Journal de développement

Notes de terrain sur les agents de code, le choix des modèles, le calcul des tokens et les rouages des API.

Rien dans cette langue pour le moment. Voici les entrées en anglais.

• a-monthly-ai-budget-your-small-team-can-actually-run-54.mdA Monthly AI Budget Your Small Team Can Actually Run token-math/ · EN · 2026-10-25

  Set a team AI allowance, track real token charges, and establish clear spending checkpoints with copyable terminal commands.

• reading-a-request-bill-input-output-store-hit-56.mdReading a Request Bill: Input, Output, Store, Hit token-math/ · EN · 2026-10-25

  Audit four token charges from your terminal, calculate request costs, and compare a cache write with a later cache hit.

• debug-ui-screenshots-with-gemini-4-argon-51.mdDebug UI Screenshots with Gemini 4 Argon model-bench/ · EN · 2026-10-25

  Capture a broken UI, send its screenshot and source through Fluxo, then verify the fix with repeatable terminal commands.

• which-model-handles-rust-borrow-checker-errors-best-49.mdWhich Model Handles Rust Borrow-Checker Errors Best? model-bench/ · EN · 2026-10-25

  Compare two models on the same Rust borrow error using reproducible commands, compiler checks, behavioral tests, and a practical scorecard.

• deepseek-v4-flash-for-bulk-refactoring-chores-46.mdDeepSeek V4 Flash for Bulk Refactoring Chores model-bench/ · EN · 2026-10-25

  A terminal-first workflow for batching small refactors through Fluxo, generating reviewable patches, and testing before anything lands.

• translate-developer-docs-with-qwen37-plus-47.mdTranslate Developer Docs with Qwen3.7 Plus model-bench/ · EN · 2026-10-25

  A terminal-first walkthrough for translating Markdown documentation with Qwen3.7 Plus through Fluxo, with exact commands and review checks.

• claude-opus-5-for-security-sensitive-code-review-45.mdClaude Opus 5 for Security-Sensitive Code Review model-bench/ · EN · 2026-10-25

  Review a Git diff with Claude Opus 5 through Fluxo, using exact terminal commands, constrained prompts, and reproducible verification.

• which-model-writes-the-clearest-commit-messages-44.mdWhich Model Writes the Clearest Commit Messages? model-bench/ · EN · 2026-10-25

  Compare two models against the same staged diff with exact terminal commands, a fixed prompt, and a practical commit-message scorecard.

• fable-51-technical-docs-from-the-terminal-43.mdFable 5.1: Technical Docs from the Terminal model-bench/ · EN · 2026-10-25

  Build a source-grounded documentation workflow with Claude Fable 5.1, Fluxo, Python, and repeatable terminal commands.

• gemini-31-pro-for-large-system-architecture-reviews-42.mdGemini 3.1 Pro for Large-System Architecture Reviews model-bench/ · EN · 2026-10-25

  A practical terminal walkthrough for reviewing large-system architecture with Gemini 3.1 Pro through Fluxo and OpenAI-compatible commands.

• picking-models-for-sql-generation-and-query-review-38.mdPicking Models for SQL Generation and Query Review model-bench/ · EN · 2026-10-25

  A terminal-first walkthrough for comparing SQL generation and review models against a reproducible PostgreSQL fixture.

• comparing-reasoning-models-on-dependency-upgrades-40.mdComparing Reasoning Models on Dependency Upgrades model-bench/ · EN · 2026-10-25

  Compare two reasoning models on the same dependency snapshot, then test their upgrade plans in isolated Git worktrees.

• context-windows-build-a-whole-repo-prompt-that-fits-39.mdContext Windows: Build a Whole-Repo Prompt That Fits model-bench/ · EN · 2026-10-25

  Build a repository snapshot, estimate its token budget, and send a focused prompt with exact terminal commands.

