Results for “rl”

Anthropic's method for training harmless AI through self-improvement. Two-phase approach - supervised learning with self-critique/revision, then RLAIF (RL from AI Feedback). Use for safety alignment, reducing harmful outputs without human labels. Powers Claude's safety system.

Read-only audit of app-layer auth — route×gate matrix, session lifecycle, OAuth, and provider traps (getSession vs getUser, middleware-as-only-gate, CVE-2025-29927). Use when "audit our auth", "check middleware protection". RLS → plan-rls-audit. OWASP checklist → audit-security.

OWASP Top 10 + Supabase-first hardening burndown. Use when "security audit plan", "OWASP audit", "hardening plan", or "security burndown". App-layer auth flows → audit-auth-flows. Table RLS → plan-rls-audit. Key rotation → plan-secrets-audit. App LLM attacks → audit-llm-security.

Audit a Supabase/Postgres project for Row-Level Security and access-control gaps, then produce a phased remediation plan. Use when "RLS", "is my Supabase secure", "anyone can read my data", "lock down my tables". App-layer session/route gates → audit-auth-flows.

Deep Supabase reference — auth flows, RLS patterns, edge functions, realtime, storage, Next.js SSR integration.

Supabase administration, RLS policies, migrations, and schema design. Use for database architecture, Row Level Security, performance tuning, auth integration. Activate on "Supabase", "RLS", "migration", "policy", "schema", "auth.uid()". NOT for Supabase Auth UI configuration (use dashboard), edge functions (use cloudflare-worker-dev), or general SQL without Supabase context.

Use when tuning RL reward function weights, fixing reward hacking, or addressing P&L gradient issues

Strategic patterns for solving intractable problems through cascading approximation, self-improvement, and heterogeneous evaluation from DeepMind's AlphaGo system

Debugging constant Markov regime features in RL observations - when HMM probabilities show uniform values instead of dynamic regime estimates

Use when implementing risk-adjusted rewards, discussing Sharpe ratio in RL training, or tuning reward components for risk awareness

Scales AI and Python applications across clusters with distributed computing primitives for ML workloads.

Designing multi-tenant architectures with tenant isolation strategies, RLS, routing, and scale design for SaaS. Use when designing multi-tenant SaaS systems or tenant isolation.

Build and manage Supabase projects including Database, Auth, Edge Functions, Realtime, Storage, Vectors, Cron, and Queues with security best practices.

Maintain durable session memory across agent disconnects by writing structured checkpoints to the repo's session directory, enabling context recovery.

Fix HOLD bias in RL reward function. Trigger when: (1) model learns to always HOLD, (2) trade rate is too low (<10%), (3) slippage penalty exceeds typical price moves.

Use this skill for auth boundaries, input validation, secrets, RLS, injection risks, destructive-action review. Trigger when the task involves security work related to Security Hardening, production implementation, audits, debugging, strategy, or validation.

Post-train LLMs with reinforcement learning using the slime framework, which integrates Megatron-LM for training and SGLang for rollout generation.

Add account state (P&L, win rate, drawdown) to RL observations + drawdown penalty in rewards. Trigger when: (1) model needs account awareness, (2) training should penalize drawdowns, (3) upgrading obs_dim 5300→5600.

Populate symbol database with crypto symbols before training. Trigger when: (1) live trader fails with 'CRYPTO VIOLATION', (2) no crypto models in models/rl_symbols/, (3) db.get_candidates(asset_types=['crypto']) returns 0, (4) starting fresh training without crypto.

Fine-tune LLMs using reinforcement learning with TRL - SFT for instruction tuning, DPO for preference alignment, PPO/GRPO for reward optimization, and reward model training. Use when need RLHF, align model with preferences, or train from human feedback. Works with HuggingFace Transformers.

Fine-tune LLMs using reinforcement learning with TRL - SFT for instruction tuning, DPO for preference alignment, PPO/GRPO for reward optimization, and reward model training. Use when need RLHF, align model with preferences, or train from human feedback. Works with HuggingFace Transformers.

Fine-tune LLMs using reinforcement learning with TRL - SFT for instruction tuning, DPO for preference alignment, PPO/GRPO for reward optimization, and reward model training. Use when need RLHF, align model with preferences, or train from human feedback. Works with HuggingFace Transformers.

Audit the working tree and git history for exposed credentials and mis-scoped keys, then a rotate-vs-relocate plan. Use when "hardcoded secrets", "did I commit a key", "secret scan", "is my .env safe", or "rotate keys". Do not fire for "RLS audit" or generic "security burndown".

Turn a refined research proposal or method idea into a detailed, claim-driven experiment roadmap. Use when the user asks for a detailed experiment plan, ablation matrix, evaluation protocol, run order, compute budget, or paper-ready validation that supports the core problem, novelty, simplicity, and any LLM / VLM / Diffusion / RL-based contribution.

