A new teammate clones the repo but never gets your conventions. Why, and the fix?
Space to flipThe instructions live in user-level ~/.claude/CLAUDE.md, which is not shared via version control. Move them into project-level config (.claude/CLAUDE.md or root CLAUDE.md).
What are the three levels of the CLAUDE.md hierarchy?
Space to flip~/.claude/CLAUDE.md (personal).claude/CLAUDE.md or root CLAUDE.md (shared)CLAUDE.md inside a subdirectoryTwo ways to keep CLAUDE.md modular instead of one monolithic file?
Space to flipUse @import (@.claude/rules/testing.md) so each package pulls only relevant standards; or split into topic files under .claude/rules/ — testing.md, api-conventions.md, deployment.md.
Behavior is inconsistent across sessions — which command diagnoses it?
Space to flip/memory — it shows which memory (CLAUDE.md) files are currently loaded, so you can spot a missing or unexpected one.
Where do you create a /review command so every developer gets it on clone?
.claude/commands/ — project-scoped, shared via version control. Not ~/.claude/commands/ (personal), not CLAUDE.md, and there is no .claude/config.json commands array.
Where do skills live, and what configures their behavior?
Space to flip.claude/skills/ with a SKILL.md file whose frontmatter supports context: fork, allowed-tools, and argument-hint.
What does context: fork do in skill frontmatter?
Runs the skill in an isolated sub-agent context, so verbose output (codebase analysis) or exploratory context (brainstorming) doesn't pollute the main conversation.
← → navigateWhat do allowed-tools and argument-hint do?
allowed-tools restricts tool access during skill execution (e.g. limit to file writes to prevent destructive actions). argument-hint prompts the developer for required parameters when the skill is invoked without arguments.
When choose a skill vs CLAUDE.md — and where do personal skill variants go?
Space to flipSkill = on-demand, task-specific workflow; CLAUDE.md = always-loaded universal standards. Create personal variants in ~/.claude/skills/ under a different name to avoid affecting teammates.
Test files are scattered across directories — how do you apply one convention automatically?
Space to flipA .claude/rules/ file with YAML frontmatter paths: glob — e.g. paths: ["**/*.test.tsx"] — which loads only when Claude edits a matching file.
What's the benefit of path-scoped rules' conditional loading?
Space to flipRules load only when editing matching files, so irrelevant context stays out and token usage drops — only the conventions relevant to the current file are present.
← → navigateWhen choose path-specific rules over a subdirectory CLAUDE.md?
Space to flipWhen a convention must apply to files of a type spread across many directories (**/*.test.tsx). A subdirectory CLAUDE.md is directory-bound and can't follow scattered files.
Trap: why not use root-CLAUDE.md headers or per-type skills for path conventions?
Space to flipRoot headers rely on Claude inferring which section applies (not explicit matching); skills require manual invocation (not automatic path matching). Path-scoped rules match explicitly and load automatically.
← → navigateWhen do you use plan mode?
Space to flipFor complex tasks — large-scale changes, multiple valid approaches, architectural decisions, multi-file modifications (e.g. a microservice restructuring or a library migration affecting 45+ files).
← → navigateWhen is direct execution appropriate?
Space to flipFor simple, well-scoped changes — a single validation check, a one-file bug fix with a clear stack trace, adding a date-validation conditional.
← → navigateWhy enter plan mode first for an architectural task — not "only if complexity emerges"?
Space to flipPlan mode enables safe exploration and design before committing to changes, preventing costly rework. For architectural work the complexity is already known from the requirements.
← → navigateWhat is the Explore subagent for, and how does plan mode combine with direct execution?
Space to flipThe Explore subagent isolates verbose discovery output and returns summaries, preserving the main context window. Combine the two: plan mode to investigate, then direct execution to implement the planned approach.
← → navigateProse descriptions get interpreted inconsistently — most effective fix?
Space to flipProvide 2–3 concrete input/output examples of the transformation — examples communicate expected behavior more reliably than prose.
← → navigateWhat is test-driven iteration?
Space to flipWrite a test suite (behavior, edge cases, performance) first, then iterate by sharing failing tests to guide progressive improvement. Give specific input/expected-output cases for edge cases (e.g. null values in migration scripts).
← → navigateWhat is the interview pattern, and when use it?
Space to flipHave Claude ask questions to surface considerations you may not have anticipated (cache-invalidation strategy, failure modes) before implementing — especially valuable in unfamiliar domains.
← → navigateWhen provide all issues in one message vs sequentially?
Space to flipOne detailed message when the issues interact (fixes affect each other); fix them sequentially when the issues are independent.
← → navigateYour CI job running claude "…" hangs on interactive input. Fix?
Add the -p (--print) flag to run non-interactively. (Wrong answers: CLAUDE_HEADLESS=true, --batch, and redirecting stdin from /dev/null — none of these is the correct mechanism.)
How do you get machine-parseable findings to post as inline PR comments?
Space to flipUse --output-format json together with --json-schema to enforce structured output that automation can parse and post.
How do you give CI-invoked Claude Code project context, and avoid duplicate tests?
Space to flipDocument testing standards, fixture conventions, and review criteria in CLAUDE.md. Provide existing test files in context so generated tests don't duplicate scenarios already covered.
Re-running a review after new commits — how do you avoid duplicate comments?
Space to flipInclude the prior review findings in context and instruct Claude to report only new or still-unaddressed issues.
← → navigateWhy not reuse the same session that generated the code to review it?
Space to flipSession context isolation: the session that generated the code is less effective at reviewing its own changes. Use an independent review instance.
← → navigateKeyboard: ← → navigate · Space flip · S shuffle · R restart · G got it · V review again