Mariano215
UserRun a different Claude Code setup in every terminal. Profile switcher with per-terminal config isolation.
Categories
Indexed Skills (19)
fullreview
Comprehensive multi-domain code + security + ops review driven by parallel agent teams, looped until clean. Use when the user types /fullreview or asks for a "comprehensive review", "full review", "full audit", "ship-readiness audit", or wants multiple domains (security, backend, frontend/backend sync, background jobs, service integrations, perf, observability, migrations, config, telemetry) reviewed simultaneously across several iterations.
gauntlet-loop
Run a Gauntlet Loop (Matt Shumer's method), driving work toward a concrete reference standard with paired builder and critic agents that loop until the output matches the bar. Use when the user says "gauntlet loop", "/gauntlet", "run a gauntlet", "make this match [reference]", "keep looping until it looks like X", or asks for output judged against a named example rather than a checklist. Best for visual, creative, or taste-driven work (game feel, landing pages, prose voice, design polish) where "done" is a comparison, not a passing test. Not for work with an objective pass/fail signal, use /tdd or /debug there.
handoff
Generate a structured handoff document (project identity, current state, architecture map, recent decisions, next steps/risks) so a new AI session or teammate can pick up full context fast. Use when the user types /handoff, asks to "write a handoff", "hand this off", "prep a handoff doc", or is wrapping up a session and wants context preserved for next time.
ingest-scrubbing
Sanitize documents at ingestion time, before they're embedded and indexed in a RAG/vector store. Different attack surface from prompt-injection-defense: the poison sits in the corpus and fires whenever a later, unrelated query happens to retrieve it.
prompt-injection-defense
Harden any endpoint that feeds untrusted text (user messages, tool results, fetched web content, bank/API data) into an LLM prompt. Zero-cost baseline layers plus guidance on when heavier ML/human-review layers earn their cost.
git-best-practices
Enforce git workflow conventions: commit message format (conventional commits), branch naming (feature/, bugfix/, hotfix/), pre-commit checks (lint, tests, docs, secrets scan), and pre-push validation (rebase, squash, full tests, CHANGELOG).
repro
Use when the user invokes /repro or asks to iterate a verification or review task a bounded number of times. Triggers on "/repro <task>", "do this at most N times", "repeat until clean", "loop X checking Y". Parses the prompt, announces the loop plan, executes iteratively with per-iteration status lines, and produces a final report.
security-audit
OWASP Top 10 checks, SQL injection prevention, XSS/CSRF protection, secrets management, dependency vulnerability scanning, and secure coding practices.
goal-iteration
Pair a superpowers methodology skill with the built-in /goal loop for work that has a verifiable end state. Use when the user wants test-driven implementation, a bug fixed, or a feature built, and the finish line can be proved from Claude's own output (tests pass, build exits 0, bug no longer reproduces). Not for open-ended design, prose, or anything a command cannot check.
harness-builder
Use when the user wants a new tack harness or profile for a kind of work (legal, data, a client, a stack), asks "build me a harness for X", "add a profile", "what should a X profile have", or wants an existing profile redesigned.
pre-push-auto-fix
Auto-fix lint, test, and format issues before pushing. Use when the user says ready to push, can I push, or check before push. Conservative limits: 3 iterations, 10 minutes. Trigger: /pre-push
smart-agent-spawner
Selects model tier (opus/sonnet/haiku) and effort level for spawned agents/subagents based on task complexity, and escalates on failure. Applies automatically before any Agent, Task, or Workflow agent() call.
build-with-goal
Use for feature/build/create/implement requests that require design, code, tests, and verification. Starts with repo exploration and planning before implementation.
document-with-goal
Use for README, docs, changelog, API docs, runbooks, migration guides, or documenting existing behavior.
graphify
Use when a repo has graphify output, the user asks about a knowledge graph, or codebase navigation would benefit from graph relationships.
debug-with-goal
Use for bugs, failures, errors, crashes, regressions, broken behavior, or repeated failed fixes. Requires root-cause investigation before fixes.
ghost
Use to review prose for LLM-tell phrasing, banned punctuation, vague marketing language, or client-facing writing polish.
pre-push-review
Use before pushing, publishing, opening a PR, or when the user asks whether a branch is safe to push.
tdd-with-goal
Use when the user asks for TDD, tests, test-first work, coverage, or implementation with tests.
Bio shown is the top-scored skill's repo description as a fallback — real GitHub bios land in a future update.