Analyze betting market sentiment including line movement, sharp vs public money, steam moves, reverse line movement, and market signals. Use when reading line movement to determine where the smart money is going. Also trigger for 'line movement', 'sharp money', 'public money', 'steam move', 'reverse line movement', 'RLM', 'where the money is', or 'wiseguy action'.

Turn a refined research proposal or method idea into a detailed, claim-driven experiment roadmap. Use after `research-refine`, or when the user asks for a detailed experiment plan, ablation matrix, evaluation protocol, run order, compute budget, or paper-ready validation that supports the core problem, novelty, simplicity, and any LLM / VLM / Diffusion / RL-based contribution.

Use when doing ANY task involving Supabase. Triggers: Supabase products (Database, Auth, Edge Functions, Realtime, Storage, Vectors, Cron, Queues); client libraries and SSR integrations (supabase-js, @supabase/ssr) in Next.js, React, SvelteKit, Astro, Remix; auth issues (login, logout, sessions, JWT, cookies, getSession, getUser, getClaims, RLS); Supabase CLI or MCP server; schema changes, migrations, security audits, Postgres extensions (pg_graphql, pg_cron, pg_vector).

Supabase-first reference and implementation skill for Nick's app stack. Use when writing Supabase queries, auth flows, RLS policies, storage rules, realtime features, edge-function integrations, migrations, or debugging Supabase behavior. Prefer this instead of guessing Supabase APIs or patterns. Bias toward product-led implementation choices, trust-aware access control, MVP discipline, and using only the Supabase features the product actually needs.

Use when doing ANY task involving Supabase. Triggers: Supabase products (Database, Auth, Edge Functions, Realtime, Storage, Vectors, Cron, Queues); client libraries and SSR integrations (supabase-js, @supabase/ssr) in Next.js, React, SvelteKit, Astro, Remix; auth issues (login, logout, sessions, JWT, cookies, getSession, getUser, getClaims, RLS); Supabase CLI or MCP server; schema changes, migrations, security audits, Postgres extensions (pg_graphql, pg_cron, pg_vector).

Design and implement adaptive filters using DSP System Toolbox System objects. Use when working with adaptive filtering, system identification, noise cancellation, echo cancellation, active noise control (ANC), channel equalization, inverse system identification, or adaptive prediction. Covers dsp.LMSFilter, dsp.RLSFilter, dsp.FilteredXLMSFilter, dsp.FrequencyDomainAdaptiveFilter, dsp.AffineProjectionFilter, dsp.BlockLMSFilter, dsp.AdaptiveLatticeFilter, dsp.FastTransversalFilter, maxstep(), and algorithm selection for adaptive filtering problems. Replaces deprecated adaptfilt.* objects (removed R2020a).

Skilled agent architecture replacing multi-agent system for RL training. Trigger when: (1) planning agent-guided training, (2) implementing tool-augmented LLM consultations, (3) comparing skilled vs multi-agent approaches, (4) designing simulate-verify loops for training, (5) implementing prompt evolution / learnable parameters, (6) understanding Claude Agent SDK integration in training, (7) debugging SkilledTrainer consultations or tool calls, (8) configuring agent safety bounds for training actions.

Scales reinforcement learning with PufferLib — high-throughput parallel training (PuffeRL), vectorized environments, and native multi-agent systems achieving 2-10x speedups over standard implementations. Use when scaling RL to millions of steps per second, running vectorized or multi-agent setups, building custom PufferEnv tasks, or integrating game environments (Atari, Procgen, NetHack, PettingZoo). For standard single-agent algorithm implementations (PPO/SAC/DQN) or quick prototyping prefer alterlab-stable-baselines3. Part of the AlterLab Academic Skills suite.

Supabase-first database schema design for Nick's app stack. Generate production-ready Postgres schemas, Supabase SQL migrations, RLS policies, TypeScript types, seed data, role-aware access patterns, and schema checklists from plain English requirements. Use when starting any new project database, designing or reviewing schemas, adding tables, planning migrations, or turning product requirements into Supabase-ready data models. Bias toward product-led schema design, MVP discipline, core-workflow-first modeling, and avoiding premature table/role sprawl.

Audits codebases for common security vulnerabilities that AI coding assistants introduce in "vibe-coded" applications. Checks for exposed API keys, broken access control (Supabase RLS, Firebase rules), missing auth validation, client-side trust issues, insecure payment flows, and more. Use this skill whenever the user asks about security, wants a code review, mentions "vibe coding", or when you're writing or reviewing code that handles authentication, payments, database access, API keys, secrets, or user data — even if they don't explicitly mention security. Also trigger when the user says things like "is this safe?", "check my code", "audit this", "review for vulnerabilities", or "can someone hack this?".