ElephantHunters
UserA personal library of Claude workflow skills for builders who orchestrate AI tools
Categories
Indexed Skills (7)
adversarial-code-review
Use this skill whenever Naveen has completed a build and wants the code reviewed before shipping. Triggers when Naveen types /adversarial-code-review, /code-review, or /code_critique, says "review this code", "check this before I ship", "what did you miss", or pastes code and asks for a critique. This skill prevents the class of problems where code that looks complete has subtle bugs, edge cases, or assumptions that break in production. Always apply after any build session before going live. Claude reviews its own code adversarially — or provides the exact prompt for a second AI to do so.
build
Use this skill whenever Naveen starts a new build, project, or feature request. Triggers when Naveen types /build, says "build me", "create", "make", "I want to build", or describes something they want to ship. This skill prevents the class of problems where Claude jumps straight to implementation without understanding the problem, choosing the right approach, or planning phases. Always apply this skill before writing any code. Research and planning first — implementation last.
context-loop
Use this skill to manage session continuity across Claude conversations using AGENTS.md as a persistent project memory file. Triggers whenever Naveen starts a new session and pastes AGENTS.md, types /context-loop, says "let's pick up where we left off", "here's my AGENTS.md", "briefing", or asks Claude to capture session learnings at the end of a conversation. Also triggers at session end when Naveen types /learnings. This skill prevents the class of problems where Claude loses all project context between sessions and Naveen has to re-explain everything from scratch. Always apply this skill at the start AND end of every project session — it is the save-and-restore system for all builds.
harness-first
Use this skill whenever Naveen is starting a new coding project, script, or automation. Triggers when Naveen types /harness-first, says "start a new project", "scaffold this", "set up the structure", or begins any build where code will be written across multiple sessions. This skill prevents the class of problems where builds lack a testing structure, making it impossible to verify that individual components work before wiring them together. Always apply at the start of Phase 4 (Implement) in the /build skill. Harness-first is how Naveen's code stays reliable across multi-session builds.
jargon-jail-break
Use this skill whenever the user pastes technical text, code snippets, error messages, architecture notes, or developer jargon and wants it explained in simple, plain language. Also trigger when the user says things like "what does this mean?", "break this down for me", "I don't understand this", "explain this like I'm not a developer", "translate this", "what is [technical term]?", or pastes a block of text with words like DOM, raster, vector, pagination, cron job, CNAME, DNS, tier, cold-start, API, SDK, JSON, dependency, deploy, cache, async, webhook, namespace, transpile, compile, middleware, scaffold, or similar. Always trigger proactively if the user has indicated they are not from a CS/engineering background.
really-honest
Use this skill whenever Naveen wants Claude to re-examine its own previous response with genuine scrutiny. Triggers when Naveen types /really-honest, says "but for real", "are you actually sure", "what did you miss", "what's the weakest part of what you just said", or "be honest about this". This skill prevents the class of problems where Claude's first response is polished and complete-sounding but contains glossed-over assumptions, overconfident claims, or subtle errors. Apply after any high-stakes response, strategy output, plan, or technical recommendation. Sits alongside /honest in the Thinking & Reasoning shortcode category — but operates differently.
silent-fail-check
Use this skill ONLY when Naveen explicitly types /silent-fail-check after a file has been generated. Do NOT trigger automatically after every file generation. This is a manual, on-demand verification skill — not an auto-firing protocol. When triggered, verify the most recently generated file: check file existence, size sanity, file type validation, and content extraction. This skill prevents the class of bugs where a script exits with returncode 0, prints a success message, but produces an empty, corrupt, or near-empty file because a library silently ignored invalid parameters or failed to write content. Trigger: /silent-fail-check — and only this. Never auto-apply.
Bio shown is the top-scored skill's repo description as a fallback — real GitHub bios land in a future update.