GitHub Copilot Prompts

Use agent mode to implement [FEATURE] in this repository. Context: the feature should [BEHAVIOR]; it touches [AFFECTED_AREAS, e.g. the API layer, the auth module, and the frontend settings page]; and it must follow the existing patterns in [REFERENCE_FILES]. Plan: (1) explore the relevant files and summarize your implementation plan before changing anything, (2) make the edits across all needed files, (3) add or update tests in [TEST_DIR], (4) run the test suite and fix failures. Constraints: do not modify [OFF_LIMITS_FILES]; keep functions under [LINE_LIMIT] lines; match the code style in [STYLE_GUIDE]. Report each file you changed and why.

Review this pull request as a senior engineer on our team. PR context: [PR_DESCRIPTION]; files changed: [LIST_OR_LINK]. Check for: correctness and edge cases, security issues (injection, auth, secrets), performance regressions, consistency with the patterns in [REFERENCE_FILES], and test coverage of the new logic. For each issue found, cite the file and line, explain why it matters, and suggest a concrete fix. Separate must-fix items from nice-to-have nits. End with an approve or request-changes verdict and a one-paragraph rationale. Be direct; do not praise boilerplate.

Write a commit message for these staged changes: [DESCRIBE_CHANGES_OR_PASTE_DIFF]. Follow our convention: [CONVENTION, e.g. Conventional Commits with scope]. The subject line must be under [CHAR_LIMIT] characters and describe the why, not just the what. Body: [NUMBER] bullets explaining the reasoning and any trade-offs, plus a breaking-change note if applicable. If the diff mixes unrelated changes, split it into separate commit messages instead of one. Do not invent ticket numbers; use [TICKET_ID] only if I provided it.

Help me debug this issue in a chat session. Symptom: [SYMPTOM]. Expected behavior: [EXPECTED]. Relevant code: [PASTE_CODE_OR_FILE_PATHS]. Environment: [ENV_DETAILS, e.g. Node 20, Postgres 15, the failing test name]. Steps I already tried: [TRIED]. Work like a debugging partner: first restate your hypothesis of the root cause, then suggest the single most informative next check (a log line, a breakpoint, a query), and only then propose a fix. Do not rewrite large blocks of code until we agree on the cause. If you need more info, ask exactly [NUMBER] targeted questions.

Turn this task into an implementation plan I can execute in Copilot Workspace: [TASK_DESCRIPTION]. Repository context: [REPO_CONTEXT]. The plan must: break the work into [NUMBER] ordered steps, name the files each step touches, define done criteria per step (tests passing, behavior verified), and flag steps that need human review (migrations, public API changes, security-sensitive code). Identify anything ambiguous in the task and propose a default decision for each ambiguity so I can approve or override. Keep the plan realistic for [TIME_BUDGET].

Act as my GitHub Copilot CLI assistant. I need a [SHELL] command that [GOAL, e.g. finds all TODO comments added in the last 30 days, lists merged branches excluding main]. Constraints: it must be safe to run (no destructive action without a dry-run preview), work on [OS], and handle filenames with spaces. Give me the exact command, explain each flag in one line, show what the output looks like on a sample repo, and provide the dry-run version first. If there are multiple ways to do it, recommend one and say why.

Refactor [FILE_OR_MODULE] using agent mode. Current problems: [PROBLEMS, e.g. oversized functions, duplicated validation, no tests]. Goals: extract well-named functions or modules, remove duplication, add type annotations, and keep every existing behavior identical (no feature changes). Rules: run the existing tests via [TEST_COMMAND] before and after; any behavior change must be flagged, not silently made; follow the style in [STYLE_REFERENCE]. Deliver a summary of what moved where, the before/after line counts, and any risky spots you want a human to double-check.

Generate tests for [FILE_OR_FUNCTION] in [TEST_FRAMEWORK]. Cover: happy paths, edge cases ([EDGE_CASES]), and error handling for [FAILURE_MODES]. Each test needs a descriptive name stating the behavior under test, not a generic label. Mock [EXTERNAL_DEPENDENCIES] following the pattern in [EXISTING_TEST_FILE]. Aim for meaningful assertions, not coverage theater: every test must fail if the behavior breaks. After generating, run [TEST_COMMAND]; fix failures yourself and report which tests you adjusted and why.

Write the pull request description for this branch: [BRANCH_SUMMARY_OR_DIFF]. Audience: [REVIEWERS]. Structure: a 2-sentence summary, 'What changed' bullets grouped by area, 'How to test' steps a reviewer can follow in under [MINUTES] minutes, a note on screenshots or recordings for [UI_CHANGES], and a checklist (tests added, docs updated, no secrets committed). Link related issue [ISSUE_NUMBER] with the right keyword (Closes/Fixes). Tone: factual and scannable; a reviewer should understand the PR in 60 seconds.

I am new to this repository: [REPO_NAME_AND_STACK]. Act as my onboarding guide. Explain: the overall architecture in a few paragraphs, what each top-level directory does, where [FEATURE_AREA, e.g. authentication, billing] lives and how a request flows through it, and the [NUMBER] files I must read first to be productive. Point out non-obvious conventions in [CONFIG_OR_DOCS]. End with [NUMBER] small, safe first tasks (docs fix, tiny bug, test addition) ordered by difficulty. Ask me [NUMBER] questions about my role so you can tailor the next round.