• grok-47-research-questions-from-your-terminal-35.mdGrok 4.7 Research Questions From Your Terminal model-bench/ · EN · 2026-10-25

  A terminal-first walkthrough for asking Grok 4.7 focused coding questions through Fluxo, saving answers, and checking claims.

• glm-53-for-bilingual-product-copy-and-ui-strings-34.mdGLM-5.3 for Bilingual Product Copy and UI Strings model-bench/ · EN · 2026-10-25

  Build an English–Chinese UI copy workflow with GLM-5.3, exact terminal commands, placeholder checks, and reviewable JSON output.

• deepseek-v4-pro-backend-code-from-the-terminal-31.mdDeepSeek V4 Pro: Backend Code from the Terminal model-bench/ · EN · 2026-10-25

  Generate a small Python backend through Fluxo, inspect the output, and test its HTTP contract with copyable terminal commands.

• qwen38-max-for-budget-minded-internal-tools-33.mdQwen3.8 Max for Budget-Minded Internal Tools model-bench/ · EN · 2026-10-25

  Build a terminal-first ticket triager with Qwen3.8 Max through Fluxo, then measure token usage before expanding the rollout.

• gemini-38-flash-for-long-logs-and-large-diffs-30.mdGemini 3.8 Flash for Long Logs and Large Diffs model-bench/ · EN · 2026-10-25

  A terminal-first walkthrough for reviewing long logs and large diffs through Fluxo, with reproducible commands and evidence-focused prompts.

• kimi-k3-for-long-chinese-english-technical-docs-32.mdKimi K3 for Long Chinese-English Technical Docs model-bench/ · EN · 2026-10-25

  A practical terminal-first workflow for translating, extracting, and reviewing long Chinese-English technical documents with Kimi K3 through Fluxo.

• gpt-61-sol-for-everyday-scripting-28.mdGPT-6.1 Sol for Everyday Scripting model-bench/ · EN · 2026-10-25

  A terminal-first walkthrough for drafting, reviewing, and testing small Python utilities with GPT-6.1 Sol through Fluxo.

• hand-off-an-unfinished-agent-task-without-losing-state-25.mdHand Off an Unfinished Agent Task Without Losing State agent-loops/ · EN · 2026-10-25

  A terminal-first walkthrough for saving unfinished code, recording test failures, and giving the next coding agent a reliable restart point.

• make-your-coding-agent-respect-formatter-and-lint-rules-23.mdMake Your Coding Agent Respect Formatter and Lint Rules agent-loops/ · EN · 2026-10-25

  A terminal-first walkthrough for giving coding agents explicit style rules, reproducible checks, and a lint gate they cannot quietly bypass.

• review-claude-code-changes-before-you-merge-24.mdReview Claude Code Changes Before You Merge agent-loops/ · EN · 2026-10-25

  A terminal-first checklist for reviewing Claude Code patches, testing failure paths, inspecting dependencies, and approving only verified changes.

• safe-database-migrations-with-an-ai-pair-programmer-22.mdSafe Database Migrations with an AI Pair Programmer agent-loops/ · EN · 2026-10-25

  A terminal-first PostgreSQL walkthrough: review AI-generated SQL, rehearse changes, backfill in batches, and keep rollback boring.

• refactoring-2000-lines-with-an-agent-one-slice-at-a-time-20.mdRefactoring 2,000 Lines with an Agent, One Slice at a Time agent-loops/ · EN · 2026-10-25

  A practical Codex walkthrough: isolate a large TypeScript file, capture behavior, extract small modules, and verify each commit.

• gemini-cli-meets-a-local-cleanup-notebook-17.mdGemini CLI Meets a Local Cleanup Notebook agent-loops/ · EN · 2026-10-25

  Clean a messy CSV with Gemini CLI and Jupyter, verify every transformation, and use Fluxo for an optional model review.

• vs-code-claude-code-without-file-collisions-21.mdVS Code + Claude Code, Without File Collisions agent-loops/ · EN · 2026-10-24

  A practical Git worktree walkthrough for running VS Code and Claude Code in parallel, reviewing changes, and merging without file collisions.

