Agent Skills

A collection of skills for coding agents to enhance AI-assisted development workflows.

Skills Included

Development Workflow

Skill	Description	Claude Code	Claude.ai
sgai-goal	Compose GOAL.md files for SGAI workspaces through interactive conversation	*
setting-up-a-project	Author CLAUDE.md and AGENTS.md with project purpose, tech stack, and development practices	*
repairing-project-setup	Audit a repo against what setting-up-a-project would generate and repair only the gaps	*
working-on-an-issue	Implement a GitHub issue hands-off, with scoped recon and a PR as the approval gate	*
working-on-an-epic	Coordinate an epic from a Conductor session: child issues run in parallel Conductor workspaces, tracked to reviewable PRs	*
supervising-conductor-workspaces	Dispatch one Conductor workspace per issue and supervise their PRs to merge: keep them current, resolve conflicts, rerun flaky CI, archive when merged	*
project-analysis	Analyze project codebase structure, architecture, key files, and dependencies	*	*
project-planning	Orchestrate end-to-end project planning with issues, diagrams, dependencies, and timelines	*	*
issue-decomposition	Decompose projects into well-structured GitHub issues with user stories and estimates	*	*
estimating-work-items	Size issues with story points on a powers-of-two scale and total epics from their children, recorded as a label, an Estimate section, and a project field	*	*
estimating-change-footprint	Estimate an issue's change footprint (lines changed and files read, each bucketed) to predict coding-agent token spend, recorded as labels, a Footprint section, and a project field	*
dependency-mapping	Generate Mermaid dependency graphs showing issue relationships and critical paths	*	*
timeline-planning	Generate Mermaid Gantt charts for project timelines with phases and milestones	*	*

Requirements & Planning

Skill	Description	Claude Code	Claude.ai
requirement-elicitation	Conversational wizard for eliciting functional, nonfunctional, and edge-case requirements	*	*
idea-to-design	Turn an idea note into a design document	*	*
writing-product-specs	Write comprehensive product specification documents	*	*
writing-user-stories	Write properly formatted user stories with Given-When-Then acceptance criteria and an INVEST quality checklist	*	*
writing-verification-plans	Acceptance test verification plans (real-world testing, never mocks)	*	*
context-aware-questions	Identify information gaps and generate questions to surface missing requirements	*	*

Project Management

Skill	Description	Claude Code	Claude.ai
at-risk-detection	Identify at-risk issues and PRs: stale items, blocked work, deadline risks, scope creep	*	*
triage-new-issues	Review new GitHub issues, assess priority, suggest labels, and recommend assignees	*	*
stakeholder-tracking	Define stakeholder personas and track their goals	*	*
stakeholder-updates	Craft clear, concise stakeholder communications	*	*
prepare-meeting-agenda	Generate time-boxed meeting agendas from recent issues, PRs, and notes, saved as dated files so action items carry over	*	*

Documentation & Communication

Skill	Description	Claude Code	Claude.ai
architecture-diagramming	Generate Mermaid architecture diagrams showing system components and data flows	*	*
mermaid-diagrams	Guide for creating syntactically correct Mermaid diagrams for GitHub	*	*
markdown-formatting	Format AI outputs into consistent, readable Markdown	*	*
build-faq-from-issues	Extract common questions from closed issues and generate an FAQ document	*	*
summarize-conversation-thread	Summarize GitHub issue/PR threads into key decisions and action items	*	*
research-topic-summarize	Research topics via web search and synthesize detailed summaries with sources	*	*

Personal & Utilities

Skill	Description	Claude Code	Claude.ai
daily-planning-ritual	Interactive daily planning across work and personal life dimensions	*	*

Installation

Claude Code: Plugin Marketplace (Recommended)

Add this repository as a plugin marketplace:

/plugin marketplace add britt/agent-skills

Install all skills:

/plugin install agent-skills@britt

Install a curated bundle:

