Prompt Optimizer Plugins

A dual-protocol marketplace containing focused prompt optimizers for GPT-5.6 and Claude Fable 5. The same Skill files are packaged for both Codex and Claude Code, so behavior stays consistent across the two runtimes.

Included plugins

Plugin	Target	Codex Skill	Claude Code Skill
optimize-gpt-5-6-prompts	GPT-5.6	$optimize-gpt-5-6-prompts	/optimize-gpt-5-6-prompts:optimize-gpt-5-6-prompts
claude-fable-5-prompt-optimizer	Claude Fable 5	$optimize-claude-fable-5-prompts	/claude-fable-5-prompt-optimizer:optimize-claude-fable-5-prompts

Both optimizers preserve explicit requirements, ask targeted questions when missing information would materially change the result, and keep runtime configuration separate from prompt text.

Install with Codex

Add the marketplace once:

codex plugin marketplace add IchenDEV/prompt-optimizer-plugins

Install either or both plugins:

codex plugin add optimize-gpt-5-6-prompts@optimize-gpt-5-6-prompts
codex plugin add claude-fable-5-prompt-optimizer@optimize-gpt-5-6-prompts

The Codex marketplace ID remains optimize-gpt-5-6-prompts so existing GPT-only installations can continue to upgrade after the repository rename.

Install with Claude Code

Add the same GitHub marketplace:

claude plugin marketplace add IchenDEV/prompt-optimizer-plugins

Install either or both plugins:

claude plugin install optimize-gpt-5-6-prompts@prompt-optimizer-plugins
claude plugin install claude-fable-5-prompt-optimizer@prompt-optimizer-plugins

Claude Code namespaces plugin skills with the plugin name. Use the full slash-command names shown in the table above.

What the optimizers do

• Preserve the original goal, facts, constraints, output contract, and prompt-layer boundaries.
• Ask one to three focused questions when outcome, input, evidence, schema, or permission boundaries are materially unclear.
• Avoid producing a provisional prompt while blocking information is missing.
• Rewrite clear requests into direct prompts suited to the target model.
• Define assessment versus implementation, authorization boundaries, evidence, verification, and completion where relevant.
• Replace requests for hidden chain of thought with concise rationale, supporting evidence, and checks.
• Use complex structure only when it improves behavior.

Dual-protocol layout

.agents/plugins/marketplace.json         # Codex marketplace
.claude-plugin/marketplace.json          # Claude Code marketplace
plugins/
├── optimize-gpt-5-6-prompts/
│   ├── .codex-plugin/plugin.json
│   ├── .claude-plugin/plugin.json
│   └── skills/optimize-gpt-5-6-prompts/
└── claude-fable-5-prompt-optimizer/
    ├── .codex-plugin/plugin.json
    ├── .claude-plugin/plugin.json
    └── skills/optimize-claude-fable-5-prompts/

Each plugin owns its Skill directory. Nothing depends on paths outside the plugin root, so both runtimes can safely copy plugins into their versioned caches.

Validation

The repository is checked with:

• the Codex Plugin validator for both plugins;
• the Codex Skill validator for both skills;
• claude plugin validate for the marketplace and both plugins;
• clean-environment installation tests for both Codex and Claude Code;
• representative forward tests for clarification, prompt-only output, external actions, long-running work, hidden-reasoning requests, and long-document question answering.

Official sources

• Codex: build skills
• Codex: build plugins
• Claude Code: create plugins
• Claude Code: create a plugin marketplace
• Prompting Claude Fable 5
• Using GPT-5.6

License

MIT