Prompting & Context Compared

# ── CLAUDE CODE ──────────────────────────────────────────────
# No @ syntax — describe files naturally in your prompt
claude "Look at src/auth.js and explain the session management"

# In interactive mode, pin a file so Claude always reads it:
#   /add src/auth.js
# Claude will then reference it in every subsequent response.

# ── GEMINI CLI ───────────────────────────────────────────────
# @ prefix for a single file
gemini "@src/auth.js explain the session management"

# @ prefix for an entire directory (uses 1M context window)
gemini "@src/ summarize every module and its responsibilities"

# @ prefix for a live URL — loads an API spec and generates client code
gemini "@https://api.stripe.com/openapi.json generate a Python client"

# Combine multiple @ references in one prompt
gemini "@src/auth.js @tests/auth.test.js what's not covered by tests?"

# ── GITHUB COPILOT CLI ───────────────────────────────────────
# No file @ syntax — translates requests to shell commands
gh copilot suggest "show the content of src/auth.js and summarize it"
# Result: suggests `cat src/auth.js` — not an AI analysis of the file

# ── CLAUDE CODE ──────────────────────────────────────────────
# Session context is maintained automatically.
# Subagent example: parallel tasks within one session
claude "Write the implementation of the payment retry logic
        AND write tests for it at the same time."
# Claude spins up a subagent for tests, main agent for impl,
# both report back — you see both files when done.

# Reset session when starting a new task:
#   /clear
# CLAUDE.md is still loaded — only the conversation resets.

# ── GEMINI CLI ───────────────────────────────────────────────
# Save a named session to resume later
gemini
# (interactive mode)
# > Analyze the payment service architecture
# > /chat save payment-analysis

# Next day — resume exactly where you left off
gemini
# > /chat resume payment-analysis
# > Now add the retry logic we discussed

# Add a persistent memory note (survives session restarts)
# > /memory add "This project uses pydantic v2 exclusively"

# ── GITHUB COPILOT CLI ───────────────────────────────────────
# Each call is completely stateless
gh copilot suggest "list files in src directory"
# Returns: ls src/ or Get-ChildItem src/

# Next call has NO memory of the previous one
gh copilot suggest "now show me the auth module"
# Returns: cat src/auth.js  (no "now" context — clean slate)

# ── CLAUDE CODE: reasoning before code ───────────────────────
claude "Explain your reasoning about the best approach, then write
        the retry middleware for the Stripe webhook handler.
        Keep the existing error logging pattern unchanged."

# ── CLAUDE CODE: tests first ──────────────────────────────────
claude "Write tests for the payment retry logic BEFORE writing the
        implementation. Make the tests fail first, then fix them."

# ── GEMINI CLI: aggressive @ loading ─────────────────────────
gemini "@src/payments/ @src/webhooks/ @migrations/ \
  What's missing from our error-handling strategy?"

# ── GEMINI CLI: /plan before executing ───────────────────────
gemini
# > /plan refactor the payment module to add idempotency keys
# Review the plan, ask questions, THEN approve execution

# ── GEMINI CLI: model selection ──────────────────────────────
gemini --model gemini-2.5-pro \
  "@src/ Design the retry architecture for webhook failures"

gemini --model gemini-2.5-flash \
  "What flag does grep use for case-insensitive search?"

# ── COPILOT CLI: command-type flag ────────────────────────────
gh copilot suggest -t git "undo last commit but keep changes staged"
gh copilot suggest -t gh "create a PR from current branch to main"
gh copilot suggest -t shell "find all .env files recursively"

# ── COPILOT CLI: explain a command ───────────────────────────
gh copilot explain "git rebase -i HEAD~3"