# claude-oss-skills

A set of practical Claude Code runbooks for developing public open-source projects on resource-constrained hardware. Eight manually-invoked skills cover the workflow from an empty directory to a deployed release, and two shell files help keep execution cost low: one best-effort `PreToolUse` command guard and one output-filter utility. Skills regulate code quality and execution cost; the command guard can block recognized Bash commands, and the output filter reduces command output

## Purpose

Discourage generated-code clutter, unnecessary abstraction, excessive comments, speculative architecture, heavy validation commands, and misleading public documentation - while keeping CPU, RAM, disk, battery, and thermal use low

## Skills and hooks

Two kinds of thing live here, and they operate at different layers:

- Workflow skills are procedural runbooks you invoke by command (`/oss-plan`, `/implement-minimal`, and so on). Each sets `disable-model-invocation: true`, so Claude never runs one on its own; you choose when to enter the runbook - The two shell files act per command, not per workflow. One is a best-effort `PreToolUse` command guard (`block-expensive-command.sh`) that can block recognized commands but is not a sandbox or a complete shell-security boundary. The other is a standalone output-filter utility (`filter-command-output.sh`) rather than a native Claude Code hook. Neither component guarantees complete command or output safety

## Design principles

- Small files, predictable structure, concise language - Progressive disclosure: each `SKILL.md` routes; stack-specific commands, recovery recipes, and report formats live in `references/` and `templates/` - Conservative shell scripting with no third-party dependencies - Explicit limitations and truthful documentation - Shared constraints live in one file; skills reference it instead of repeating it

## Directory structure

```text claude-oss-skills/ ├── README.md ├── LICENS