Charlie Hills 🦩's Post

Charlie Hills 🦩 · article · 2026-04-16 source ↗ Reader 🖥️

Agent Skills Agentic Engineering Capability Provisioning (right-sizing to the task)Claude Code Agent Context Files (CLAUDE.md / AGENTS.md)Context Rot & Context Limits Guardrails Multi-Agent Orchestration Prompting as Communication

Effective use of Claude Code requires deliberate setup, disciplined workflow habits, context management, precise prompting, and exploitation of power features — and learning these on day one would save weeks of friction.

practical · expository — A listicle of 32 practical tips for using Claude Code, organized into five thematic sections. No argumentation or narrative — pure prescriptive advice.

Structure

Part	Role
Introduction (experience claim + framing)	Establishes authority (47 days daily use) and sets up the listicle premise
Section 1: Setup (5 tips)	Foundation layer — configure CLAUDE.md, commands, skills, integrations, updates
Section 2: Workflow (10 tips)	Process discipline — git hygiene, scope limits, planning, review cadence
Section 3: Context management (6 tips)	Session hygiene — compacting, model selection, task isolation
Section 4: Prompting (6 tips)	Input quality — specificity, error-pasting, step-by-step thinking
Section 5: Power moves (5 tips)	Advanced techniques — subagents, hooks, voice, fast mode
Closing (links + engagement ask)	Social media engagement pattern; pointers to deeper resources

Key terms

CLAUDE.md Agent Context Files (CLAUDE.md / AGENTS.md)
A project-level configuration file that sets persistent rules and context for every Claude Code session
/commands
User-defined reusable shortcuts within Claude Code (slash commands)
/skills Agent Skills
Specialist agent configurations within Claude Code, each scoped to a particular job
scope limit Guardrails
Restricting Claude Code's file access to only explicitly named files to prevent unwanted changes
/compact Context Rot & Context Limits
A Claude Code command that summarizes and compresses the conversation context to free up token budget
subagents Multi-Agent Orchestration
Parallel Claude Code agent instances that can handle separate tasks concurrently

Propositions

explicitcorroboratedCLAUDE.md sets persistent rules that apply across every Claude Code session in a project.conf: high
supports ←Claude Code: Best practices for agentic coding · p2 — This guide is the canonical source for CLAUDE.md as an auto-loaded file carrying persistent per-session, per-project rules.
explicitMost people lose time due to poor workflow discipline, not poor prompting.conf: high
explicitYou should commit your code before letting Claude make changes, to preserve a known-good state.conf: high
implicitcorroboratedLong Claude Code sessions degrade output quality because context becomes polluted.conf: medium
refines ←[FULL WORKSHOP] AI Coding For Real Engineers - Matt Pocock, AI Hero · p2 — Pocock supplies the mechanism behind the long-session tip: attention degrades quadratically with context length, collapsing around ~100k tokens.
supports ←Claude Code: Best practices for agentic coding · p12 — This guide is the canonical source of the claim that long sessions degrade output because context fills with irrelevant content, and prescribes /clear as the remedy.
explicitdevelopedUsing Opus for planning/thinking and Sonnet for implementation is the optimal model allocation strategy.conf: medium
refines ←[FULL WORKSHOP] AI Coding For Real Engineers - Matt Pocock, AI Hero · p26 — Agrees on asymmetric model allocation but assigns the strong model to review rather than planning, with the cheap model implementing.
extends ←[FULL WORKSHOP] AI Coding For Real Engineers - Matt Pocock, AI Hero · p26 — Pocock generalizes Hills's Opus-plans/Sonnet-builds heuristic into a principle: assign the stronger model wherever judgment concentrates, including review.
extends ←GitHub - musistudio/claude-code-router: Use Claude Code as the foundation for coding infrastructure, allowing you to decide how to interact with the model while enjoying updates from Anthropic. · p9 — Generalizes and automates the manual 'Opus for planning, Sonnet for implementation' allocation into config-driven per-scenario routing across arbitrary providers, plus /model, custom router, and subagent pinning.
explicitDescribing the problem rather than prescribing a solution produces better Claude Code output.conf: high
explicitPasting raw error output is more effective than describing errors in natural language.conf: high
explicitReviewing every diff line before approval is necessary to maintain code quality when using Claude Code.conf: high
implicitcorroboratedRunning two subagents in parallel can double throughput on independent tasks.conf: medium
supports ←Android Studio Quail 2 Redesigns Agent Mode, Streamlines AI-Assisted Coding · p1 — Android Studio productizing parallel independent-task agent conversations corroborates the claim that parallelizing independent agent work increases throughput.
explicitLetting Claude interview you before writing code produces better results than immediately briefing it.conf: medium

Arguments

You should adopt these 32 practices to be effective with Claude Code from day one. strength: low

Explicit premises:

The author has used Claude Code daily for 47 days.
These are things he wishes he knew on day 1.

Implicit premises:

47 days of daily use constitutes sufficient experience to generalize.
What works for one user's workflow generalizes to other users.
The cost of not knowing these tips is significant wasted time.

Inferences:

Authority from personal experience transfers to prescriptive advice.

Workflow discipline matters more than prompting technique. strength: medium