agent-config

A personal configuration repository for Claude Code — managing instructions, skills, commands, and subagents in one place, version-controlled and symlinked into ~/.claude/.

Security notice: The contents of this repository flow directly into Claude's system prompt. Review all files before use, and never commit secrets or credentials. A gitleaks pre-commit hook is included to help catch leaks before they happen.

What's in here

Directory	Purpose
CLAUDE.md	Global instructions loaded by Claude Code on every session
commands/	Custom slash commands (invoke with /command-name)
skills/	Reusable skill modules auto-discovered by Claude Code
subagents/	Specialized agents Claude can delegate work to
prompts/	Standalone prompts for specific tasks
scripts/	Setup scripts

Setup

git clone https://github.com/coreydaley/agent-config.git
cd agent-config
brew install gitleaks
make all

make all creates symlinks into ~/.claude/ and installs the gitleaks pre-commit hook. It is idempotent — safe to run multiple times.

What gets symlinked

CLAUDE.md   → ~/.claude/CLAUDE.md
commands/   → ~/.claude/commands/
skills/     → ~/.claude/skills/
subagents/  → ~/.claude/agents/

Existing files at symlink destinations are backed up with a .old extension before replacement.

Individual targets

make symlinks   # Create ~/.claude/ symlinks only
make hooks      # Install gitleaks pre-commit hook only
make help       # Show all available targets

Commands

Invoke with /command-name in Claude Code. Commands with disable-model-invocation must be called explicitly; others may also trigger automatically based on context.

Command	Description
/commit	Analyze uncommitted changes and create grouped conventional commits
/tag	Analyze commits since last tag and propose the next semantic version
/create-task	Create a new task note in the Obsidian vault
/create-knowledge	Create a new knowledge note in the Obsidian vault
/create-draft	Create a new draft note in the Obsidian vault
/create-read-later	Add a URL to the Read Later list in the Obsidian vault
/sandbox-script	Create a new standalone script in the sandbox
/sandbox-project	Create a new project directory in the sandbox
/sprint-plan	Multi-agent collaborative sprint planning
/sprint-work	Execute the next sprint from local docs
/audit-security	Dual-agent security review → findings report
/audit-design	Dual-agent UI/UX review → findings report
/audit-accessibility	Dual-agent WCAG 2.1/2.2 review → findings report
/audit-architecture	Dual-agent architecture review → findings report
/create-blog-post	AI-powered blog post creation workflow

Skills

Skills are auto-discovered from ~/.claude/skills/. Each skill has a SKILL.md with YAML frontmatter describing when it applies. Skill descriptions are always loaded into context; skill bodies load only when triggered.

Skill	Description
github	gh CLI operations — issues, PRs, releases, branches
obsidian	Obsidian vault operations via the obsidian CLI
orbstack	OrbStack management — Linux machines, Docker, Kubernetes
frontend-design	Production-grade UI component creation
ledger	Sprint ledger tracking
generate-post-image	Hugo blog post image generation
skill-creator	Guide for creating new skills

Subagents

Subagents are specialized agents Claude delegates work to via the Agent tool. Their full workflow runs in an isolated context, keeping the main conversation clean. Defined in subagents/ (symlinked to ~/.claude/agents/).

Subagent	Description
audit-security	Dual-agent security audit — 5-phase workflow (orient, independent reviews, synthesis, devil's advocate, report)
audit-accessibility	Dual-agent WCAG 2.1/2.2 audit — same 5-phase pattern with accessibility-specific finding schema
audit-architecture	Dual-agent architecture audit — findings anchored to named principles with migration cost estimates
audit-design	Dual-agent UI/UX audit — findings anchored to Nielsen heuristics and project design system

Audit output

Each audit run writes timestamped artifacts to ~/Reports/<repo-path>/:

$REPORT_TS-audit-security-claude.md          ← Claude's independent review
$REPORT_TS-audit-security-codex.md           ← Codex's independent review
$REPORT_TS-audit-security-synthesis.md       ← unified findings
$REPORT_TS-audit-security-devils-advocate.md ← Codex challenge pass
$REPORT_TS-audit-security-report.md          ← final findings report

The report is a reference document. To act on findings, run /sprint-plan and use the report as the seed.

Sprint planning output

/sprint-plan writes all artifacts to ~/Reports/<repo-path>/ (derived from pwd). Files marked * are only created when the corresponding optional phase ran:

$REPORT_TS-sprint-plan-intent.md
$REPORT_TS-sprint-plan-claude-draft.md
$REPORT_TS-sprint-plan-codex-draft.md                  * (Compete)
$REPORT_TS-sprint-plan-claude-draft-codex-critique.md  * (Compete)
$REPORT_TS-sprint-plan-codex-draft-claude-critique.md  * (Compete)
$REPORT_TS-sprint-plan-merge-notes.md                  * (Compete)
$REPORT_TS-sprint-plan-devils-advocate.md              * (Devil's Advocate)
$REPORT_TS-sprint-plan-security-review.md              * (Security Review)
$REPORT_TS-sprint-plan-architecture-review.md          * (Architecture Review)
$REPORT_TS-sprint-plan-test-strategy-review.md         * (Test Strategy Review)
$REPORT_TS-sprint-plan-SPRINT-NNN.md                   ← final (renamed after approval)

The ledger (ledger.tsv) lives in the same ~/Reports/<repo-path>/ directory. /sprint-work reads sprint documents and the ledger from there.

Disclaimer

This repository contains AI-generated content. Review all configurations and instructions before use. The creators assume no liability for problems caused by using resources from this repository. See SECURITY.md for details.