Review implementation plans and specs for design quality before coding. Use when a plan needs UI/UX critique, missing state coverage, AI-slop detection, responsive/accessibility checks, or unresolved design decisions.

Product discovery via YC-style forcing questions and 10-star product thinking. Use when starting a new feature, evaluating a product idea, or reframing a request into its most ambitious version.

Run synthetic user research sessions natively — no backend required. The agent plays an AI-generated persona and simulates a user interview based on the 4 Ps framework (Persona, Problem, Promise, Product). Use when a user wants to run a user research session, interview a synthetic persona, validate product ideas, generate user personas, or simulate customer conversations. Triggers on "user research", "synthetic persona", "simulate a user", "userstudies", "interview a persona", "validate my idea", or "talk to a user".

Startup advice frameworks from Sahil Lavingia (Gumroad) based on The Minimalist Entrepreneur. Use when someone is starting a business, validating an idea, finding customers, setting prices, building an MVP, creating a marketing plan, defining company values, or making any business decision through a minimalist lens. Also use for 'minimalist entrepreneur', 'sahil', 'gumroad advice', 'bootstrap business'.

Lightweight brainstorm — 5 versions with estimated probabilities + assumption surfacing. Use when: brainstorm this, give me options, what are my options, quick brainstorm, /quick-brainstorm.

Plan mode for Hermes Agent — inspect context, write a markdown plan into the active workspace's `.hermes/plans/` directory, and do not execute the work.

Run a multi-perspective council analysis on any question, plan, or decision. Spawns parallel cognitive perspectives (Architect, Skeptic, Pragmatist, Innovator, User Advocate, Temporal Analyst) and synthesizes via structured dialectical analysis. Use when: making a significant decision, stress-testing a plan, red-teaming an idea, exploring alternatives, or needing a "devil's advocate" on something. Trigger phrases include "another perspective", "council analysis", "stress test this", "red team this", "poke holes in this", "multiple viewpoints", "what am I missing", "challenge this idea".

How to clean up duplicate/untracked skills in the skills repo

Reference for the two-tier skill loading system — preloaded vs category-gated skills, how to add/remove preloaded skills, and the external-services category layout.

Eval-driven skill optimizer: runs a skill repeatedly, scores outputs against binary evals, mutates the prompt via structured edits, and keeps only changes that improve a held-out validation score (cross-model optimizer/target, three-way splits, golden cases, checkpoint resume). Use when: optimize/improve this skill, make this skill better, run autoresearch on, self-improve skill, benchmark/eval my skill, run evals on.

Create new skills, modify and improve existing skills, and measure skill performance. Use when creating or editing ANY skill — including rewrites, description tweaks, or adding examples. TDD pre-flight required: write failing test first (3+ pressure scenarios run WITHOUT the skill) before touching SKILL.md. Also use for running evals, benchmarking skill performance with variance analysis, and optimizing skill descriptions for better triggering accuracy.

Audit and score any skill against best practices. Use when: audit this skill, review this skill, check this skill, score this skill, is this skill good, skill health check, skill review, rate this skill. Takes a skill directory path, evaluates structure/content/patterns against a checklist, and outputs a scorecard with specific fixes.

Iteratively optimize an AI agent's system prompt using a keep/revert autoresearch loop. Use when asked to optimize, improve, or tune a system prompt, run prompt optimization, or do a Karpathy-style autoresearch loop on an agent prompt. Also use when asked to run evals and improve the prompt based on results.

Use the mcporter CLI to list, configure, auth, and call MCP servers/tools directly (HTTP or stdio), including ad-hoc servers, config edits, and CLI/type generation.

Autonomously improve AI agent skills through eval-driven mutation loops. Use when asked to "improve a skill", "optimize agent performance", "evolve a skill", "run an improvement loop", "make this skill better", "optimize this prompt", "self-improve", "mutation loop", "eval-driven optimization", or "GEPA-style evolution". Also use when diagnosing why a skill underperforms or when building eval datasets from real session history. Works against any agent type: ADK, LangChain, CrewAI, AutoGen, custom Python, HTTP API, or CLI agents.

Retrieve memory from past sessions. Use whenever the user asks about past conversations, prior decisions, "what did we do about X", "when did we last...", "do you remember...", or anything where the answer might live outside the current session's context. Uses progressive disclosure across six tiers (hot memory → dated episodes → raw sessions) with an uncertainty gate — stops expanding context when confidence plateaus.

Set up a persistent memory system for any AI coding agent (Claude Code, OpenCode, Assistant Runtime, OpenClaw, etc.). Covers workspace files, memory entry format, session-end extraction, daily garbage collection, multi-tier recall, and optional vector search. Use when someone wants their agent to remember things across sessions, set up memory, or configure persistent context.

Daily memory garbage collection for MEMORY.md / USER.md. Apply decay rules, drain .pending.md, consolidate near-duplicates, maintain canonical theme tags, prune old episode and session files. Invoke when asked to "run memory GC", "clean up memory", "apply memory decay", or from the scheduled cron job.

Use for Fastlane AI at usefastlane.ai, the short-form content platform and REST API for generating, remixing, scheduling, and analyzing organic TikTok, Instagram Reels, YouTube Shorts, and related content. Use when an agent needs to plan or automate Fastlane campaigns, call the Fastlane API, create content angles, tune Blitz preferences, generate content, schedule/cancel posts, inspect analytics, design account warmup/posting strategy, or build reusable prompt/workflow files for Fastlane. Do not use for fastlane.tools mobile app deployment automation.

Use when creating, scheduling, or managing social posts via Typefully.

Use the Stably CLI to create, run, fix, and maintain Playwright tests in the bloom-tests repo. Use when running tests (stably test), auto-fixing failures (stably fix), or generating new tests from a prompt (stably create).

Manage Render.com services, deploys, databases, logs, and infrastructure using the official Render CLI (`render`). Use this skill whenever the user asks about Render deployments, service management, viewing Render logs, restarting Render services, running database queries on Render Postgres, triggering deploys, managing environments/projects, or anything related to their Render infrastructure. Also use when the user mentions render.com, Render dashboard actions, or wants to automate Render workflows.

Search TikTok viral videos, App Store rankings, hook analysis, app strategy, and content research via SGE Prometheus MCP / Passistant. Requires SGE_API_KEY. Use when researching viral video hooks, benchmarking app rankings, analyzing content trends, pulling App Store review data, or when a prompt mentions "prometheus", "Passistant", "SGE", "Social Growth Engineers", "viral content", "app store rankings", "TikTok research", or "hook analysis".

Manage Porkbun domains, DNS records, SSL certificates, URL forwarding, and hosting blueprints via the Porkbun API. Use when the user asks about domain management, DNS, SSL certs, URL redirects, or connecting a domain to a hosting provider.