Claude Skills by shadok-ai

Use when PRs are open on shadok-ai/shadok-ai and need bringing to a merge — a periodic pass over the PRs, a green PR to merge, a PR behind main (strict mode), a rebase conflict, a red CI verify to repair, or doubt about whether a PR is eligible for automatic merging.

Create and drive isolated Claude Code agents through the shadok-ai server (git worktrees, prompts, dialogs, diff). Use when the user wants to delegate a task to a shadok-ai agent, launch agents in parallel, or inspect/drive existing shadok-ai sessions.

Verify whether something is already resolved/decided before you assert or act on its status, and record resolutions/decisions so sibling agents don't re-surface them. Use before reporting a bug, proposing a feature, saying "X isn't done", or acting on a premise like "the PR isn't merged" — and whenever you resolve, decide, or change something notable. Not for code/PR status (use git/gh); for everything without a system of record (infra, marketing, site, ops, tasks).

Reload (respawn) your OWN agent session to pick up a changed pilot prompt or newly-seeded skills — for example after an operator enables a feature. You keep your history (it is a resume), you just come back as a fresh process with the new prompt/skills active. Use when told a change won't take effect until you reload, or when you (a lead) refresh an agent.

Schedule recurring prompts (monitoring, reporting, watching) for the CURRENT shadok-ai channel — with an optional deterministic "check" that runs WITHOUT the LLM so routine "nothing to report" runs cost ZERO tokens. Use whenever the user asks to monitor / report / watch / alert on something on a recurring basis (every N minutes, hourly, daily, each morning…).

Store a credential you OBTAINED yourself (a CLI login, a provisioned key, an API token you just created) in shadok-ai's secret vault, so the next agent doesn't have to obtain it again. Use when you end a task holding a credential that outlives the session. Never for a secret the user typed in the chat.

> **For agentic workers:** REQUIRED SUB-SKILL: Use superpowers:subagent-driven-development (recommended) or superpowers:executing-plans to implement this plan task-by-task. Steps use checkbox (`- [ ]`) syntax for tracking. **Goal:** A project skill (`.claude/skills/shadok-ai-agents/`) that lets Claude Code create and drive shadok-ai agents through the web server, with a `pilotctl.mjs` thin client offering one-shot commands and JSON output. **Architecture:** `pilotctl.mjs` speaks the server's ...