adityaarakeri
UserEight portable SKILL.md playbooks that make coding agents work like careful senior engineers. One-command install across Claude Code, Codex, Antigravity, OpenCode, Copilot, and Cursor.
Categories
Indexed Skills (9)
evaluate-ai-changes
Evaluate changes to prompts, model or provider versions, RAG retrieval and ranking, tool schemas, agent policies, structured outputs, safety filters, and model routing. Use when AI-system quality, safety, latency, or cost may change. Do not use for ordinary deterministic code changes; use tdd-loop for those and combine both skills when an AI change also modifies deterministic adapters.
git-hygiene
Execute explicitly requested version-control mutations safely, including staging, commits, branches, merges, rebases, conflict resolution, undo operations, and pull-request preparation. Always use before force push, reset, clean, history rewrite, or deletion. Do not trigger for read-only status, diff, log, blame, or show requests that must leave repository state unchanged.
plan-first
Produce and maintain an implementation plan for risky, ambiguous, irreversible, or coordinated changes, and whenever the user asks for a plan or approach. Trigger on public contracts, schemas, auth, payments, migrations, build or deployment systems, and material design choices. Do not use for low-risk mechanical work with an obvious established implementation.
repo-recon
Map an unfamiliar repository, service, or module before non-trivial work. Use when entering a codebase not explored in the current session, onboarding, tracing an unknown flow, or recovering from a wrong structural assumption. Do not use for a scoped, obvious change when the relevant files, commands, and conventions are already known.
safe-refactor
Change internal code structure while preserving observable behavior through small verified steps. Use for renames, extractions, moves, decomposition, and duplication removal when behavior must remain stable. Do not use for bug fixes, public contract changes, dependency or framework migrations, schema migrations, or other work expected to change compatibility or runtime behavior.
self-review
Review a completed diff or change set for correctness, security, regressions, scope, and missing tests. Use in author mode before handoff or in read-only mode when the user asks to review, inspect, check readiness, or report findings. Do not replace runtime completion evidence from verify-done, and do not fix findings during a review-only request.
tdd-loop
Implement a known feature, behavior change, or diagnosed bug with a red-green-refactor loop. Use when the user asks to add, build, implement, or fix behavior and automated tests exist or should exist. For an unknown failure cause use debug-protocol first; for model, prompt, RAG, or agent quality use evaluate-ai-changes alongside deterministic tests.
verify-done
Gate completion claims with observed build, test, runtime, documentation, and diff evidence. Use at the end of implementation or when the user asks whether existing work is complete. Do not use as the primary code-review workflow, and do not repair failures during a verification-only request unless the user separately authorizes implementation.
debug-protocol
Diagnose bugs, failing tests, crashes, regressions, flaky behavior, and performance failures when the cause is unknown. Use for stack traces, wrong output, intermittent failures, environment mismatches, or repeated unsuccessful fixes. Do not use when the root cause is already established and the task is to implement the known fix; hand that work to tdd-loop.
Bio shown is the top-scored skill's repo description as a fallback — real GitHub bios land in a future update.