Rapid pass over the rest of the collection, with Theo's read on each.

teach — stateful learning

  • User-invoked only (disable model invocation); you trigger it with /teach (19:47).
  • Premise: the request to be taught something is stateful — the user intends to learn a topic across multiple sessions (19:47).
  • Mechanism: treat the current directory as a teaching workspace; the state of your learning lives in several files in that directory. Theo's takeaway — you're meant to run it in its own folder (20:14).

ask-matt — the router

  • Purpose-built to help you figure out which skill or flow fits your situation (20:14).

triage & wayfinder

  • triage — moves issues through a state machine of triage roles (20:14).
  • wayfinder — planning huge chunks of work as a map of decisions to be settled (21:03, 22:38).

wizard — the one he was most excited about

  • Purpose: walk a human through the steps only they can perform (21:26).
  • Ideal cases: an agent can't reach a dashboard it needs, or there's a sudo call it can't run. The wizard has the agent set everything up so you run one script and handle the rest from there (21:26).
  • Live install: Theo asks his fleet thread whether wizard was ever set up, noting there's one more machine in the fleet now (a new Mac Mini) (21:50, 22:20).
  • Result: Opus caught every change to his systems. The answer came back direct — no, wizard was never set up; it ranked seventh in the earlier report, but he only picked seven skills and it wasn't among them; it's installed now (25:14, 25:46). It appeared on all his machines (27:52).
  • He has nothing that needs it yet but can think of four things from the previous day he should have used it for (27:52, 29:28).

Side-by-side: the unslop effect, again

That wizard answer becomes an accidental A/B. Theo scrolls back to older threads in the same project, same codebase, same agents and models, from before the skills were installed, and reads a "recent changes summary" out loud (25:46, 26:10):

  • It's dense with em dashes and hedged asides ("rebasing his branch under him felt more intrusive than the situation called for", "worth watching whether the check goes green") (26:37).
  • "I can't even read this, and I'm trying to read it out loud for you guys in a video." (27:01)

writing-for-agents — the renamed "writing great skills"

  • Scope: a reference for writing any document an agent consumes — a skill, an AGENTS.md/CLAUDE.md, or a doc reached by a pointer (30:28).
  • Why it matters: agents are bad at writing instructions to other agents. If you let them write skills, markdown, or sub-agent prompts unaided, the output is poor (28:39, 29:04).
  • Theo's actual use: prompting sub-agents, where it's been very helpful (30:28).
  • Core concept — context pointers: a reference held in the agent's context that names some out-of-context material and encodes the condition for reaching it. A skill's description is one; a line in AGENTS.md naming a doc is the same object. The pointer's wording — not its target — decides when and how reliably an agent reaches the material. A must-have target behind a weakly worded pointer is a variance bug (30:28, 30:54).
  • When writing a skill specifically, read the skill-mechanics.md file for frontmatter, invocation choice, and router skills (30:28).
  • Criticism: strong ideas, unreadable delivery — too many em dashes and dense phrasing (30:54).

wait-what — the repair tool

  • User-invoked, roughly three lines long (29:04, 29:48).
  • Job: get a simpler, more digestible restatement when a response makes no sense. It pulls in "simplify technical English" (29:04).
  • Honest limitation stated in the skill itself: it repairs one message — it doesn't prevent the next sloppy one (29:04).
  • The real cure for jargon is a shared language built up front using grill-with-docs. Reach for wait-what only when you haven't done that setup yet (29:28).
  • It's functionally near-identical to pstack's bro skill (33:04).

domain-modeling

  • Theo already does this by hand — writing grammars and glossaries with agents in every project, defining what terms mean. Making it a skill makes sense to him (23:30).

On the docs themselves

  • "This is a good docs page. There's a lot to learn from here in the best sense." (29:48)
  • He's tempted to install everything but holds back — he's happy with his current setup and doesn't want to muddy it (23:56).