dotfiles

Onboard one or more job applications from role URLs, pasted adverts, or explicitly identified browser pages into Flo's Obsidian application dashboard. Use when Flo asks to evaluate, rank, inspect, triage, or add job opportunities or applications as dashboard candidates; reconstruct current goals and evidence from the vault and current CV, inspect application forms without submitting, then create or safely update each candidate record. Do not use for a Welcome to the Jungle recommendation review whose requested output is only the daily journal; use review-wttj-recommendations instead.

Review Flo's Welcome to the Jungle job recommendations in his logged-in Chrome or Vivaldi session, rank them for genuine senior AI-engineering fit, flag existing application-dashboard duplicates, and capture the review in today's Obsidian journal. Use when Flo asks to scan, review, shortlist, rank, journal, or revisit Welcome to the Jungle recommendations without onboarding them as applications.

Produce a short, visual, evidence-backed health report for Flo's chezmoi-managed dotfiles from any working directory. Use only when explicitly invoked as $chezmoi-status or when the user unmistakably asks for the state, status, or health of chezmoi or the dotfiles repository. Do not trigger for a generic project-status or Git-status request.

Inspect and organize ambiguous, heterogeneous, or multi-agent working trees into coherent thematic commits. Use when the commit scope is unclear, changes may contain multiple themes, staging needs review, work has accumulated over time, or the user explicitly asks to triage or split changes. Do not use for a straightforward commit when the conversation already establishes one clear, cohesive scope; commit those changes directly.

Legacy adapter that prepares a WalkAndLearn LangGraph input file and UI state JSON from a conversation copied to the macOS clipboard. Use only when the user explicitly asks for the legacy LangGraph workflow, LangGraph Studio/UI, or a WalkAndLearn graph state block. Do not use for Codex-native WalkAndLearn summarization; use generate-walkandlearn-summary instead.

Generate, correctness-check, compare, surgically repair, and fidelity-audit three emotional and three technical WalkAndLearn summary candidates from a macOS clipboard conversation or an absolute transcript file, then publish the accepted evaluated set to the Obsidian WalkAndLearn DebugSandbox. Use when the user invokes $generate-walkandlearn-summary, asks for a Codex-native WalkAndLearn summary, wants technical claims checked against primary sources, wants failed winners patched instead of regenerated, or wants to resume a preserved WalkAndLearn run ID.

Create or update explanatory Obsidian notes in Flo's agent sandbox. Use when the user asks to create a sandbox file, agent handbook note, explanation note, visual explainer, or communication artifact in the agent sandbox; when they mention the Agent Sandbox, sandbox folder, INDEX.md, or helping agents understand a project/concept.

Update Flo's Structured daily planner from natural language. Use when the user asks to update, reorganize, clean up, reschedule, color-code, icon-code, split, log, or reconcile tasks/events/time blocks in Structured, especially today's plan, including messy voice-transcribed updates, quarter-hour scheduling, overlap handling, recurring task instances, travel/appointment splitting, and Structured built-in icons/colors.

Triage and onboard new agent assets discovered in watched Claude, Codex, or Agents folders. Use when watch-dirs reports unmanaged files, when the user asks to onboard a skill/hook/agent asset, or when migrating a Claude/Codex skill into shared `.agents` ownership without letting the agent decide ownership silently.

Explicitly invoked workflow for migrating a clean repository's Claude Code instructions, rules, hooks, skills, commands, subagents, MCP config, and plugin-adjacent assets into shared `.agents` assets plus Claude Code and Codex adapters. Use when the user asks to migrate Claude Code repo assets, reduce duplication with symlinks, preserve Git history, or make Claude/Codex agent assets consistent.

Rewrite markdown documents to match Flo's documentation style — scannable, outline-dense, callout-rich, emoji-semantic. Use this whenever the user says "apply my style", "style this", "make it look like mine", "format this doc in my style", or wants a markdown document reformatted to their personal conventions. Also trigger when the user references their "style guide" or "style reference" in the context of reformatting a document. Even if the user just says "now make it pretty" or "clean up the formatting" after drafting content, this is likely the right skill.

Fetch the current weather in London and recommend what to wear. Use this whenever the user asks about London weather, what to wear in London, whether they need a coat or umbrella in London, or is heading out in London and wants a clothing recommendation — even if they don't explicitly say "weather skill."

