NeAGI
OrganizationLocal-first distillation wiki for AI agents. Turns chat noise into reusable decisions, skills, and project memory.
Categories
Indexed Skills (12)
local-agent-delegation
Use when Codex needs to delegate a bounded prompt to installed local agent CLIs or Hermes on the VPS, wait for their answers, collect artifacts, and compare results. Supports Claude Code, Mimo Code, Gemini CLI, Hermes VPS over SSH, and an Odysseus API placeholder. Do not use for web-only agents or for uncontrolled edits; use read-only mode by default and a fresh task worktree for edit scopes.
reefiki-dual-remote-publish
Use when publishing REEFIKI work to git, including "push", "пуш", "отправь на git", merge, PR, public snapshot, cleanup of task worktrees, or any dual private/public remote publication.
source-command-connect
Подключить кодовый проект к REEFIKI. Создаёт junction, маркер и wiki-проект если нужно. Запускать из корня.
source-command-new
Создать новый проект из шаблона. Интерактивно заполняет домен. Запускать в корне REEFIKI/.
source-command-sync-template
Синхронизировать обновления из _template в существующие проекты. Запускать из корня после правок шаблона.
browser-runtime-smoke-gate
Use after frontend, route, page, component, form, layout, auth, sync, local web app, deployed web app, or browser-visible behavior changes. Verifies real browser behavior with console and page error gates.
durable-harvest-closeout
Use when completed work produces a reusable procedure, decision, synthesis, debug runbook, audit outcome, release proof, project rule, or when the user says harvest, save, remember, фиксируй, сохрани, or запомни. Stores durable knowledge in the correct REEFIKI project.
reefiki-experience-monitor
Use before non-trivial work on a project that has REEFIKI wiki records. Retrieves relevant durable skills, decisions, concepts, synthesis and log entries so the agent does not repeat known mistakes or ignore project-specific rules.
safe-task-worktree-bootstrap
Use before starting a non-trivial coding task in an existing git repository, especially when the checkout may be dirty, multiple agents may work in parallel, or coordination files may be touched. Establishes a clean task branch, worktree, base, scope and cleanup boundary.
source-of-truth-diagnostics
Use when a user asks about config, installed tools, model/runtime capabilities, MCP/plugins, paths, versions, logs, health, access, or any fact likely to drift. Verifies the live source of truth before answering from memory.
staged-scope-and-publish-guard
Use before commit, push, publish, release, harvest, or handoff in dirty, multi-project or public/private repositories. Checks staged paths, generated artifacts, private/public boundaries and project publish gates.
verification-before-completion
Use before declaring a task done, especially after code, docs, adapters, data, release, deployment or infrastructure changes. Maps every completion claim to fresh evidence and calls out missing proof.
Bio shown is the top-scored skill's repo description as a fallback — real GitHub bios land in a future update.