# project-foundations: setup, issue decomposition, user stories,
# verification plans, and architecture/dependency/Mermaid diagrams
/plugin install project-foundations@britt

# llpm: product/project-management skills - requirements, specs, user
# stories, stakeholders, meetings, triage, risk, timeline/dependency planning,
# and the project-planning orchestrator (add architecture-diagramming separately)
/plugin install llpm@britt

# chat: for chat agents rather than coding agents - writing and
# communication, research, diagrams, requirements, specs, and daily planning.
# Nothing in it needs a repository, git, or the gh CLI.
/plugin install chat@britt

Install individual skills:

/plugin install architecture-diagramming@britt
/plugin install at-risk-detection@britt
/plugin install build-faq-from-issues@britt
/plugin install context-aware-questions@britt
/plugin install dependency-mapping@britt
/plugin install estimating-change-footprint@britt
/plugin install estimating-work-items@britt
/plugin install idea-to-design@britt
/plugin install issue-decomposition@britt
/plugin install markdown-formatting@britt
/plugin install mermaid-diagrams@britt
/plugin install prepare-meeting-agenda@britt
/plugin install project-analysis@britt
/plugin install project-planning@britt
/plugin install requirement-elicitation@britt
/plugin install research-topic-summarize@britt
/plugin install stakeholder-tracking@britt
/plugin install stakeholder-updates@britt
/plugin install summarize-conversation-thread@britt
/plugin install timeline-planning@britt
/plugin install triage-new-issues@britt
/plugin install working-on-an-issue@britt
/plugin install working-on-an-epic@britt
/plugin install supervising-conductor-workspaces@britt

Some skills are only available via a bundle, not standalone: daily-planning-ritual, repairing-project-setup, setting-up-a-project, sgai-goal, writing-product-specs, writing-user-stories, writing-verification-plans. Install agent-skills@britt to get all of these, project-foundations@britt for the project-setup and user-story/verification skills, llpm@britt for writing-product-specs, writing-user-stories, and writing-verification-plans, or chat@britt for daily-planning-ritual, writing-product-specs, and writing-user-stories.

Claude Code: Manual Installation

Skills can be installed globally or per-project.

Global installation (available in all projects):

# Clone the entire repository
git clone https://github.com/britt/agent-skills.git ~/.claude/skills/agent-skills

Project-specific installation:

# Clone the entire repository into the project
git clone https://github.com/britt/agent-skills.git .claude/skills/agent-skills

Restart Claude Code after installation to load new skills.

Codex CLI

This repo is a Codex plugin marketplace (.codex-plugin/plugin.json + the existing .claude-plugin/marketplace.json, which Codex also reads):

codex plugin marketplace add britt/agent-skills
codex plugin install agent-skills@britt        # all skills
codex plugin install project-foundations@britt       # curated bundle only
codex plugin install llpm@britt                      # curated bundle only
codex plugin install chat@britt                      # curated bundle only

Cursor

This repo ships a native Cursor plugin manifest (.cursor-plugin/plugin.json) and marketplace catalog (.cursor-plugin/marketplace.json). Cursor's marketplace-add flow is dashboard-based (no CLI yet): Dashboard → Plugins → Team Marketplaces → Add Marketplace → Import from Repo, pointing at britt/agent-skills, then install agent-skills (or project-foundations, llpm, chat) from Customize in the sidebar.

Codex CLI, OpenCode, and Cursor: Manual Installation

All three also discover skills the same way Claude Code does — a SKILL.md with name/description frontmatter, auto-triggered by relevance — under the shared .agents/skills/ path. This repo already exposes every skill there (.agents/skills/<name> symlinked to skills/<name>, so there's only ever one copy of each skill's content). This is the option to use for OpenCode, which has no marketplace, or if you'd rather not use Codex's/Cursor's plugin flow above.

Global installation (available in all projects):

git clone https://github.com/britt/agent-skills.git ~/.agents-skills-src
mkdir -p ~/.agents/skills
for dir in ~/.agents-skills-src/skills/*/; do
  ln -s "$dir" ~/.agents/skills/"$(basename "$dir")"
done

Project-specific installation: clone into the project and symlink, or run ./sync-agents-skills.sh after copying skills/ into your project's own .agents/skills/.

Codex CLI also reads ~/.codex/AGENTS.md / project AGENTS.md, OpenCode reads AGENTS.md and natively falls back to .claude/skills/, and Cursor also reads .cursor/skills/ — so a plain git clone into any of those tool-specific directories works too. Restart the tool after installation to load new skills.

Note: Some skills (like sgai-goal) reference Claude Code-specific tooling (e.g. CLAUDE.md, TodoWrite) and may need minor adaptation to work identically in other tools; most skills are plain instructional markdown and are unaffected. setting-up-a-project is multi-agent aware — it writes both a CLAUDE.md and an AGENTS.md adapted for Codex CLI, OpenCode, and Cursor. Granular per-skill plugin installs (like /plugin install architecture-diagramming@britt for Claude Code) aren't set up for Codex/Cursor yet — only the two bundles above; use the manual .agents/skills/ installation for individual-skill granularity on those tools.

Claude.ai (Web and iOS)

Skills can be added to Claude.ai projects as project knowledge:

1. Download the skill zip file

   • Go to the Releases page
   • Download the zip file for the skill you want (e.g., writing-product-specs.zip)

2. Add to your Claude.ai project

   • Open claude.ai and go to your project
   • Click on Project knowledge in the sidebar
   • Upload the zip file

3. Use the skill

   • The skill will be available in all conversations within that project
   • Claude will automatically use it when the context matches the skill's description

Note: Some skills (like sgai-goal and setting-up-a-project) require Claude Code-specific features and won't work in Claude.ai.

Rules

The rules/ directory contains reusable rule sets that can be dropped into your project.

TDD.rules.md

Strict Test-Driven Development rules for coding agents. Used by the setting-up-a-project skill, which references them from both CLAUDE.md and AGENTS.md.

Key rules:

• No production code without a failing test first (RED-GREEN-REFACTOR)
• 90%+ test coverage required on all metrics (configurable)
• Violations mean delete and start over
• Tasks aren't complete until tests pass, build succeeds, and no linter errors

Usage: The setting-up-a-project skill copies this file to your project root and fills in its placeholders, then references it from CLAUDE.md as @TDD.rules.md. You can also copy it manually:

curl -o TDD.rules.md \
  https://raw.githubusercontent.com/britt/agent-skills/main/rules/TDD.rules.md

Replace every {{PLACEHOLDER}} ({{TEST_COMMAND}}, {{BUILD_COMMAND}}, coverage thresholds, test file convention) with values for your tech stack.

rules/TDD.rules.md is a mirror. The canonical copy lives at skills/setting-up-a-project/TDD.rules.md so single-skill installs stay self-contained; ./sync-agents-skills.sh refreshes the mirror.

CLAUDE-specific-fuckups.md

A running list of mistakes Claude Code makes repeatedly, tied to gaps in its training data rather than general project conventions. Grows over time as new patterns are found; entry 1 is checking for outdated GitHub Actions versions and deprecated runtimes before adding or editing a uses: step.

Usage: The setting-up-a-project skill copies this file to your project root verbatim and references it from CLAUDE.md as @CLAUDE-specific-fuckups.md. It is Claude Code-specific and is not referenced from AGENTS.md. You can also copy it manually:

curl -o CLAUDE-specific-fuckups.md \
  https://raw.githubusercontent.com/britt/agent-skills/main/rules/CLAUDE-specific-fuckups.md

rules/CLAUDE-specific-fuckups.md is a mirror. The canonical copy lives at skills/setting-up-a-project/CLAUDE-specific-fuckups.md; ./sync-agents-skills.sh refreshes the mirror.

Documentation

• Contributing Guide - Repository structure, creating skills, and contribution guidelines
• Claude Code Skills Documentation - Official documentation

License

MIT