vivswan-skills

Message a Claude Code session on another machine with the same push delivery two sessions get on one machine: two ssh socket forwards, no files copied, no polling.

1 skill

Standards for maintainable code, applied while writing and folded into reviews.

1 skill

Keeps the Codex desktop Chrome plugin driving Brave and other Chromium browsers: cache sync after app updates, browser-only auth under a custom model provider, and browser-family selection.

1 skill

Create skills and memories to a quality bar, and repair them at their source when they fail.

1 skill

Art direction for public-facing web pages, dealt from a library of researched directions and held to a no-slop house style with a browser inspector

1 skill

Distinctive, intentional visual design for new or reshaped UI: aesthetic direction, typography, and choices that do not read as templated defaults

1 skill

Design direction for frontend work: commands to design, critique, audit, polish, or harden a web or native interface, with detector rules against AI-template tells

1 skill

Write and edit prose that avoids the documented signs of AI writing from Wikipedia's field guide.

1 skill

Never fix the same failure class twice. Climb every correction to the most durable response reachable: architecture and data structures first, then CI checks, then rules.

1 skill

Refactor invariants out of runtime checks and into the type system: parse at boundaries, then make invalid internal states impossible to construct.

1 skill

Run a session as an orchestrator: parallel worktree subagents, gated landings (direct or PR-based), fleet monitoring.

1 skill

PR bodies and issues that show the change instead of describing it, shaped to the change, with the human part short and anything written for a tool collapsed at the bottom.

1 skill

Carries a pull request from opened to landed: draft state that tracks convergence, same-cycle review-comment triage, line accounting against the stated purpose, and a human hand on the merge by default (a standing-approval label applied by an authorised approver, where the repository uses one, hands over the merge).

1 skill

Name a project, package, or paper from a table of checked candidates, scored against the names the owner already likes.

1 skill

Replies to issue reporters and reviews of someone else's PR: plain words first, the reporter's own log lines, questions only when the report does not answer them, and one pending review the user publishes.

1 skill

Cross-model code review using a second agent, tool, or read-only CLI fallback.

1 skill

One daemon per session, a background task of that session, polls any source and delivers its events as messages in the peer envelope, with no text quoted from the polled system, and as a log agents read with wait.

1 skill

Cuts AI tells from any writing

1 skill

Controls before a zero or absent reading, an alarming finding, a success claim, or stillness becomes a conclusion: evidence-bearing probes, positive and negative controls, second derivations, postconditions over logs, and a re-measure at the moment of sending.

1 skill

One daemon per session, a background task of that session, watches pull requests and branch tips and delivers comments, reviews, CI verdicts, head moves, label moves, conflicts, merges, and closes as messages instead of polls.

1 skill

Working text that scans well and names real code: the device picked from the content, a paragraph-cap and path probe, fact-preserving restructures, and the architecture page whose boxes name real symbols.

1 skill

Safety rules for git worktrees: verified removal, explicit handovers, one branch per worktree, shared-config hazards, and per-file ownership for parallel actors.

1 skill

Installable, plugin-ready skills that give coding agents disciplined workflows, from code standards and review gates to multi-agent orchestration