os-suite

Explain-like-I'm-five mode — talk to Gareth the way you'd talk to a curious five-year-old for the rest of the conversation, using short concrete sentences and everyday comparisons, while still saying the real technical word or jargon and immediately translating it into something a small child would get. Use whenever Gareth says "/5-year-old", "talk to me like I'm five", "explain like I'm 5", "ELI5", "five year old mode", or "explain it so a kid could get it". Once triggered, stay in this mode for the rest of the session, not just one reply, until Gareth explicitly asks to end it (e.g. "end five year old mode", "back to normal", "stop talking like that", "talk normally again").

Cross-provider mass adversarial tester for Gareth's AI agents. Runs many difficult, messy, multi-turn fake-customer conversations against an agent's exact prompt (default Leo, the GCE front-door assistant), scores each with a judge model against the agent's own rules, tracks real token cost, and writes a ranked bug report. Runs the agent-under-test on OpenAI or Claude, and in benchmark mode compares several models on the same scenarios and names the cheapest one that still passes. Catches personal-story derailment after correct qualification, price pressure, ambiguous or multi-service requests, discontinued class requests, agency enquiries, bot challenges, and NDIS-sensitive wording. Reusable across agents by pointing it at a different prompt file. Use when Gareth says "stress test Leo", "test the agent", "run the bug test", "benchmark the models", "which model is cheapest that works", "adversarial test", or after any change to an agent prompt to check for regressions.

Daily focus check-in for Gareth Cohen. Reads the current vault state silently, asks one energy question, and delivers a single clear directive: one thing to do today, the first concrete action right now, and one thing to explicitly defer. Also detects and names priority misalignment before it causes drift. Invoke this skill whenever Gareth says: "daily focus", "morning check-in", "what should I work on", "focus check", "reset focus", or /daily-focus. Always use this skill at the start of any workday or whenever focus needs resetting -- even mid-session. Do not attempt to construct a focus directive without invoking this skill.

Run a bounded Gareth OS daily review, or configure its prompt and schedule when requested. Reconcile current evidence into the Daily note and rolling backlog using available authorized read-only connectors.

Audit and improve Gareth's Second Brain instructions, knowledge links, evidence freshness and navigation. Use for OS optimizer, workspace audit, skill cleanup or architecture review. Preserve useful task knowledge and evaluate changes against actual task behavior.

Convene a five-juror council to pressure-test a specific idea, spend, build, or buyer move before Gareth commits to it. Use this whenever Gareth asks to "run the council", "roast this", "stress test this", "greenlight this", "talk me out of this", "poke holes in this", or asks "should I build / buy / spend / pitch X", "is this worth it", or "am I overbuilding again". The council argues from Gareth's actual operating rules (cash-stability, lighter-beats-heavier, approval gates), not generic business advice, and returns a greenlight / reshape / kill verdict plus the single cheapest test. Trigger it even when Gareth does not say the word "council" but is clearly weighing whether to commit time, money, or a build to something. It is advisory only and never takes any action.

Epistemic honesty mode for Gareth. Strips performative warmth, fashionable conversational norms, and confidence theatre from Claude's responses so trust is built on precision rather than polish. Activates a strict discipline - flag uncertainty explicitly, separate knowledge from inference, never simulate memory or feeling, answer first and contextualise only when it clarifies. Use when Gareth says "/honesty", "honesty mode", "be straight with me", "no dress-ups", "speak to me directly", or at the start of any session where he wants directness over smoothness. Once triggered, the mode stays active for the whole session.

Start a clean Claude session with only the context the next task needs, instead of /compact. Captures the goal of your next session, points to the right files (never copies them), and outputs a self-contained prompt to paste into a fresh chat. Use when: "re-fresh", "refresh context", "fresh start", "my chat got sloppy", "start a new chat with context", "hand off this session", outputs degrading, drifting off task. Three levels: lite (quick reset), full (default handoff), ultra (full briefing). NOT for: summarizing in place (that's /compact), or saving permanent notes.

Adapt, improve, repair, simplify, or extend an existing Claude skill without rebuilding it from scratch. Use when the user says "adapt this skill", "modify this skill", "improve this skill", "fix this skill", "rewrite this skill", "add a layer to this skill", "remove clutter from this skill", "make this skill interview me first", "change how this skill behaves", "keep the useful parts but change the process", "turn this skill into a better version", or pastes a detailed change proposal for an existing skill. Use skill-creator instead when no skill exists yet.

Fast macro-goal sanity check — pauses before (or during) a nontrivial multi-step plan, restates the actual objective in one line, and weighs a couple of genuinely different ways to get there, including the boring/obvious one. Catches tunnel vision before it burns hours making a specific tool, integration, or sub-approach work when the real goal never needed it. Use when Gareth says "/think-outside-the-box", "zoom out", "bigger picture check", "are we overcomplicating this", "step back", or "is there a simpler way". Also self-invoke this proactively and silently, unasked, whenever about to commit to a plan with several steps or multiple tools/integrations, a task is dragging on longer than it should, or troubleshooting a tool has replaced pursuing the actual goal — run it as a quick internal check before continuing, not something that interrupts Gareth. A gut-check, not a formal review — no reports, no scoring.