Create cheat sheets that match Flo's preferred learning style — practical, opinionated, problem-first documentation with real-world examples and honest tradeoff assessments. Use this skill whenever the user asks for a cheat sheet, reference guide, quick guide, learning document, or says things like "write me a cheat sheet on X", "create a guide for Y", "help me understand Z", "document how to use X". Also trigger when the user asks to turn a conversation or learning session into a reusable document. Always use this skill for any documentation or cheat sheet creation, even if the topic seems simple.

**Multi-Agent Exploration & Visualization Pipeline**: Spawns multiple brainstorming agents with different perspectives, synthesizes their findings into a comprehensive Markdown reference doc, then builds a stunning interactive HTML showcase. Optionally writes utility code and updates project instructions. - MANDATORY TRIGGERS: explore and present, deep dive, brainstorm, research and visualize, multi-perspective analysis, explore all angles, comprehensive analysis, "what can we do with", research this thoroughly - Also use when the user asks to analyze a dataset/codebase/system from multiple angles, wants to understand all possibilities before making a decision, or says things like "I want to really understand X", "explore everything about Y", "give me the full picture on Z" - Use this skill even for requests like "think about this from different angles" or "I want creative ideas about X" — any time multiple perspectives would add value and the user wants a polished deliverable, not just a chat response.

Audit Claude Code sessions for the AskUserQuestion silent auto-complete bug where empty answers are hallucinated as real selections. Triggers on "audit ask questions", "empty answer bug", "check phantom answers", "session audit", "did I actually answer", "transcript audit", "check for hallucinated answers", "AskUserQuestion bug".

Spawn a diverse review panel (junior dev, staff engineer, product owner, end user) to pressure-test specs, architecture docs, API designs, or any planning document before implementation. Each reviewer runs as an isolated sub-agent to ensure fresh, unbiased perspectives. Use whenever the user wants feedback on a document — triggers on 'review this', 'what am I missing', 'poke holes in this', 'is this ready to implement', 'review panel', 'spec review', 'get feedback on this', 'sanity check this design', 'fresh eyes on this', 'critique this', or when presenting a planning document and asking for input. Also use when the user mentions wanting multiple perspectives, asking 'what would a junior/senior/PO/user think', or says things like 'run the panel on this'. Even if the user doesn't explicitly name the skill, trigger it when they share a spec or design document and ask for review or feedback.

Interactive zettel ideation partner for Flo's Obsidian Zettelkasten. Helps unblock the thinking process before writing zettels — discovers vault connections via graph traversal, helps decide if something is zettel-worthy, catches split points, and plays back outlines from conversation. Use this skill whenever the user wants to write a zettel, has an idea they want to develop, says things like "I want to write a zettel about X", "is this zettel-worthy?", "help me think through this idea", "turn this into a zettel", or shares a raw thought and wants help shaping it. Also trigger when the user shares an insight and wonders whether it belongs in their Zettelkasten. Even if they just say "that's a good idea, I should write that down" — this is likely the right skill.

Clean up stale git worktrees and their associated branches. Checks merge status, detects rebased work on main, flags uncommitted changes, and only deletes what's safe. Use this skill whenever the user mentions worktrees, stale branches, cleaning up branches, or worktree cleanup. Trigger on: 'clean worktrees', 'delete worktrees', 'stale worktrees', 'worktree cleanup', 'clean up branches', 'remove worktrees'.

Run project linting, formatting, and type checking, then auto-fix all mechanical issues and present structural ones for user decision. Never suppresses warnings — always fixes root causes. Use this skill whenever the user mentions linting, formatting, type errors, code cleanup, preparing for commit, or finishing implementation work. Trigger on: 'lint', 'fix lint', 'lint-fix', 'clean up the code', 'run checks', 'fix errors', 'make it pass lint', or after implementing features/phases. Even if the user just says 'check the code' or 'is this clean?', use this skill.

Read a spec/design file and interview the user in depth using AskUserQuestion to surface gaps, tradeoffs, edge cases, and implicit assumptions — then write the deepened spec back to the file. Use when the user points at a spec, design doc, RFC, milestone brief, or feature sketch and wants it fleshed out via questioning. Triggers on 'interview me on this spec', 'deepen this spec', 'flesh out this spec', 'ask me questions about X', 'dig into this spec', 'spec interview', 'interview me about this', or any request to be interrogated about a document until it feels complete. Also trigger when the user hands you a thin/skeletal spec and asks what's missing — the answer is to interview, not to guess.