Prompt library

Do a read-only test gap review. Look at recent changes, critical paths, error handling and edge cases. Explain what the existing tests actually prove, and list missing tests by risk; for each, give the failure mode, setup, assertions and where it should live. Do not edit any files. Stop once you've produced the above; don't go on to anything else.

Do a read-only analysis of dependency upgrades for this project. Identify the runtime, package manager, lockfile and CI; split the upgrades into batches by risk and coupling, and list breaking changes, verification commands and rollback points. Do not install or change any dependencies. Stop once you've produced the above; don't go on to anything else.

Do a read-only quick accessibility audit of the current UI, covering semantic structure, keyboard use, focus, form labels, status feedback, contrast, 200% zoom and reduced motion. For each issue, give the location, user impact, repro steps and the smallest fix, ordered by severity. Stop once you've produced the above; don't go on to anything else.

Do a read-only assessment of whether this project is ready to release. Check the build, tests, data migrations, environment variables, secrets, caching, health checks, monitoring, backups, smoke tests and rollback. Sort the results into pass, blocker and needs human sign-off. Do not deploy. Stop once you've produced the above; don't go on to anything else.

Find the 3 highest-risk behaviors in this project that no test covers (look first at recent changes, critical paths and error handling), add one test for each, and run them to confirm they pass. Only add tests; don't change application code. If you spot a suspected bug, put it in the summary and don't fix it. Stop after those 3, and finish by listing which tests you added and what each one proves.

Do a read-only check of this project's docs against the implementation. Focus on install, architecture, public behavior, config, deployment and troubleshooting. For each mismatch, cite the doc location and the implementation evidence, explain the impact, and propose the smallest fix. Do not edit any files. Stop once you've produced the above; don't go on to anything else.

Write a codebase tour for someone new to this project. Trace the entry points, one end-to-end flow, data boundaries, config, tests and deployment; cite real files, and list safe first changes, dangerous shortcuts and unknowns. Show me the outline first, and only after I approve, create a single doc file. Stop once you've produced the above; don't go on to anything else.

Turn this conversation into a handoff note the next agent can work from. Keep only confirmed decisions, user preferences, constraints, key evidence, unfinished work and next steps; list open questions and rejected options separately. Do not fill in missing facts. Stop once you've produced the above; don't go on to anything else.

Turn the current troubleshooting into a handoff: impact, timeline, environment, repro steps, key logs, confirmed facts, hypotheses still to test, actions already tried and the next safe check. Redact credentials. Do not run any new commands. Stop once you've produced the above; don't go on to anything else.

Turn the repeated work in this conversation into a workflow: trigger, inputs, ordered steps, checks, stop conditions, approval gates and final deliverable. Separate steps that need judgment from steps that can be automated. End with a prompt I can copy. Do not run the workflow. Stop once you've produced the above; don't go on to anything else.

Review the prompt below and find ambiguity in its goal, scope, permissions, stop conditions and definition of done. Pay special attention to deleting, overwriting, installing, committing, publishing, sending messages, credentials and actions on external accounts. List the risks first, then give the smallest safe rewrite that keeps the original goal. Stop once you've produced the above; don't go on to anything else.

Turn the material below into a short backlog. Separate confirmed decisions, open questions, actionable tasks and parked ideas; for each task, add the user outcome, dependencies, definition of done and priority. Don't decide product direction for me, and don't do the tasks. Stop once you've produced the above; don't go on to anything else.

Do a read-only audit of this project's long-term agent instructions. Find AGENTS.md, CLAUDE.md, the README and any tool-specific rules; group them by scope, and list duplicates, conflicts, stale paths and content that belongs in regular docs. Output the audit, a draft merge and the questions I need to decide. Do not edit, install, commit or publish anything. Stop once you've produced the above; don't go on to anything else.

Read my recent session history and summarize my work habits and preferences: how I trade off speed against quality, what you must ask me before doing, what you shouldn't decide on your own, and what I often correct you on. Keep it to work only; don't speculate about my personality. First strip any keys, accounts and other people's information from the history. Give the evidence for each point and mark whether it appeared "multiple times" or "only once". Show me the draft first, and once I approve it, write it to my personal memory file (for Claude Code, ~/.claude/CLAUDE.md), not into the project repo. Stop after writing the draft.

Look back over this session for approaches I corrected you on or explicitly approved, and turn them into a few short, actionable rules, each with a one-line reason. Don't repeat anything that's already recorded. Show me the draft first, and once I approve it, write it to the memory file. Stop after you've drafted these.

Do a read-only analysis of this project and write short notes for an agent: how to install, build, test and run it, where the code conventions differ from the usual, and the pitfalls that are easy to hit. Include only what you can't tell from the code and what's useful every time, and keep it to one screen. If the project already has a notes file, suggest edits to that file instead of starting a second one. Show me the draft first, and write it once I approve. Stop after you've written the draft.