ivuorinen

ivuorinen/skills

Command

Hostile single-shot audit of the signal surface a codebase emits: assume every production failure is invisible until the emitted logs, metrics, traces, and alerts prove otherwise.

ivuorinen/skills

Command

Hostile single-shot performance audit: assume every data path degrades superlinearly until traced and proven otherwise.

ivuorinen/skills

Command

Turn a change request into an implementation plan that survives hostile scrutiny before a line of code is written. Assumes every plan is naive — hiding unhandled failure modes, security holes, over-engineering, and false assumptions — until the audit lenses prove otherwise. Produces a plan document and stops…

ivuorinen/skills

Command

Hostile but constructive review of a diff or set of changed files, packaged as copy-paste-ready markdown for a GitHub PR comment. Stdout-only: this command never writes a findings file — it overrides the findings-store protocol in conventions.md.

ivuorinen/skills

Command

Hostile single-shot data-privacy audit: assume every element of personal data is stored and transmitted without the control its data class requires until the code proves the control exists.

ivuorinen/skills

Command

Hostile audit of a codebase that integrates a language model: assume every string reaching the model is attacker-controlled, every token the model emits is attacker-authored, and every tool the model can call will be called with the worst possible arguments — until the code proves otherwise. The model is an untrusted…

ivuorinen/skills

Command

Pass/fail gate over the findings store. Writes nothing, fixes nothing.

ivuorinen/skills

Command

Hostile audit of behavior when a dependency fails, a call is retried, or a message is redelivered: assume every retry double-applies, every external call hangs forever, and every crash lands in the worst window — then prove where. A call that is idempotent, bounded, timed-out, and backed-off survives; anything else is…

ivuorinen/skills

Command

Re-verifies every open finding in the store against the current code and gives each an evidence-backed disposition: resolve the proven-fixed and proven-invalid, keep the still-live open, flag the unverifiable for a human. It changes no code and files no new findings — it only re-checks and resolves findings already on…

ivuorinen/skills

Command

Hostile code review of a diff, file, or module: assume the code is broken and prove it. The job is finding bugs, not being helpful.

ivuorinen/skills

Command

Automated, tool-driven security audit: probe which security scanners are installed, run each one, normalize and deduplicate the output, and file consolidated findings. No tool is assumed present — every tool is probed before use.

ivuorinen/skills

Command

Teach the user a skill or concept inside a persistent teaching workspace. This is a stateful, multi-session request: the user intends to learn the topic over time, and each run builds on the last. The topic is the text after the command (/nitpicker teach rust ownership); with no topic, read the existing workspace and…

ivuorinen/skills

Command

Hostile audit of the test suite itself: assumes every test is weaker than it looks and proves it — a green suite is the null result, not evidence of quality. Findings are about tests only; fixes add or strengthen tests, never touch production source.

ivuorinen/skills

Command

Scans a repo (or the changed files) and emits a ranked run-plan of which nitpicker commands to run, each justified by a cited repo signal. This is a selector, not an auditor: it recommends commands, files no findings, and runs nothing itself.

ivuorinen/skills

Command

Hostile audit of the static type layer: assume every any, every suppression, and every unchecked cast is hiding a real type error until proven otherwise. The type checker is only as strong as the escape hatches punched through it — this command finds every hole.