Summary
The model-specific routing in prompt-master is written around Claude Opus 4.7 and GPT-5.x. Both have since been superseded (Opus 4.8 and GPT-5.5), and a few of the prompting recommendations are now out of date. I went through Anthropic's and OpenAI's current official docs and have a set of updates ready. Opening this first to check you're interested before sending a PR.
What's outdated (factual)
Claude block — now Opus 4.8:
- Effort is now the primary intelligence/cost lever (max/xhigh/high/medium/low), more so than on any prior Opus. The current block doesn't mention it.
- Thinking is off by default unless adaptive is enabled; fixed
budget_tokens is no longer supported.
- Opus 4.8 auto-calibrates response length to task complexity, so the "always specify length" advice now has a clearer rationale.
- Literal-instruction-following is stronger (state scope explicitly); it favors reasoning over tool calls and spawns fewer subagents by default.
ChatGPT block — now GPT-5.5:
- Outcome-first prompting (describe the goal + success criteria + stopping rules; drop step-by-step process unless the path matters) is OpenAI's headline shift and isn't reflected.
reasoning.effort (default medium) and text.verbosity are now separate levers, and higher effort isn't automatically better.- New guidance: avoid unnecessary absolute rules (use "if X then Y" decision rules), put tool guidance in tool descriptions, use Structured Outputs over in-prompt schemas, and don't include the current date.
Sources: Anthropic prompting best practices (platform.claude.com/docs/en/build-with-claude/prompt-engineering/claude-prompting-best-practices) and OpenAI's "Using GPT-5.5" + prompt guidance (developers.openai.com/api/docs/guides/latest-model).
Optional addition (my own preference, take it or leave it)
For my own use I also added (a) a mandatory ⚙️ Settings output line that names the recommended model + effort/thinking setting per prompt, and (b) a small "Model & Effort Settings Reference" table to back it. This reflects my workflow rather than a correctness fix, so I've kept it separate — happy to include it, drop it, or split it into its own PR, whatever fits your vision for the skill.
Proposed next step
If you're open to it, I'll send a PR with the factual updates (and the optional addition flagged separately so it's easy to accept or reject). Let me know which parts you'd want. Thanks for the skill — it's great work!
Summary
The model-specific routing in
prompt-masteris written around Claude Opus 4.7 and GPT-5.x. Both have since been superseded (Opus 4.8 and GPT-5.5), and a few of the prompting recommendations are now out of date. I went through Anthropic's and OpenAI's current official docs and have a set of updates ready. Opening this first to check you're interested before sending a PR.What's outdated (factual)
Claude block — now Opus 4.8:
budget_tokensis no longer supported.ChatGPT block — now GPT-5.5:
reasoning.effort(default medium) andtext.verbosityare now separate levers, and higher effort isn't automatically better.Sources: Anthropic prompting best practices (
platform.claude.com/docs/en/build-with-claude/prompt-engineering/claude-prompting-best-practices) and OpenAI's "Using GPT-5.5" + prompt guidance (developers.openai.com/api/docs/guides/latest-model).Optional addition (my own preference, take it or leave it)
For my own use I also added (a) a mandatory
⚙️ Settingsoutput line that names the recommended model + effort/thinking setting per prompt, and (b) a small "Model & Effort Settings Reference" table to back it. This reflects my workflow rather than a correctness fix, so I've kept it separate — happy to include it, drop it, or split it into its own PR, whatever fits your vision for the skill.Proposed next step
If you're open to it, I'll send a PR with the factual updates (and the optional addition flagged separately so it's easy to accept or reject). Let me know which parts you'd want. Thanks for the skill — it's great work!