session-viz

A backlog that goes both ways — describe something and it is filed with where it came from and why it matters; ask for work and it returns what to pick up next together with the criteria it ordered by. Local per project, or the shared team queue. Use when the user runs /qbl, wants to park something for later, asks what to work on next, or asks what is left.

Send this machine's delivery ledger to the shared reference as nine bounded columns — kind, task class, CLI band, ISO week, terminal state, delivery state, error class, and log2 buckets for output tokens and tool calls. No prompts, no paths, no repo names, no timestamps finer than a week. Use when the user runs /qcontrib, asks how to contribute findings, or the console says "Send your first findings".

Where the tokens actually go — cache-read versus generated output, and how much context each agent family replays per run. Use when the user runs /qcost, asks about token spend, why their bill is high, what an agent costs, or wants to reduce context usage.

Audit this repository's Claude Code configuration against how the user's other repos are set up — CLAUDE.md, slash commands, hooks, and whether permissions cover Write. Use when the user runs /qdoctor, asks whether a repo is set up properly, why headless or scheduled runs fail on permissions, or wants to compare configuration across projects.

File tasks into your team's queue from findings that passed a gate. Nothing is filed until you say so.

Analyse the current Claude Code session — prompting quality, friction metrics, and a derived intent breakdown carried forward from this project's earlier sessions — render it as an interactive HTML document opened in a preview window, and produce a copy-pasteable /compact instruction tuned to what the session was actually about. Use when the user runs /qpact, or asks to visualise, audit, or summarise the current session before compacting.

The delivery ledger for autonomous work — every scheduled run, subagent and workflow agent on this machine, with what it actually shipped. Shows which recurring tasks produce nothing, which runs died on a permission prompt, and how each one terminated. Use when the user runs /qruns, asks what their agents or cron jobs are doing, why a scheduled task is not producing output, or wants to see subagent activity.

Connect this machine to a session-viz workspace by signing in through the browser — the only way in; no token is ever typed, pasted or shown in the terminal.

Choose which projects, sessions or autonomous runs your team can see. Nothing is shared by default.

Turn prompts you keep retyping into slash commands. Pools repeated prompts across every session, separates rituals (a procedure you retype) from misses (a prompt that failed and got re-sent), and writes a .claude/commands file for the ones worth promoting. Use when the user runs /qship, mentions typing the same thing repeatedly, or asks what to turn into a command.

Work with the shared session-viz cloud — federated Obsidian vaults across projects and people, and task handoff between teammates. Resolve a [[wikilink]] against every vault you can see, find links that resolve nowhere, hand a task to a colleague with the context they need, or list what has been handed to you. Use when the user runs /qteam, mentions handing work over, asks who else has notes on something, or wants to find a concept documented in another project's vault.

Analyse every Claude Code session on this machine — friction and craft trends over time, an incident taxonomy with cost attribution, per-project comparison, and real repeat/correction exemplars — then render it as an interactive HTML report opened in a preview window. Use when the user runs /qtrends, or asks about prompting trends, historic patterns across sessions, whether they are improving, or which projects go badly.