Each is a question you ask in plain language, and each says where every claim came from. They're instructed never to act unasked — and telos is built so it structurally can't. Together they stack: what it's all for → what's promised → what happened.
Build a purpose file from the evidence your machine can see, then get interrogated about it — hardest at the top.
The near-term goals are worked out from your actual work diary and repos, with receipts. The big-picture layer is never guessed for you — a pattern mined from logs presented as your life's purpose would be a bad week dressed up as an identity — so it's questions, answered by you or left honestly open. It never inserts itself into your sessions; it speaks only when you invoke it. And scope follows the notebook: a machine wired to a work-only store builds its telos from work evidence alone, because the personal evidence isn't reachable from there.
Say a commitment out loud today; ask “where do I stand?” in a fresh session tomorrow.
Your promises, kept in your own words — “review it deeply enough that it can merge by Wednesday” carries meaning a due-date field can't. When you ask, it checks the live facts (did that get merged? is the ticket overdue?), flags the promise nothing has moved on, and lets you reorder the list by just saying so. It never writes down a status — those are checked fresh every time, so they can't go stale.
Reconstruct what you actually did, from three sources — because git alone under-reports your most valuable work.
Weighing options, reviewing someone else's change, investigating something and deciding no — none of it shows up in the code you shipped. week merges the work diary, your project histories, and your team's review tools where connected, then drafts the summary you can actually send. It never invents activity, and says “no evidence” rather than “no work.”
keel explained from the live tree, never from memory of it.
Reads the sibling skills' own files, runs the CLI's own help, and answers from what's actually on disk — so its explanations cover every skill keel grows next with zero maintenance. Walkthrough mode — a dry look at week, deck, or telos — runs only read-only steps and stops where the writes would begin. Every explanation keeps the exit visible.
The install ceremony, as a conversation.
Inspects first, asks only about the gaps, names every write before making it, and drives the CLI's flag form so you never babysit a TTY prompt. Derives your project registry from the activity log and git remotes, asking only for what machinery can't know. Never destructive; safe to re-run any time.
keel status · log · doctor · setup · settings · update ·
migrate · link
Small commands the skills stand on — mostly for reading, careful about
writing, and safe to run twice. In practice the assistant runs them for you.
keel doctor is the noisy one: it checks everything the rest of
the system stays politely quiet about, and tells you exactly what to fix.
The technical version →