vivswan-skills

Use when writing code changes or asked to apply house code standards.

Use when creating a skill or memory, when one fails in practice, or when the user reports one misbehaving or stale.

Use when asked to humanize text, remove AI tells, check whether text reads as AI-generated, or draft prose that must not sound AI-written.

Use when you just corrected an agent or fixed the same failure class twice, or when a change removes a rule, guard, or check.

Use when code or a design guards invariants with repeated runtime checks, lifecycle flags, or the same guard at every consumer, or when asked to make invalid states unrepresentable or parse-don't-validate.

Use when asked to act as orchestrator, orchestrate with worktrees, run a fleet of parallel subagents on multi-track work, or build and manage many PRs in parallel.

Use when opening a pull request, changing its body, title, or draft state, triaging review comments, writing an issue, or deciding who merges.

Use when asked to rubber duck a change, get a second opinion, or run an independent cross-model review of code changes.

Use when about to report a zero or absent reading, an alarming finding, a success claim, or stillness from a probe or status read, so the reading is evidence-bearing and controlled first.

Use when pushing commits to a remote with CI, merging a PR, or when asked whether a pipeline passed, so every push and merge gets a background watcher.

Use when creating, removing, or handing over git worktrees, or deleting their branches, or when several actors share one repository through worktrees.