that combine an MCP server and widget UI. | MODULE.md |
codex-agent-dev-kit |
Convert Claude Code agent-kit setups into Codex-compatible skills, project instructions, guardrail scripts, command workflows, and subagent guidance. | MODULE.md |
context-budget |
Audit actual context consumption and routing overhead across instructions, skill metadata and bodies, tool schemas, agent definitions, conversation output, and project f… | MODULE.md |
ai-model-route-evaluator |
Compare current AI models, providers, gateways, subscriptions, local routes, and free/paid endpoints for a concrete task. | MODULE.md |
Supporting files (open only when the module points to them):
writing-skills: anthropic-best-practices.md, CLAUDE_MD_TESTING.md, persuasion-principles.md, testing-skills-with-subagents.mdsafe-skill-updater: evaluation-and-rule-promotion.md, model-provider-policy.md, quality-preservation-rubric.md; scripts:modules/safe-skill-updater/scripts/analyze_skill_corpus.pyskill-package-security-audit: review-checklist.md; scripts:modules/skill-package-security-audit/scripts/inspect_skill_package.pymcp-builder: evaluation.md, mcp_best_practices.md, node_mcp_server.md, python_mcp_server.md; scripts:modules/mcp-builder/scripts/connections.py,modules/mcp-builder/scripts/evaluation.pychatgpt-apps: app-archetypes.md, apps-sdk-docs-workflow.md, interactive-state-sync-patterns.md, repo-contract-and-validation.md, search-fetch-standard.md, upstream-example-workflow.md, window-openai-patterns.mdcodex-agent-dev-kit: claude-to-codex-map.md, codex-plugin-layout.mdai-model-route-evaluator: evaluation-scorecard.md