AI Agent Workflow Collection

Use when the user gives a short or messy request that doesn't explain the full details, but the 'right way' to do it is already visible in the environment. It should be triggered by casual phrasing like “finish the rest of the list,” “do it just like the others,” “update the records using the existing style,” or “add a new entry to the site following the current layout.” This applies when an agent must look at previous examples, templates, or sibling folder structures to figure out what fields to fill, what formats to use, or where exactly to put a file without being told explicitly.

Use when the user asks a question in one language (e.g., German, Spanish, Chinese, Arabic) but the information needed is stored in documents of a different language (e.g., English), or split across documents in multiple languages. Trigger it for requests like 'find the answer from these English reports and tell me in German', 'compare what the local news and the international news say', or 'check if the model flips to English when I ask a question in Spanish about an English text'. This skill targets the unique challenge of performing complex reasoning (comparison, aggregation) across language boundaries while maintaining the correct response language.

Use when the user wants messy-data questions where the agent has to clean first and analyze later.

Use when the user wants translation data with short replies, subtitles, or conversational fragments where part of the meaning is missing unless you read the previous line. Trigger it for requests like "the second sentence leaves words out," "the answer is elliptical," or "test whether the translator can fill in what is omitted from context instead of translating the fragment too literally.

Use when the writing task requires moving beyond a simple list of facts to produce reasoned insights, justify a verdict, or synthesize massive amounts of fragmented evidence. Trigger it for requests like 'write an investigative report,' 'analyze the trends,' 'give me a 360-degree view of these 100 documents,' 'explain why this claim is false based on the evidence,' or 'draft a legal reasoning section.' It perfectly resolves tasks where raw data must be explicitly mapped to overarching conclusions, truths, or rules.

Use this when the user wants to solve a large, multi-faceted project by breaking a distal goal into a sequence of abstract middle-states, sub-goals, or typed nodes like research versus writing. Trigger it for requests like 'break this big project into different types of steps', 'recursively decompose this goal until you have simple steps', 'give me a high-level strategy and then the moves to do it', or 'organize a plan where composition steps wait for reasoning steps.

Use when transforming massive documents or multi-source corpora into structured, coherent formats. Trigger for requests like 'summarize this huge PDF without losing the thread', 'draft a defense statement based on this complaint and the evidence list', 'create a presentation script from this report', or 'write the official trial facts by resolving contradictory testimonies into a ground-truth timeline'. This is critical when shifting text registers, interleaving visual anchors, or composing professional narrative sections from conflicting multi-source contexts.

Use when the user expects the agent to act as a deeply personalized assistant by synthesizing facts, schedules, and persona traits scattered across long conversation histories, or when the agent must proactively use past history to steer the conversation. It is triggered when the agent must detect implicit user preferences, identify schedule conflicts, distinguish between different speakers in group chats, or proactively mention past topics to build rapport. Everyday examples include: 'What should I do this weekend based on what I like?', 'Do you remember which one of us ordered the vegetarian meal?', and 'Bring up things we talked about last week to cheer me up.

Use when the user wants questions where the docs look relevant but still do not contain the answer. Trigger it for requests like 'make the model say it can't tell', 'give me questions with not enough information', 'test whether it refuses instead of guessing', or 'include related documents but no real answer'. This skill is for data that teaches evidence-bounded abstention in a read-only RAG setting.

Use when the user wants translation data where a repeated term must stay the same across a whole document, such as product names, policy labels, medical terms, or recurring entities. Trigger it for requests like "keep the wording consistent across paragraphs," "do not rename the same thing every time," or "test whether the translator locks repeated terms instead of drifting between alternatives.