armor

Deep-dive interview to flesh out a spec

Walkthrough branch changes with teaching-focused feedback

Write a concise recap to LOG.md

Review recent changes using specialized agents

Two-pass security review with beatrix: find, then filter false positives

Type design analyst — use when introducing new types, reviewing type changes in PRs, or refactoring existing types. Provides qualitative feedback and quantitative ratings on encapsulation, invariant expression, usefulness, and enforcement.

Security reviewer — use when reviewing a diff or PR for security, auditing changed code for vulnerabilities, or threat-modeling a change. Finds high-confidence, genuinely exploitable findings, not theoretical ones, and proves the attack chain before reporting. Triggers on "security review", "is this safe", "vulnerability", "exploit", "security audit", "threat model this change".

Error handling auditor — use when reviewing code changes for silent failures, inadequate error handling, inappropriate fallback behavior, or any code that could suppress errors. Invoke after completing work involving catch blocks, fallback logic, or error paths.

Test reviewer — use when reviewing tests in a diff, PR, or recently modified spec files. Drives a discussion by asking the questions that separate valuable tests from theater. Flags tautologies, over-mocking, framework-testing, vague names, and tests that would still pass if the prod code were gutted.

Expert second opinion — use when stuck on complex coding problems, architectural decisions, or technical challenges. Consults an external model for validation and synthesizes the response.

Use this skill when the user wants to set up a new TypeScript project or add standard tooling to an existing one. Triggers on phrases like "set up this repo", "initialize the project", "add linting", "scaffold this", "set up tooling", "configure typescript", or "sow".

Fetch content directly from URLs and return it as markdown, text, or raw HTML. Use when the user wants to read a webpage, inspect a URL, pull docs from a site, extract page content from a link, or fetch web content without opening a browser. Prefer this over browser automation when a normal HTTP fetch is enough. Supports GitHub blob URLs and sites that return LLM-optimized markdown for Accept: text/markdown.

Choose between two things by feeling which one has more life. Read each design until you can hold it whole in your mind, set the ego aside, then ask which one is a truer picture of your self — which one you would rather become. Trust the gut before the reasons, and check whether another good engineer's gut agrees. A felt-sense tie-breaker for two options that both work. Use when stuck between two designs, APIs, schemas, or structures and the tests cannot separate them. Triggers on "which feels right", "which is more alive", "gut check between two designs", "which one is better", "can't decide between".

Break a tie between two implementations that both work. A forced pairwise comparison: ask which one you would rather maintain, not which one is cleverer. Use when stuck between two designs, APIs, names, schemas, or refactors that both pass tests and abstract criteria cannot separate. Triggers on "which approach", "can't decide between", "option A or B", "break this tie", "help me choose between two implementations".

Mine a codebase's recurring solutions into a named pattern language, or apply existing patterns to a new problem with the local adaptation. The constraint system that keeps design intuition grounded — without it, felt-sense wanders into impractical excursions. Use when documenting conventions, onboarding code into an existing repo, or asking how a thing is usually done here. Triggers on "what's our convention", "how do we usually", "extract patterns", "document conventions", "pattern language".