JetBrains AI Prompts

Brief Junie, the JetBrains autonomous coding agent, to implement [FEATURE_DESCRIPTION] in [PROJECT_NAME] inside [IDE_NAME]. Scope it to the [MODULE] module ([FILES] files). Require Junie to: 1) produce a plan first, 2) write the multi-file code, 3) run [TEST_COMMAND] and iterate on failures using debugger-driven iteration, 4) summarize every change. Review the plan before it codes, and set the quality bar: [CRITERIA]. Approve the final diff only after tests are green.

Use Junie Plan Mode in [IDE_NAME] to design [FEATURE_OR_REFACTOR] before any code is written. Ask Junie to produce a requirements summary and a technical design covering [COMPONENT_1], [COMPONENT_2], the data flow for [FLOW], and the test strategy for [TEST_SCOPE]. Review the plan against [ARCHITECTURE_DOC], challenge the [RISKY_DECISION], and only then authorize implementation. Save the approved plan to [PLAN_FILE] as the implementation contract.

Configure unlimited Mellum-powered code completion in [IDE_NAME] for daily [LANGUAGE] development at [COMPANY]. Enable Next Edit Suggestions so the IDE predicts my next change, not just the current line. Since Mellum completion costs no AI credits, use it freely across [PROJECT_1] and [PROJECT_2], and reserve paid cloud model credits for [CLOUD_USE_CASE]. Measure the acceptance rate over [WEEKS] weeks and report it to [STAKEHOLDER].

Run an AI chat refactor session in [IDE_NAME] on [FILE_OR_MODULE] ([LINES] lines of [LANGUAGE]). Ask the chat, with full project context, to: explain what [FUNCTION_NAME] does, propose [NUMBER] refactorings that reduce [CODE_SMELL], and apply the safest one. Then ask it to generate the missing docstrings and a commit message following [COMMIT_STYLE]. Review each change in the diff view; the IDE deep code understanding makes suggestions structurally sound, but you still own the review.

Connect [MCP_SERVER_NAME] to JetBrains AI via Model Context Protocol in [IDE_NAME] so agents can [CAPABILITY, e.g. query our internal API docs]. Configure the MCP server in [CONFIG_PATH] with [AUTH_METHOD], expose the tools [TOOL_1] and [TOOL_2], and test that Junie can call them during a [TASK] task. Document the setup in [DOC_FILE] and restrict the tools Junie may call to the allowlist in [ALLOWLIST_FILE] for the [ENVIRONMENT] environment.

Run [EXTERNAL_AGENT, e.g. Claude Code] inside [IDE_NAME] through the Agent Client Protocol instead of switching tools. Point it at [PROJECT_PATH], give it the task '[TASK_DESCRIPTION]', and let it use the IDE file operations and terminal. Compare its output against Junie on the same task for [METRIC_1] and [METRIC_2]. Document which agent you prefer for [TASK_TYPE_A] vs [TASK_TYPE_B] in [COMPARISON_DOC] for the team.

Configure JetBrains AI with Bring Your Own Key plus local models for [TEAM_NAME]. Add API keys for [PROVIDER_1] via [AUTH_METHOD] for cloud features, and connect Ollama with [LOCAL_MODEL] for offline work on [SENSITIVE_PROJECT]. Define the routing rule: [RULE, e.g. local model for drafts, cloud model for reviews]. Verify no code from [SENSITIVE_PROJECT] leaves the machine, and document the setup in [SETUP_GUIDE] for [NUMBER] engineers.

Set up Junie code review on pull requests for [REPO_NAME] via GitHub Actions. Configure the workflow to trigger on PRs to [BRANCH], have Junie review with focus on [FOCUS_AREAS], and post findings as review comments. Define severity handling: [SEVERITY] findings block merge, lower ones are advisory. Compare Junie findings against [OTHER_TOOL] on the first [NUMBER] PRs and tune the focus areas in [WORKFLOW_FILE] until the signal is clean.

Plan JetBrains AI credit usage for [TEAM_SIZE] engineers: 1 credit equals USD 1, and only cloud features (chat with frontier models, Junie runs) consume credits; Mellum completion and local models are free. Allocate [CREDITS] credits per engineer per month, prioritizing [PRIORITY_USE]. Set up [ALERT_METHOD] alerts at [PCT]% spend, define the top-up approval flow through [APPROVER], and choose the tier ([AI_PRO_OR_ULTIMATE]) based on the [WEEKS]-week pilot burn rate.

Use the AI-powered full method generation in [IDE_NAME] (2026.2+) for [PROJECT_NAME]. When calling a method that does not exist yet, let the AI generate the signature stub and the complete implementation, then tab-to-accept. Apply it to [METHOD_1] in [CLASS_1] and [METHOD_2] in [CLASS_2], following the [STYLE_GUIDE]. Review each generated method for [CONCERN, e.g. null safety], add tests in [TEST_FILE], and report the time saved versus hand-writing to [STAKEHOLDER].