ketchup

• SKILL.mdcore · 8,480 bytes
• agents/openai.yamlasset · 43 bytes

# Ketchup

Invoke manually as `/pstack:ketchup` in Claude Code, or select `pstack:ketchup` or mention `$pstack:ketchup` in Codex. Do not load it implicitly.

Catch the user up on everything in this chat since the last actual human-authored message before this invocation. Exclude the invocation itself, system messages, heartbeat and scheduler prompts, automated notifications, and subagent results when choosing that anchor, even if the host presents them as user-role events. Automated events after the anchor still belong in the catch-up. If the true anchor or intervening activity cannot be established, state the gap and use only confirmed current-chat context. Do not search unrelated chats. Assume the user remembers their last message and nothing after it. Help them understand each change, not only where it stands. Then lay out every call waiting on them so they can make it from your reply alone.

Use [`/bro`](../bro/SKILL.md)'s voice, not its task: plain words, short sentences, no jargon, like one person catching up another. Say what changed for the user, not how you did it. Say "the tests pass", not "CI is green". Leave out hashes, file paths, commands, build numbers, and test counts unless the user needs one to act. Call each PR, branch, or agent by a name the user would know, not by a label you invented. If a fact needs context to make sense, give the context or leave the fact out.

## Gather what changed

Keep this turn read-only. Collect your own work since that message, reports from subagents, and messages from anyone else in this chat. Check the current state of each PR, CI run, or agent this chat touched through existing CLI, API, or connector readers. Inspect a UI only under explicit user authorization for the current task and scope. General task authorization or an app mention does not grant it. If status requires unauthorized UI or an unavailable reader, name the concrete limitation and leave that status unchecked. Say which tasks are still running.

Treat earlier agent prose as claims, not verified current state. Check it against available artifacts and human instructions before repeating its implications. Distinguish verified implemented behavior, proposed options, and unknowns. A proposed parameter does not prove its mechanism exists. Operational bounds, costs, guarantees, and causal rationale require evidence for the underlying mechanism and supported assumptions; arithmetic from option labels alone is not that evidence. Before writing a comparison, identify the mechanism in the inspected artifact. An admission predicate is not a cache store, and a proposed expiry setting does not establish a freshness bound or lower backend load. If that mechanism is absent, label each option as a proposed policy with unknown effects. Check every visual label against this distinction before sending; a shorter label must not turn a proposal into a guarantee.