Agent System Prompt Scaffold
Generates a production agent system prompt with a scoped role, tool-use policy, refusal boundaries, escalation rules, and a matching set of test cases.
{{PURPOSE}}{{TOOLS}}{{USERS}}+3 moreLibrary
Copy-ready prompts that hold up in real work. Every entry lists the variables you need to fill in, the models it has been tested against, and a short note on the technique it uses — so you can adapt it rather than just paste it.
Generates a production agent system prompt with a scoped role, tool-use policy, refusal boundaries, escalation rules, and a matching set of test cases.
{{PURPOSE}}{{TOOLS}}{{USERS}}+3 moreReads an agent execution trace to find the first step where reasoning went wrong, then maps the failure to a class and a specific prompt or tool fix.
{{SYSTEM_PROMPT}}{{TOOL_DEFINITIONS}}{{TRACE}}+1 moreA code review prompt that separates correctness bugs from style opinions, rates severity, and refuses to comment on anything outside the diff.
{{STACK}}{{INTENT}}{{CONVENTIONS}}+1 moreTurns an error and its stack trace into ranked hypotheses with the specific evidence that would confirm or eliminate each one, before proposing any fix.
{{STACK_TRACE}}{{CODE}}{{RECENT_CHANGES}}+1 moreForces the model to characterize existing behavior with tests before touching any code, so refactors preserve behavior instead of quietly changing it.
{{CODE}}{{STACK}}{{GOAL}}Converts a business question into SQL by first restating assumptions and definitions, then writing the query with a validation check the analyst can run.
{{QUESTION}}{{DIALECT}}{{SCHEMA}}+1 moreReviews summary statistics and sample rows to separate genuine data quality problems from real-world variation, ranked by impact on the intended analysis.
{{ANALYSIS_GOAL}}{{PROFILE}}{{SAMPLE_ROWS}}+1 moreBuilds image generation prompts using real photographic language — lens, lighting setup, surface, and camera position — plus a negative list and variation axes.
{{PRODUCT}}{{MATERIAL}}{{USE_CASE}}+2 moreGenerates three distinct conceptual approaches for an article illustration, each with the visual metaphor stated first and the rendering specification second.
{{THESIS}}{{PUBLICATION_STYLE}}{{PLACEMENT}}+2 moreDerives a positioning statement from customer pain and competitive alternatives, then generates headline variants that each test a different message hypothesis.
{{PRODUCT}}{{SEGMENT}}{{ALTERNATIVE}}+3 moreTurns one customer case study into a four-email outbound sequence where each message uses a different angle and no email repeats the previous ask.
{{CASE_STUDY}}{{PROSPECT_ROLE}}{{PROSPECT_COMPANY_TYPE}}+2 moreExtracts decisions, owned actions, and unresolved questions from a transcript, and refuses to assign an owner or date the meeting never established.
{{TRANSCRIPT}}{{ATTENDEES}}{{PURPOSE}}Sorts a task list against stated goals and a real hours budget, forcing explicit cuts and naming the cost of each thing that does not get done.
{{GOALS}}{{TASKS}}{{HOURS}}+2 moreProduces a research brief where every claim is tagged with its source and confidence, and unsupported assertions are quarantined instead of blended into the text.
{{QUESTION}}{{SOURCES}}{{DECISION_CONTEXT}}Builds a comparison matrix across competitors using criteria derived from buyer needs, with explicit unknowns rather than plausible-sounding filler in empty cells.
{{CATEGORY}}{{COMPETITORS}}{{BUYER_PROFILE}}+1 moreTurns a topic and audience into a structured outline with per-section word budgets, a stated thesis, and an explicit list of what the post will not cover.
{{TOPIC}}{{AUDIENCE}}{{WORD_COUNT}}+1 moreConverts rough notes into a first draft that imitates a supplied writing sample, with strict rules against inventing facts the notes do not contain.
{{VOICE_SAMPLE}}{{NOTES}}{{FORMAT}}+1 moreA two-phase editing prompt: the model first diagnoses concrete problems in your draft against named criteria, then rewrites only the passages it flagged.
{{DRAFT}}{{VOICE_NOTES}}The AI signal without the hype — new models, tools worth your time and what actually shipped. One email, no vendor pitches.
The AI signal without the hype — new models, tools worth your time and what actually shipped. One email, no vendor pitches.