Summary of "How Claude Code’s Creator Starts EVERY Project"

How Boris Cherney (creator of Claude Code) starts every project

Boris organizes every project around six repeatable practices designed to reduce assumptions, speed iteration, and build systems that scale.

Start ~80% of sessions in “plan mode” (enter in Claude Code by pressing Shift+Tab twice).
Use an interview-first approach: have Claude ask clarifying questions (core problems, users, success metrics, constraints) and lock a plan before building.
Rationale: avoids vague prompts, wrong assumptions, and time spent debugging solutions that solved the wrong problem (example: Claude changed a DB value instead of fixing a display bug).

A project- or user-specific cheat sheet Claude reads at chat start to stay on your conventions.
Keep it minimal: add rules only when repeated errors occur. If it becomes bloated, delete and start fresh or run a “trim” command to remove obsolete/contradictory content.
Example maintenance prompt: “Based on this conversation, can you update Claude.md so this doesn’t happen again?”

“If Claude.md becomes bloated, delete and start fresh.” — Boris’ counterintuitive advice to avoid ever-growing, contradictory instructions.

Give Claude tools to see outputs (browser, test harness, reports) and explicitly tell it about those tools.
Ask Claude to create and run a verification plan before building; periodically request it to “go back and verify all of your work.”
Boris’ two-step rule:
1. Give Claude access to the output.
2. Tell Claude about that tool.

Run multiple, partitioned Claude sessions in parallel, each focused on a non-overlapping task.
Benefits: fresh context windows see things an entrenched session can miss; avoids sessions working on the same files/context and creating conflicts.

Identify repeated daily tasks (“inner loops”) and convert them to slash commands or Claude skills (documented, callable processes).
Claude skills are repeatable plays you can invoke consistently (e.g., generate a standardized report with variable data).
Discovery prompt: “Based on the project I’m working on, what Claude skills should I create?”

Don’t over-optimize prompts or add brittle scaffolding—models improve quickly; general models usually outperform hand-tuned specific workarounds (the “bitter lesson” mindset).
Focus effort on information mode: the context, data, and systems that compound model improvements over time rather than micro-optimizing prompts.

Useful example prompts:

Planning interview: Before we start building, interview me about this. What are the core problems this solves? Who is this for? What does success look like? What should this not do? Summarize it back to me before we write any code.
Claude.md maintenance: Update my Claude.md to remove anything that's no longer needed, contradictory, duplicate information, or unnecessary bloat impacting effectiveness.
Verification: Please go back and verify all of your work so far. Make sure you use best practices, were efficient, and didn't introduce any issues.
Skills discovery: Based on the project I’m working on, what Claude skills should I create?

The main video is a how-to/guide outlining Boris’ workflows and giving prompts and practical commands.
There is a separate deep-dive video on Claude skills (tutorial on creating/using skills and slash commands) referenced by the creator.

Real-world bug example: Claude fixed a symptom by altering DB values, creating more breakage—underscores the need for planning and verification.
Boris’ counterintuitive advice: delete a bloated Claude.md and start fresh rather than continually piling on instructions.
Emphasis on systems (information mode) over tiny prompt tweaks because models keep improving.

Boris Cherney (creator of Claude Code) — interviews, tweets, and public posts quoted.
Video narrator / creator — a researcher/host who compiled Boris’ public guidance and added personal examples and implementation tips.