Andersonlimahw

Recursive error-fix loop. Use when a bash command fails with a build/test/typecheck error, when the same class of error appears more than once, or when the harness itself needs a persistent rule so the error does not come back. Detect → root-cause → minimal fix → test → lint/typecheck → persist rule (SKILL.md / AGENTS.md / docs/rules) → audit with evidence. Cross-harness (Claude Code, OpenCode, Codex, Gemini, Agy).

Wraps any external CLI so the AI harness can invoke it without context bloat. Captures --help output, documents every subcommand/flag in a compact digest, saves token metrics (raw vs wrapped), and exposes a standardized interface (invoke, explain, audit). Use before calling an unfamiliar CLI or when the harness needs to interact with a CLI repeatedly.

One-shot setup that configures the CLI wrapper ecosystem across ALL AI harnesses (Claude Code, Gemini, Codex, Agy, OpenCode). Creates symlinks from Claude (single source of truth) to every other CLI, registers wrapper skills/agents in each harness, and injects the "prefer wrapped CLI" policy into each CLI's config file. Run once per machine or after adding a new CLI to the ecosystem.

Cross-harness skill that teaches Claude Code, Codex (OpenAI), and Antigravity (Agy) how to use OpenCode as a subagent. Covers delegation triggers, task formatting, provider routing, result handling, and the complete protocol for efficient subagent orchestration.

Automatically routes tasks to the optimal AI agent, model, or provider based on complexity, cost, and capability. Use when implementing features, fixing bugs, or any multi-step development work. Triggers on "implement", "build", "create", "fix", "add feature", "develop", or when the user asks to do any coding task.

Surface opportunities to deepen a codebase. Finds modules that are shallow (thin pass-throughs, anemic types, missing domain modeling) and proposes how to make them deep (rich domain types, co-located behavior, testable seams). Outputs a visual HTML report and then runs an interactive grilling loop that pressure-tests each opportunity. Use when the codebase feels "flat", domain logic leaks into controllers/UI, types carry no behavior, or AI agents struggle to navigate the module boundaries. Integrates with domain-modeling, codebase-design, and ADR workflows.

Loop de diagnóstico para bugs difíceis e regressões de performance. Feedback loop → reprodução → hipóteses → instrumentação → fix + teste → cleanup. Use quando o usuário disser "diagnosticar"/"debugar isso", ou relatar algo quebrado/lançando exceção/falhando/lento.

Interrogatório implacável para planos e designs que cria documentação (CONTEXT.md/glossário, ADRs) durante o processo, enquanto grilla. Combina grilling de decisão com domain-modeling — desafia a linguagem vaga contra o glossário existente, stress-testa cenários concretos, e captura termos e decisões inline conforme cristalizam. Use when user wants to stress-test a plan against their project's language and documented decisions, mentions "grill me", "grilling", or needs to sharpen terminology while deciding.

Visualize an agent's perceive→think→act loop. Inspects tool calls, latency per step, decision branches, retry patterns. Surfaces bottlenecks. Triggers on "trace agent", "why is agent slow", "agent loop analysis", "what did the agent do".

Compute context window budget for a prompt/agent — sums system prompt, tool descriptions, examples, conversation history; flags overruns; suggests cuts. Triggers on "context budget", "how much context", "will this fit", "budget my prompt".

Treat dataset / eval JSONL as code — lint, validate schema, detect duplicates, leakage between train/eval, label imbalance, PII. Karpathy "Software 2.0" lens. Triggers on "audit dataset", "lint jsonl", "check for leakage", "validate eval data".

Enforce "evals over vibes" — before any prompt, agent, or model change, require an eval set with success criteria and baseline measurement. Blocks vibe-only changes. Triggers on "tune prompt", "improve agent", "change model" without prior eval set.

Build a 10-30 case eval dataset from a PRD, skill, or prompt — covering golden path, edge cases, and adversarial inputs. Output as JSONL with input/expected/category. Triggers on "build eval set", "create evals", "generate test cases for prompt".

Break down any ML/AI concept to first principles — strip jargon, derive from basics, build physical/mathematical analogy. Karpathy intuition-building style. Triggers on "first principles", "explain like I'm 5", "derive from scratch", "why does X actually work".

Generate minimal, runnable from-scratch implementation of LLM concepts (attention, BPE tokenizer, backprop, RoPE, KV-cache, MoE, LoRA, RLHF). Karpathy-style pedagogy — code first, theory in comments. Triggers on "explain X from scratch", "minimal X implementation", "show me how X works in code", "build X from zero".

Add a new entry to the existing llm-wiki — following the project's template (concept → intuition → math → code → references). Maintains consistency across entries. Triggers on "add to llm-wiki", "new wiki entry", "document concept X".

Run the SpecLoop QA loop: load specs and business rules, enforce read-only safety, explore critical flows, write structured findings, triage accepted reports, and suggest regression tests.

Create, edit, or run SpecLoop scenario files under `.specloop/scenarios/`, including local localhost checks, remote landing/blog screenshot flows, and manual browser QA runbooks.

Run a repeatable, agent-driven workflow enforcing the 7 mandatory phases (Idea, Plan, Implement, Test, Gate, Ship, Reflect) to keep projects improving while maintaining SpecLoop quality standards.

Run a backend validation loop that generates Postman collections, executes RESTful tests, and validates API compliance using OpenAPI/Swagger specs or code routes.

Run an agentic testing loop for Mobile applications (SwiftUI, iOS, Android) that navigates the UI, analyzes accessibility, takes screenshots, and verifies visual specs.