Pustaka Prompt
AI Agents
Ready prompts per unit. Fill in the blanks, check the preview, copy it into the AI tool you are allowed to use at work.
25 prompts
Prompt: draft a project instruction file
Use it when: A developer agent is about to work in your project and you need starting rules that can be checked, not general advice.
Works with any AI assistant.
Your prompt
Goal: I need a draft project instruction file for a developer agent in the ‹proyek› project. Reader: the AI agent that reads it at the start of every session, and the teammate who reviews it. Material: our check commands: ‹perintah›. Past incidents: ‹kejadian›. Limits: at most 15 rules; every rule must be checkable from the trace or the commit; no general explanations. Shape: a numbered list, one line per rule, grouped: checking work, how we work, prohibitions. Check: mark any rule you think cannot be checked, and propose a version that can.
Summariser agent instructions for your folder
Use it when: You want a read-only agent to summarise one folder with a source on every point, without guessing or changing files.
Works with any AI assistant.
Your prompt
Goal: I need a one-page briefing from the folder ‹folder›. Reader: ‹pembaca›. Material: only the files in this folder; do not use outside knowledge. Limits: do not guess; if it is not written, write "not found in the files". Do not change, move or delete files. Shape: numbered points about ‹topik›. Each point ends with the file name and one short quote. Check: list the files read and not read, with the reason.
Learn it in: Project: a folder summariser agent with citations
A draft instruction file for your team
Use it when: Your team is starting to work with agents and needs one shared rules file loaded at the start of every session.
Works with any AI assistant.
Your prompt
I am drafting a project instruction file for the ‹tim› team, who work with agents. Goal: rules the agent reads at the start of every session. Material: the team's work: ‹pekerjaan›. Test and build commands: ‹perintah›. Limits: only checkable rules; no adjectives like "careful" or "great". Shape: five parts: project goal, commands, working rules, what done means, steps that are never alone (‹larangan›). Check: mark every rule that must also be enforced through the permission settings.
Learn it in: Project: a team instruction file and permission rules
Help turning one real failure into a test task
Use it when: An agent just failed on a real task and you want that failure to become a test task run before every change.
Works with any AI assistant.
Your prompt
Goal: I want to turn one agent failure into a test task. Reader: the team that will run this test set before every change. Material: the user's request: ‹permintaan›. What went wrong: ‹kesalahan›. Material cleaned of personal data: ‹bahan›. Limits: do not rewrite the correct answer word for word; write conditions. Shape: input, must include, must not include, how to grade, reason. Check: make sure every condition can be ticked by someone else without asking me.
Assessing an MCP server before installing it
Use it when: Before you install someone else's MCP server into your agent, you need to weigh its tools, secrets and data flow.
Works with any AI assistant.
Your prompt
Goal: I want to assess the MCP server ‹server› before installing it on the agent ‹agen›. Material: the tool list from tools/list and notes from its author's page: ‹bahan›. Limits: install or run nothing; treat annotations as untrusted. Shape: a tool table (name, read or write, risk, proposed permission), then five answers: author, tools, secrets requested, where data goes, how to pin the version. Check: list the questions the server's author must answer before I install it.
Learn it in: MCP from the inside, and how to choose a server
