Claude Code
Tool, coding agent, by Anthropic
Assessment
The way everything on this system gets built, with every session routed through a local proxy so the engine can be a home model or the Claude API. The skills ecosystem (SKILL.md files, portable to other agents) is where the leverage is going in 2026.
2026-10-09
Strengths
- [object Object]
- Project memory through CLAUDE.md files and a persistent memory directory
- Skills (SKILL.md), hooks, slash commands and MCP servers for project-specific behaviour
Limitations
- Context is finite; long sessions get summarized and details can drop
- A small local model behind the proxy loses details like SSH keys between tool calls
Claude Code is Anthropic's terminal coding agent: it reads a repository, edits files, runs commands and tests, and commits, with the model's behaviour steered by CLAUDE.md files at the user, directory and project level. It runs as a CLI, in IDE extensions, as a desktop app and on the web.[1]
Skills
A skill is a SKILL.md file, a playbook the agent loads for a class of task, either when invoked as a slash command or when the task matches. The format has become portable: the same files work in Cursor, Gemini CLI, Codex CLI and others. Official Anthropic skills install with npx skills add anthropics/claude-code --skill <name>;[2] community libraries bundle more than a thousand. The practices here list the ones worth knowing.
Running it against a local model
An OpenAI-compatible proxy in front of llama-server lets Claude Code use a home model as its engine; the environment variable ANTHROPIC_BASE_URL points the CLI at the proxy. Small local models lose working details between tool calls, so a system that does this injects the facts they need (hosts, keys, paths) into every request and gives them helper scripts instead of raw commands.
History
- 2026-10-09: record seeded.
Comments
Public comments on each entry are coming. Nothing is collected here yet.