agent-treasures-public

Rewrite user's original prompt by integrating their critique of the agent's failed response.

Deconstruct LLM "neuroslop" — machine clichés, puffery, weasel-wording, Markdown abuse, rhythmic monotony. Use when user asks to humanize, dehumanize-AI, очистить от нейрослопа, депорезать, redact AI-generated copy, edit a draft from ChatGPT/Claude/Gemini, or refine corporate/PR/wiki text that "sounds AI".

Discover and merge useful features from sibling forks into the user's fork. Invoke only on explicit user request via /fork-merge with a fork URL.

Git workflow patterns — branching, committing, rebasing, conflict resolution. Use when working with git operations.

Relentlessly interview user to stress-test a plan/design.

Generate conversation handoff document for the next agent session.

Generate a comprehensive CLAUDE.md for the current repository — first runs Anthropic's /init for a baseline, then restructures with an extended standardized template (sections, security flows, anti-patterns, data models, CI/CD, common issues). Use when a repo has no CLAUDE.md or you want to enrich an existing one with extended structure.

Practical playbook for maximizing prompt cache hit rate and minimizing token spend on LLM/agentic systems (Anthropic, OpenAI, Google Gemini, self-hosted vLLM). Covers prompt structure rules, multi-turn cache breakpoint placement, tool definition stability, dynamic content handling, TTL choice, multi-tenant isolation, and known anti-patterns.

Professional adaptive intake interview that converts a vague task into a structured spec before planning. Uses JTBD, 5 Whys, MoSCoW, and INVEST to produce a two-layer artifact (TL;DR brief + full spec).

Aggressively critique code/design for vulnerabilities, inefficiencies, and code smells. Triggered by "roast me".

Harness-agnostic post-session friction forensics. Run at (or near) the end of a session to audit THIS session for avoidable frictions that had a cheap alternative, and emit a ranked report mapping each friction to a concrete cheap harness lever (skill / MCP tool / permission / prompt-CLAUDE.md / tool-schema / memory / delegation-policy). Use when the user wants to know what the session cost in wasted turns/tokens and what cheap harness change would stop it recurring.

Analyze the current Claude Code session and generate a retrospective summary of lessons learned. Use when the user wants to: (1) reflect on what was accomplished in a session, (2) summarize lessons learned for future reference, (3) create a blog-post-friendly writeup of a coding session, (4) understand what went wrong and how it was fixed, or (5) document techniques discovered.

Close-out checklist for a non-trivial task — run BEFORE declaring done / writing `result:`. Verifies the project's continuity docs, READMEs, and cross-session trackers actually match the work just landed, and persists durable facts to memory. Use when finishing a coding session, wrapping up, closing out a task/project, or when asked "did you update the docs". Skip for trivial one-line edits.

Create new agent skills with proper structure, progressive disclosure, and bundled resources. Use when: designing a skill, writing SKILL.md, building a skill template, or formalizing a recurring workflow into a reusable skill.

Provide broader context, high-level perspective, or architectural map. Triggered by requests to zoom out, big picture, or unfamiliar code.