skills

box a git repo, clone it locally and search the real source, with a tracked manifest. Delegates repo work to subagents when available, runs in the main thread otherwise. Triggers: a VCS URL (GitHub, GitLab, Bitbucket), a previously cloned repo name, or a request to clone/search/explore a repo. Flags: --persist, --update, --list, --no-subagents.

explain-diff: explain a code change, diff, branch, or PR as a rich self-contained HTML page. Flags: --target <branch>, --pr <n>, --staged, --unstaged, --output <path>.

fix a Linear ticket end to end: fetch it, branch, classify mode, plan, implement, review. Flags: --base <branch> (default main), --mode <mode> (default auto; values: auto, scratch, modify, debug, trivial), and a ticket ID positional arg.

fix PR review feedback: fetch unresolved threads, triage each finding skeptically, fix what holds, reply on threads, commit and push. Triggers: /fix-pr, address PR reviews. Reached by prath-mode after make-pr. Flags: --pr <n|url>, --no-push, --no-reply.

notify the user via Discord webhook embed. Triggers: /notify, send a Discord notification, alert me when done, ping me on Discord. Flags: --task, --link, --webhook, --color, --field, --content, --dry-run.

Route a task to the right skill in prathamdby/skills and run multi-step workflow chains. Verifies the leaf skill is installed before invoking it.

make a pull request with a plain-English title and a clean-room thematic summary built from the branch diff. Triggers: open a PR, create a pull request, submit a PR. Flags: --target <branch> (default main), --ticket <id> (prefix title with the ticket), --conventional (conventional-commit title).

commit changes to git with a clean-room message derived from the diff alone. Skips hooks with -n by default; --verify runs them. Triggers: /commit, commit changes, save work to git, write a commit message. Flags: --staged/--unstaged for scope, --conventional/--simple for style, --verify to run hooks.

recon builds a persistent memory of a codebase and keeps it current. Cold start explores the unknown; every later run explores only the drift. Triggers: /recon, explore or map a codebase, re-orient in a repo after time away. Flags: --refresh, and a positional focus argument.

Session mode: act as orchestrator brain only. Research and implementation go to cheaper-model subagents; the orchestrator scopes, briefs, verifies, and judges. User-invoked with the task as the argument.

Ultra-compressed replies that cut ~75% of tokens by dropping filler, articles, and pleasantries while keeping full technical accuracy. User-invoked only.

handoff the session, compact the conversation into a document a fresh agent can resume from, or restore context from a saved handoff. Triggers: /handoff, save session context, hand off to a new agent, resume from a handoff file. Flags: --resume <path>, --path <path>, and a positional focus argument.

peer-review a plan, design, or implementation, find the one critical risk, list other gaps, propose a fix for the critical risk, and update the plan. Triggers: peer review a plan, review an implementation, assess a design, check a proposal for gaps.

assign a task to an external coding agent and monitor it. Pipes the prompt via stdin to dodge quoting failures and auto-approves permissions to avoid silent hangs. Triggers: run a task with an external agent, hand off to OpenCode, Claude Code, or similar, execute a plan non-interactively. Flags: --agent <name> (default opencode), --model <model>, and --dir <path>.

deslop a diff, remove AI slop and over-engineering while preserving exact behavior. Triggers: /deslop, remove AI slop, clean AI artifacts, strip over-engineered patterns, simplify code, review changes for bloat. Flags: --staged, --unstaged, --base <branch>.