By jimeh
Eval-Driven Development: evals-first feature workflow with sub-agents for spec, implementation, and verification
Archive a completed, deferred, or dropped EDD feature
Refine and iterate on evals for a Draft-phase EDD feature
Copy EDD commands, agents, and skill into the local project for portability
Implement a feature from its spec using parallel sub-agents where possible
Bootstrap the EDD (Eval-Driven Development) system into the current repo
Adversarial requirements reviewer for EDD features. Spawned during /edd-new to suggest edge cases, error scenarios, and constraints the user may have missed. Use this agent when creating a new EDD feature or when the user wants help brainstorming acceptance criteria. <example> Context: User just created a new EDD feature user: "I created EDD feature 003 for user tagging. Can you brainstorm edge cases?" assistant: "I'll use the eval-brainstormer agent to review your feature for gaps." <commentary> User explicitly asks for requirements brainstorming on an EDD feature. </commentary> </example> <example> Context: User is writing evals and wants adversarial review user: "What could go wrong with this feature? Review my evals." assistant: "I'll use the eval-brainstormer agent to do an adversarial review." <commentary> User wants someone to poke holes in their acceptance criteria. </commentary> </example>
Implements a single task or batch of related tasks from an EDD feature's task list. Spawned by /edd-impl with a clean context scoped to specific tasks. Use this agent when executing implementation tasks for an EDD feature. <example> Context: Orchestrator is dispatching implementation tasks user: "Implement Task 1 and Task 2 for EDD feature 003" assistant: "I'll use the implementer agent to implement these tasks." <commentary> Parallel-safe tasks dispatched to an implementer with scoped context. </commentary> </example> <example> Context: Sequential task needs implementation after prerequisites user: "Task 3 for EDD-003 is ready, its dependencies are complete" assistant: "I'll use the implementer agent to implement Task 3." <commentary> Dependent task dispatched after its prerequisites are verified complete. </commentary> </example>
Derives a product spec and implementation task list from frozen EDD evals. Spawned by /edd-spec with a clean context. Use this agent when an EDD feature's evals are ready and a spec needs to be generated. <example> Context: User has frozen evals and wants a spec generated user: "/edd-spec 003" assistant: "I'll use the spec-writer agent to derive a spec from the frozen evals." <commentary> The /edd-spec command triggers spec-writer to produce spec.md and tasks.md. </commentary> </example> <example> Context: User wants to regenerate a spec after reverting to Draft user: "I updated the evals, regenerate the spec for feature 003" assistant: "I'll use the spec-writer agent to derive a fresh spec from the updated evals." <commentary> Evals were modified after a revert to Draft; spec needs regeneration. </commentary> </example> <example> Context: User reviewed spec and wants targeted changes user: "/edd-spec 003 split task 4 into smaller steps and add error handling to the API section" assistant: "I'll use the spec-writer agent to refine the existing spec based on your feedback." <commentary> Existing spec/tasks get refined rather than regenerated from scratch. </commentary> </example>
Strict, skeptical QA verifier for EDD features. Checks implementation against the evals with fresh eyes — does NOT receive the spec or task list. Spawned by /edd-verify with a clean context. <example> Context: Implementation is complete and needs verification user: "/edd-verify 003" assistant: "I'll use the verifier agent to check the implementation against the evals." <commentary> Feature implementation complete, verifier checks against evals only. </commentary> </example> <example> Context: User wants verification with browser testing user: "/edd-verify 003 http://localhost:5173" assistant: "I'll use the verifier agent to verify against evals, including browser testing." <commentary> URL provided for runtime verification of UI criteria. </commentary> </example>
Uses power tools
Uses Bash, Write, or Edit tools
Runs pre-commands
Contains inline bash commands via ! syntax
Own this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge. GitHub access is read-only (username + org membership).
Sign in to claimOwn this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge. GitHub access is read-only (username + org membership).
Sign in to claimBased on adoption, maintenance, documentation, and repository signals. Not a security audit or endorsement.
My personal AI coding agent configuration, with any quirks, oddities, opinionated rules, and hallucination-inducing prompt fragments I live with on a daily basis.
One repo, one set of rules, symlinked into every agent's config directory.
Supports Claude Code, Codex, and any tool that reads AGENTS.md.
