Prompting

Reuse curated prompt templates for coding, writing and analysis

Work with strict scientific rigor, objective critique and formal correctness

Optimize LLM prompts with few-shot, chain-of-thought and structured outputs

Present explicit choices to the user and pause execution until they answer

Switch a voice assistant to the Mars persona for deep, reflective conversations

Switch a voice assistant to the Venus persona for fast logistics and calendar work

Review agent tool placement and defensive prompts when tools or subagents change

Audit any agent system prompt against a quality bar and final review rubric

Default process for open-ended tasks that touch files, the shell and the web

Make AI replies less robotic and listy in casual or emotional chats

Turn a rambling outcome description into one polished autonomous goal prompt

Test marketing prompts with A/B evals, version history and ready-made templates

Version, A/B test and regression-check prompts before they reach production

Pause mid-task to zoom out and check direction, assumptions and bias

Optimize prompts and score LLM outputs and RAG retrieval against eval sets

Stop agents from inventing APIs by checking methods exist in your installed version

Turn a fuzzy intention into a goal with checkable done criteria and a budget

Turn a one-line idea into a standalone goal brief that an autonomous run can finish

Create, version and manage managed prompts in Agent Platform

Review recent failures and improve prompts, skills and memory in place mid-session

Coach with a nagging-mom reminder style while keeping evidence-first debugging

Add self-evolution notes, KPI summaries and flavor switching to PUA sessions

Push the agent to exhaust every fix after repeated failures or stalled work

Cancel a running PUA loop from Codex with the pua-cancel-loop command

Apply performance-review pressure to force exhaustive, structured debugging

Switch PUA personality flavors from Codex with the pua-flavor command

Push the agent to exhaust all fixes with a Japanese corporate-style debug loop

Show PUA KPI-style progress summaries from Codex with the pua-kpi command

Start a PUA improvement loop from Codex with the pua-loop command

Run the nagging-mom PUA flavor from Codex with the pua-mama command

Turn PUA mode off in Codex with the pua-off command

Run PUA offline mode from Codex with the pua-offline command

Turn PUA mode on in Codex with the pua-on command

Run the P10 CTO-level PUA persona from Codex with the pua-p10 command

Run the P7 senior engineer PUA persona from Codex with the pua-p7 command

Run the P9 tech lead PUA persona from Codex with the pua-p9 command

Run the PUA Pro extensions from Codex with the pua-pro command

Clean up orphaned PUA processes from Codex with the pua-reap-orphans command

Run a PUA survey from Codex with the pua-survey command

Check PUA agent team status from Codex with the pua-team-status command

Tear down all PUA sessions from Codex with the pua-teardown-all command

Push Trae agents past repeated failures and unverified completions on request

Switch to the encouraging PUA coaching voice from Codex with pua-yes

Inject a compact PUA persona reference into sub-agents or short sessions

Switch to an encouraging, praise-heavy PUA coaching voice for morale

Test a question at narrow and wide scopes to surface design tensions

Brainstorm diverse candidate ideas and surface the strongest options

Audit AI instruction sets for over-prompting and cut rules a smarter model needs

Run 240+ prompt patterns for summaries, analysis, threat models and reviews