Proactively log small friction points, flaky commands, failed tool calls, or setup gotchas to `PAPERCUTS.md` using `yarn papercut`. Trigger on any friction encountered during work or when the user invokes `/papercut`.
Turn real work the user has actually done into a Reddit post for technical, developer-centric, or AI-related subreddits. Use whenever the user says "write me a Reddit post", "help me post on Reddit", "turn this into a Reddit post", "share this on Reddit", or anything similar — and also proactively suggest this skill when the user just finished something post-worthy (a benchmark, a workflow they figured out, a bug/incident diagnosis, a tool/framework comparison, a meaningful failure). The skill REFUSES to generate posts from nothing — it requires real work, real numbers, real experience as the input. Karma-farming generic listicles are explicitly out of scope.
Activates an interactive interview mode to resolve design decisions, clarify vague requirements, and align on a plan. Trigger on "/align", "align", "interview me", "let's align", or when the user has an ambiguous plan and needs to be questioned to solidify it.
Run a user-given shell command directly and return raw output, no commentary.
Triggers on error messages, stack traces, "this is broken", unexpected output, or debugging requests. Focuses on root cause analysis, ranked hypotheses, and targeted fixes.
Execution mode. Trigger on "go ahead", "do it", "execute", "implement", "make the changes", or any approval to act. Stop planning, start building.
Multi-model analysis and synthesis using OpenCode CLI — replicates OpenRouter Fusion style workflows locally. Runs multiple LLMs in parallel, judges their outputs, and produces a single synthesized response. Trigger whenever the user wants to run a task against multiple models, compare model outputs, get a consensus answer, or says things like "fusion run", "multi-model", "run all models", "synthesize model outputs", or "parallel agents". Also trigger when the user wants to reduce hallucinations by cross-checking multiple LLMs, or when high-stakes tasks need more than one model's perspective.
Activates a read-only planning mode where the assistant observes, reads, and reports — but never edits, creates, or modifies anything. Use this skill whenever the user says "planning mode on", "plan:", "let's plan", "just plan this", "don't touch anything, just look", or any similar phrase indicating they want analysis without action. Also triggers for task planning requests where the user wants to understand what exists before deciding what to do. In planning mode, files and context are read, what's there and what's missing is reported, and next steps are proposed — but zero action is taken. Stays active until the user says "planning mode off", "exit planning mode", or "done planning". This skill must be used any time the user signals they want observation-only mode, even if phrased casually.