Agent Inbox

Code Harness Skills

Reusable Claude Code skills for software development workflows. Works with any project, any language.

Installation

Install the full marketplace

# In Claude Code
/plugin marketplace add 8-bit-sheep/code-harness-skills

Install individual skills

/plugin install skill-builder@8-bit-sheep/code-harness-skills
/plugin install sprint-planner@8-bit-sheep/code-harness-skills
/plugin install release-manager@8-bit-sheep/code-harness-skills
# ... etc

Local development

git clone https://github.com/8-bit-sheep/code-harness-skills
# In Claude Code:
/plugin marketplace add ./code-harness-skills

Available Skills

Meta / Tooling

Skill	Description
skill-builder	Create, test, evaluate, and optimize Claude Code skills with eval-driven iteration
headless-runner	Run Claude Code programmatically in headless mode for CI/CD, automation, and agent workflows
cloud-setup	Set up development environments in cloud/mobile Claude Code sessions

Code Quality

Skill	Description
codebase-organizer	Monitor file sizes and split large files into AI-friendly modules
test-coverage-guardian	Analyze test coverage, identify gaps, detect dead code, enforce quality gates
perf-reviewer	Review code for performance issues, run benchmarks, identify optimizations

Project Management

Skill	Description
sprint-planner	Analyze design docs, calculate velocity, create realistic sprint plans
sprint-executor	Execute sprint plans with TDD, milestone tracking, and checkpoint validation
design-doc-creator	Create structured design documents with systemic analysis
release-manager	Automate releases with version bumps, changelogs, tags, and CI verification

Collaboration

Skill	Description
github-issue-triage	Monitor and triage GitHub issues against design docs and implementation status
agent-inbox	Cross-agent messaging with inbox management and handoff patterns

Configuration

Skills use environment variables with sensible defaults. Each skill documents its configuration in a ## Configuration section within its SKILL.md.

Common variables:

Variable	Default	Used By
TEST_CMD	make test	Most skills
LINT_CMD	make lint	Most skills
BUILD_CMD	make build	Most skills
SOURCE_DIR	.	codebase-organizer
DESIGN_DOCS_DIR	design_docs/	design-doc-creator, sprint-planner, github-issue-triage
COVERAGE_TARGET	80	test-coverage-guardian
SPRINT_STATE_DIR	.sprint-state	sprint-planner, sprint-executor

Quick Start

Create your own skill

The skill-builder skill is a meta-skill that helps you create, test, and optimize new skills:

User: "Create a skill for database migration management"
Claude: [Uses skill-builder to scaffold, test, and iterate on a new skill]

Organize your codebase

User: "Check file sizes and split any large files"
Claude: [Uses codebase-organizer to find files >800 lines and propose splits]

Plan and execute a sprint

User: "Plan a sprint from the auth-redesign design doc"
Claude: [Uses sprint-planner to analyze velocity and create a milestone-based plan]

User: "Execute the sprint"
Claude: [Uses sprint-executor with TDD, running tests at each checkpoint]

Skill Structure

Each skill follows the progressive disclosure pattern:

skill-name/
  SKILL.md        # Always loaded (core docs, <300 lines)
  scripts/        # Execute as needed (automation)
  resources/      # Load on demand (detailed references)

Creating Your Own Skills

See the skill-builder for the complete guide to creating, testing, and optimizing skills.

Key principles:

• Pushy descriptions: List specific trigger scenarios to combat under-triggering
• Progressive disclosure: Keep SKILL.md lean, details in resources/
• Eval-driven: Test with the skill-builder's eval framework before shipping

Contributing

1. Fork this repository
2. Create a new skill in skills/your-skill-name/
3. Include a SKILL.md with YAML frontmatter (name + description)
4. Add scripts/ and resources/ as needed
5. Run skill-builder/scripts/validate_skill.sh skills/your-skill-name/
6. Submit a PR

License

MIT