Socratic
Meta — questioning about questioning. Question yourself till you're left with only answers. Use whenever the user asks to build, design, scaffold, or architect any system, feature, service, app, agent, pipeline, connector, or tool — especially when the request is short, vague, or underspecified. By default the agent interrogates ITSELF silently across the relevant engineering domains (frontend, b…
One Context Mcp
One local MCP server that gives Claude, Cline, Codex, and other AI tools the same project memory.
Readback
Maintain a user-approved, per-project record of architecture and design changes made during AI-assisted work. Use when an architecture direction, implementation design, or consequential decision is settled, changed, rejected, or implemented, and ask the user before creating a record.
Groundtruth
Use before starting any non-trivial build (write the plan to a persistent file, not just a chat reply); before pushing commits, opening a PR, or making a repository public (scan for leaked keys, tokens, absolute paths, and internal hostnames first); and before declaring something done, fixed, working, or ready (verify against a real run or real data, not test-green optimism alone). Three routes,…
Mentor
Use when the user is deliberately relearning or building hands-on skill in a language, framework, or system — they write every line themselves and want the agent to teach, not implement. Trigger phrases include "I want to write this myself", "explain it, don't code it", "quiz me", "review what I wrote", "don't give me the fix", or a stated rule that the agent must not write application code for t…
Commenter
Add or improve useful code comments by explaining abstractions, contracts, constraints, intent, and non-obvious design decisions; supports beginner, intermediate, and advanced detail levels.
Mandate
This skill should be used before acting on an instruction that could reach further than the user intended — writing into a project other than the one being worked in, doing work that was discussed but not requested, taking a next step the user has not reached yet, or treating an earlier approval as covering a new action. It should also be used when the user says "I didn't ask for that", "why did…
Agent Structure
Use at the start of substantial agentic AI work, including multi-step builds, redesigns, and new features, to apply structured skills for authority, memory, verification, and code-documentation clarity.
One Context
One local MCP server that gives Claude, Cline, Codex, and other AI tools the same project memory.
Architect
Use when the user wants to build hands-on but by assembling code themselves rather than writing it from a blank file — they want the agent to hand over one function or concept at a time in chat, with explanation, which they then type or paste into the project and merge on their own. Trigger phrases include "give it to me in chunks", "paste it myself", "dump it in chat, I'll merge it", "walk me th…
Node Map
Maintains a persistent, append-only knowledge graph of a codebase — one root index plus one node file per project/folder — so future sessions read a compact map instead of re-reading source code. Use this skill whenever an architectural decision is actually implemented (code committed or files changed after a design decision), when the user runs git commit in a mapped workspace, when the user ask…