[!Warning]
This is my personal config. It is not a starter kit, a framework, or a best-practices guide. If you use it as-is, things will probably work — but they'll work my way, which may not be your way. Browse for ideas, steal what's useful, but don't say I didn't warn you.
git clone https://github.com/jimeh/agentic.git ~/.config/agentic
cd ~/.config/agentic
./setup.sh # create symlinks (skips existing)
./setup.sh --force # replace existing (backs up to .bak)
This creates symlinks from the repo into ~/.claude/, ~/.agents/, and
~/.codex/, and registers plugin marketplaces and installs plugins via the
Claude CLI. Run ./setup.sh --help for details.
RULES.md — Single source of truth for all agent behavior rules.
Symlinked as the global rules file for each supported agent. Edit this file
directly — never edit the symlink targets.claude/ — Claude Code settings and statusline script.codex/ — OpenAI Codex config.skills/ — Custom skills (auto-discovered by setup.sh).plugins/ — Claude Code plugins, published via a local marketplace.docs/references/ — External articles and guides.Skills are auto-discovered — drop a directory in the right place, re-run
setup.sh, done. Commands live in plugins. Plugins are registered and installed
via the Claude CLI (claude plugin marketplace add / claude plugin install).
A PreToolUse hook that strips redundant git -C <cwd> flags from Bash
commands when the path matches the current working directory. Claude Code tends
to add these unnecessarily, and the -C flag changes the command string enough
that pre-approved git commands no longer match the allowlist — causing repeated
permission prompts.
Handles all -C syntax variants (space, =, bare, quoted) and compound
commands (&&, ;).
Slash commands for common git workflows:
/commit — Stage changes, create a well-formed commit./commit-push-pr — Commit, push, and open a PR./rebase — Rebase onto upstream main/master./clean-gone-branches — Clean up branches deleted on remote.Derived from the official commit-commands plugin, heavily modified.
Slash commands for managing AGENTS.md files:
/claude-md-to-agents-md — Convert CLAUDE.md to AGENTS.md./generate-agents-md — Generate AGENTS.md from codebase analysis./refactor-agents-md — Refactor AGENTS.md for progressive disclosure.A disciplined research-plan-implement workflow. Instead of jumping straight to code, you move through distinct phases so every decision is reviewed before implementation begins.
/research — Deep-read a codebase area, write findings to research.md./plan — Create a detailed implementation plan in plan.md./refine — Address inline notes you've added to the plan./todo — Add a granular task breakdown to the plan./implement — Execute the plan mechanically, marking progress./review — Optional post-implementation sanity check.Also includes agent-agnostic prompt snippets for use with any AI coding assistant.
Initializes a lightweight Feature Design (FD) tracking system in any project. Scaffolds directory structure, templates, index, project-local slash commands, and CLAUDE.md conventions for the full FD lifecycle.
/fd-init — Set up the FD system in the current project.Based on the Feature Design system by manuelschipper.
You can install individual plugins directly without cloning the repo. First add the marketplace, then install whichever plugins you want:
# CLI
claude plugin marketplace add jimeh/agentic
# Or from within Claude Code
/plugin marketplace add jimeh/agentic
Then install plugins:
# CLI
claude plugin install git-commands@jimeh-agentic
# Or from within Claude Code
/plugin install git-commands@jimeh-agentic
setup.sh ensures both the official claude-plugins-official marketplace and
this repo's local marketplace are registered, then installs plugins listed in
the CLAUDE_PLUGINS array at the top of the script. To add or remove
auto-installed plugins, edit that array.
npx claudepluginhub jimeh/agentic --plugin eddResearch-plan-implement workflow: phased commands for disciplined AI-assisted development
AGENTS.md management: generate, refactor, and convert from CLAUDE.md
Strips redundant git -C flags and cd prefixes when the path matches the current working directory
Feature Design (FD) system: lightweight feature tracking with templates, index, and slash commands
Git workflow slash commands: commit, push, PR, rebase, and branch cleanup
Planning and execution workflows for Claude Code. Based on obra/superpowers.
Complete project development toolkit: 23 agents, 23 slash commands, 29 lifecycle hooks, and 69 reusable skills for Claude Code workflows
Portable Development System — AI-assisted development methodology with skills for consistency and agents for scale.
Claude Code 插件開發工具包,提供七個專業技能、三個代理和完整的工作流程命令,用於構建高品質插件
11 agents, 35 skills, 18 commands, 9 hooks — spec-driven multi-agent orchestration for Claude Code, with optional cross-device semantic memory.
AI-powered agents for specialized development tasks