• let-your-coding-assistant-draft-the-pr-description-18.mdLet Your Coding Assistant Draft the PR Description agent-loops/ · EN · 2026-10-24

  A terminal-first workflow for turning a branch diff and test results into a reviewed, draft pull request description through Fluxo.

• give-codex-a-map-of-your-monorepo-19.mdGive Codex a Map of Your Monorepo agent-loops/ · EN · 2026-10-24

  Map a pnpm monorepo for Codex with scoped AGENTS.md files, exact shell commands, focused prompts, and package-level verification.

• debugging-flaky-ci-with-claude-code-and-test-logs-16.mdDebugging Flaky CI with Claude Code and Test Logs agent-loops/ · EN · 2026-10-24

  Download failing CI logs, reproduce a flaky Vitest test, and use Claude Code to build an evidence-backed fix.

• plan-your-coding-day-before-the-first-edit-15.mdPlan Your Coding Day Before the First Edit agent-loops/ · EN · 2026-10-24

  A terminal-first morning routine for turning repository context into a focused daily plan with a coding agent.

• claude-code-hooks-lint-after-every-edit-14.mdClaude Code Hooks: Lint After Every Edit agent-loops/ · EN · 2026-10-24

  Wire Claude Code’s PostToolUse hook to ESLint, return failures to the agent, and test the loop with copyable terminal commands.

• when-to-interrupt-a-coding-agent-8.mdWhen to Interrupt a Coding Agent agent-loops/ · EN · 2026-10-24

  A practical terminal walkthrough for deciding when to interrupt an agent and when to let it finish safely.

• two-coding-agents-two-branches-one-worktree-11.mdTwo Coding Agents, Two Branches, One Worktree agent-loops/ · EN · 2026-10-24

  A practical command-by-command workflow for running two coding agents in parallel on separate Git branches.

• teach-claude-code-your-teams-commit-conventions-7.mdTeach Claude Code Your Team’s Commit Conventions agent-loops/ · EN · 2026-10-24

  Turn your team’s commit rules into repository instructions and a reusable Claude Code skill, then test the workflow with exact commands.

• using-gemini-cli-for-quick-repository-archaeology-5.mdUsing Gemini CLI for Quick Repository Archaeology agent-loops/ · EN · 2026-10-24

  A terminal-first walkthrough for mapping an unfamiliar repository with Gemini CLI, exact prompts, and a repeatable verification loop.

• review-an-agents-diff-without-losing-an-afternoon-4.mdReview an Agent’s Diff Without Losing an Afternoon agent-loops/ · EN · 2026-10-24

  A command-by-command workflow for checking scope, tracing risky changes, and testing a coding agent’s diff without derailing your day.

• pair-programming-with-an-agent-on-legacy-php-6.mdPair Programming With an Agent on Legacy PHP agent-loops/ · EN · 2026-10-24

  Build a failing PHP regression test, ask an agent for a minimal patch, and verify the fix from your terminal.

• let-the-coding-agent-write-the-tests-first-3.mdLet the Coding Agent Write the Tests First agent-loops/ · EN · 2026-10-24

  A terminal-first walkthrough for making a coding agent prove the contract before implementing a small Python function.

• run-codex-in-a-worktree-keep-main-clean-2.mdRun Codex in a Worktree, Keep Main Clean agent-loops/ · EN · 2026-10-24

  Create an isolated Git worktree, configure Codex through Fluxo, review the experiment, and remove it without disturbing main.

• keep-claude-code-on-one-task-per-session-0.mdKeep Claude Code on One Task per Session agent-loops/ · EN · 2026-10-24

  A terminal-first walkthrough for scoping Claude Code sessions, running focused tests, and stopping before one fix becomes a rewrite.

Inscrivez-vous en une minute et recevez du crédit pour essayer tous les modèles cités ici. S'inscrire →