Skill-based workflow engine for AI coding agents: planning, TDD, worktree orchestration, and session learning
Skill Claude Code
Structured exploration before a design is settled. Use when a request is underspecified, when several designs are plausible and you are about to pick one without saying so, or when the user is thinking out loud rather than asking for code.
Skill Claude CodeCodex
Review gate for a change about to be handed off. Use when an implementation or task is finished and needs checking against its requirements, before opening a pull request or merging, or when review feedback has come back and needs answering.
Skill Claude CodeCodex
Root-cause discipline for a failure you cannot yet explain. Use when a test fails, a bug is reported, behavior surprises you, a build or CI step breaks, or a fix you already tried did not hold.
Skill Claude CodeCodex
Excalidraw diagram builder for an Obsidian vault. Use when the user asks to be shown something visually — an architecture, a flow, a mind map, a "draw this" — or when a note's content needs a diagram beyond an embedded Mermaid block.
Skill Claude CodeCodex
Dispatch discipline for work that can run in parallel. Use when several pieces of work could run at once, when work needs an isolated workspace to run in, when writing the brief a dispatched agent will run on, or when dispatched work reports back and has to be accepted.
Skill Claude CodeCodex
Measurement discipline for claims about agent behavior. Use when a behavioral claim needs evidence before shipping, when designing or auditing an eval scenario, or when eval numbers have come back and a verdict is about to be read out of them.
Skill Claude Code
Break work into a checkbox task list and run it to completion. Use when a change needs more than one step, when a task list is already waiting to be executed, or when you are picking between working alongside the user and running unattended.
Skill Claude CodeCodex
Integrate completed work. Use when implementation is done and you are deciding whether to merge, open a PR, keep, or discard a branch or worktree.
Skill Claude CodeCodex
The opt-in learning loop: capture a session diary, extract patterns from diaries, review what activates.
Skill Claude CodeCodex
Start and supervise an unattended loop that works a task list across fresh sessions.
Skill Claude Code
Vault interface for an Obsidian knowledge base. Use when something should be filed into the user's notes, when a question should be answered from the vault rather than from general knowledge, when vault health needs auditing, or when a new vault needs bootstrapping.
Skill Claude CodeCodex
Continuity across a break in context. Use when a session is ending or handed over, when stopping mid-task, when resuming from a handover or a named session archive, when a session should be saved under a name, or when a long session is filling up and compaction is the question.
Skill Claude Code
Spec-driven product state a project is maintained from. Use when a repo keeps living specs and a decision log under product/ and the code is about to change, when a recorded decision is being reversed, when a milestone is picked up or shipped, or when asked to start that state.
Skill Claude CodeCodex
Test-first implementation discipline. Use when adding a function or feature, fixing a bug, changing existing behavior, or when implementation code already exists with no test covering it.
Skill Claude CodeCodex
Router and index for the arcforge toolkit. Use when you are unsure which skill fits the work in front of you, want the shortest workflow for a task, or need a map of what this toolkit actually offers.
Skill Claude CodeCodex
Author an arcforge skill that actually changes agent behavior — invocation, description, form, evidence.
At most 3 mods per repository are shown here, and a mod shipped inside a plugin is left to that plugin's page — the rest are on their repository pages: