Use this skill to dispatch N parallel managed agent sessions (fire-and-forget), then gather results after they complete. For local dev workstations only — laptops where sessions are interruptible (lid close, Wi-Fi drop). Not useful for cloud runners or CI which have persistent infrastructure. The skill owns infrastructure only — session creation, ID persistence, polling, result extraction, and reporting. The caller owns prompt content, output shape, and merge logic. Trigger for prompts like 'dispatch research on X', 'run these topics in parallel', 'fire off N sessions', 'gather results', 'check on my dispatch run', 'dispatch these to the cloud', or when the user invokes `/dispatch`. Also trigger when another skill needs to fan out work to managed agents.
Auto-log working time to the Work Log Google Calendar. Three modes: (1) collect — silent heartbeat recording via UserPromptSubmit hook, (2) sync — flush pending heartbeats across all projects to Calendar (run in a dedicated session), (3) review — show all unreviewed entries for approval. Heartbeats are never purged until the user explicitly approves them. On first use per project, backpopulates 30 days of history from git log (HITL). The skill only writes to the dedicated Work Log calendar, never to any other calendar.
Use this skill when the user wants to diagnose an error, investigate a failure, or root-cause a bug — without necessarily fixing it right away. Trigger for prompts like 'troubleshoot X', 'why is Y failing', 'debug this error', 'investigate LIN-123', 'figure out what's wrong with Z', 'look into this CI failure', 'root-cause this crash', or when the user invokes `/troubleshoot`. The skill is a walk-away investigator: it runs read-only commands, tees output to files, investigates in parallel via subagents, formulates a root-cause hypothesis, and confirms it with a minimal reproducible test — all with minimal interrupts so the operator can step away and return to a finished report. Findings are posted as Linear comments or design note updates depending on context. If a fix is needed after the diagnosis, the skill hands off to /pair. Do NOT trigger when the user wants to immediately implement a fix (use /pair), when pure web research suffices (use /qsearch), or when the issue is a design decision rather than a fai
Use this skill when the user asks you to quickly research, investigate, look into, find out about, compare, summarize, or get up-to-speed on a topic that benefits from fresh information from the web. Trigger for prompts like 'research X', 'qsearch Y', 'what are the best Y for Z', 'compare A vs B', 'find me sources on...', 'is it true that...', 'what's the current state of...', 'dig into...', or any question where a good answer requires fetching and synthesizing web content. Also trigger when the user invokes `/qsearch`. This skill batches any necessary clarifying questions into ONE up-front round before executing, so the user can answer them, walk away, and return to a finished artifact instead of being pinged for follow-ups throughout. Do NOT trigger for narrow factual questions answerable from training data, for code lookups better served by Grep/Glob, when the user provides specific URLs to fetch (just fetch them), or when the user explicitly says 'no web search' / 'from what you know'.
Use this skill when the user wants to fix spelling, grammar, punctuation, or typos in a piece of text while preserving their personal voice and style. Trigger when the user pastes text and asks things like 'untypo', 'fix this', 'proofread', 'autocorrect', 'clean up grammar', or submits a draft (email, message, comment, commit message, social post) with visible errors and asks for a cleanup. Also trigger when the user invokes `/untypo`. Do NOT rewrite for clarity, shorten, restructure, change tone, translate, or 'improve' the writing — only fix objective mistakes. Do NOT trigger when the user explicitly asks for editing, rewriting, summarizing, translating, copyediting for style, or stylistic feedback — those are different tasks.