skills

use-skill when running a remote skill on demand from one or more GitHub links to skill files, fetching via gh api and executing the fetched procedure as if natively present.

box when cloning, updating, listing, or searching an external git repository from its real local source.

claude-code when launching the local Claude Code CLI (`claude`) for a first-run install, interactive session, one-shot, stream, resume, worktree, or auth, MCP, or plugin command.

codex when launching the local Codex CLI (`codex`) for a first-run install, interactive TUI, one-shot exec, stream, worktree, review, resume, login, MCP, or plugin command.

commit when saving scoped git changes with a message derived only from the committed diff, especially after tickets or reviews could bias the wording.

cursor-agent when launching the Cursor Agent CLI (`cursor-agent`) for a first-run install, interactive session, one-shot, stream, persist, worktree, MCP, plugin, worker, or Bedrock command.

deslop when removing AI-generated bloat, needless defenses, foreign patterns, or avoidable complexity from a git diff without changing behavior.

fix-pr when exhaustively handling open pull-request feedback, including nested discussions, CI, invalid suggestions, arrivals, fixes, and replies. Loads gh for hunt and reply I/O.

gh when inspecting a PR, reading review threads, diagnosing red CI, posting a thread or conversation reply, or composing non-trivial gh commands. fix-pr loads this skill for hunt and reply I/O. Node below 23 or ERR_UNKNOWN_FILE_EXTENSION is not a skip to GraphQL.

handoff when saving resumable session state or continuing work from an existing handoff document.

make-pr when publishing committed branch changes as a new pull request or updating the existing pull request for that branch.

Orchestrate user-invoked tasks that should be split across cheaper subagents while the main agent remains responsible for scope and verification.

peer-review when deciding whether an implementation plan, design, or proposed change is ready to build.

prath-mode when routing goal-shaped work to a playbook that sequences prathamdby/skills leaves, including multi-step delivery chains.

recon when mapping the current codebase, refreshing an earlier map, or re-orienting after repository changes.

todoist-task when creating or previewing Todoist tasks from natural-language requirements, including projects, subtasks, dates, and consistent wording.

upfront-design when a request is large enough that product behavior, system architecture, program design, and implementation order should be agreed with the user before code, or when given an existing design to resume or check.