CS2680 Lecture 2 (2026-09-09) takes Claude Code apart. You type one request and it calls the model many times. Each time, the model only proposes a tool to run; the harness checks permissions, runs it, and puts the result back into context, and when the context fills up it gets compacted. The second half covers memory (the CLAUDE.md you write and the auto memory Claude keeps) and four extensions: skills cover how to do things, MCP reaches outside systems, hooks cover what must always happen, and plugins handle packaging. This post first shows what one API call looks like, then walks the whole loop through fixing one test.
The first half of CS2680 Lecture 3 (2026-09-14) is about scaling yourself up. It uses two plugins, Superpowers and Caveman, to show how skills, hooks, and a proxy fit together, then covers three directions to scale: run agents in parallel to buy wall-clock time, delegate to subagents to buy focus, and let one agent run long to buy scope. The second half is about the cost of scaling: when you have an agent write code from natural language, uncertainty seeps in from four layers (meaning, context, model sampling, the tool environment), and the remedies are to define clearly, separate proposal from execution, and tighten verification. This post works through "three bugs in three worktrees" to show what parallelism actually saves.
CS2680 Lecture 4 (2026-09-16) switches from using agents to designing them. An agent is a model plus a loop, and you decide which tools it gets and what it sees at each step. The tools half covers four ways to find code, four ways to edit it, the power and risk of a shell, and a set of design principles: few orthogonal tools, purpose-built over generic, token-cheap output, errors that teach. The context half covers why context is a finite resource, four ways to manage it (from guarding the door to summarizing as a last resort), and five ways context fails. It closes with a paper showing that when context is tight, the same model with only a different harness can solve 70% to more than twice as many tasks.
CS2680 Lecture 5 (2026-09-23) is about designing multi-agent systems. There are four reasons to split work across agents: specialization, isolation, parallelism, and context management. Any architecture can be described by three questions: who coordinates, how the work is split, and how messages move. The lecture lays out eight architectures, then covers six ways to decompose a task, a granularity formula for deciding whether to split, six common decomposition failures, and why communication is really a distributed systems problem: messages get lost or resent, and two agents end up changing the same thing at once. It closes by contrasting A2A with MCP, with "Is multi-agent really necessary?" as the first design principle.
The first half of CS2680 Lecture 6 (2026-09-28) is about agent memory. Keeping one fact means deciding, in order, whether to write it, what form it takes, where it lives, when to retrieve it, whether to generalize from it, and whether it still holds, plus who owns the memory. Don't simply delete outdated facts; record when they were true and where they came from. The second half is about reliability: at 95% success per step, a 20-step run is fully correct only about 36% of the time. Failures can be traced to five layers (task spec, harness, model, tool, environment), each with its own fixes. The two halves meet at one harness-layer failure: treating a stale note as fact.
CS2680 Lecture 7 (2026-09-30) is about measuring whether an agent is any good. First, be clear whether you are measuring a component or the whole task, and whether "success" means a correct answer, working code, or an environment left in the right state. On metrics, pass@k asks whether at least one of k tries succeeds, while pass^k asks whether all k succeed; from the same set of trials the two can come out as 0.9 and 0.3. Then come eight representative benchmarks and what each one measures, and finally when a score can't be trusted: the answer may be in the training data, the agent may look it up online, it may be sitting in the environment's git history, and the grader itself can be gamed.
CS2680 Lecture 8 (2026-10-07) has two halves. The first takes apart an agent's bill: cost is a product of users × sessions per user × turns per session × steps per turn × tokens per step × price per token, and the last four are the ones you can cut. The students' own Claude Code logs show 97.6% of tokens are cached re-reads, so keeping the prefix stable matters more than keeping the prompt short. Speed is set by the critical path, not the sum of every step. The second half compares three harness architectures (monolithic, modular, microkernel) and the design choices of eight open-source and commercial harnesses. This post works one bill by hand: put a timestamp at the start of the system prompt and the same task goes from 23.6 units back to 65 or more.
Daeryuk-bong offers southeast-facing granite sport climbing on two walls in Busan. The older guide starts at 5.10a; approach via Oncheonjang and bus 203, then check current park rules and the final walking approach.
Ganhyeon combines easier slabs, face routes and 5.12–5.13 cave climbing with a relatively straightforward approach. Consider cool sunny days, check that its limited 5.11 selection suits you, and walk in from designated parking.
Gochang’s Halmae-bawi offers vertical to gently overhanging pocketed sport climbing, an alternative to steeper Seonunsan sectors. Confirm the Gochang location, current access across local land and the Korean route names before travelling.
Insubong offers granite cracks, friction slabs and chimneys in Bukhansan National Park. Insu A and Chouinard B are graded YDS 5.8, but still require trad, multipitch and descent skills. Take the Ui–Sinseol Line to Bukhansan Ui Station, then a taxi to the Baegundae trailhead to save the uphill road walk.
Janggunbong in Seoraksan has Front and Southwest Walls with granite cracks, slabs and face climbing. A2O and the Standard Route offer different grades and lengths. Choose the face and full route before arranging park authorization.
The School Crag is a streambed bouldering sector in upper Musucheon. Cheese Dorito and Pig Farm are both recorded as Font 6C/V5; arrange pads, spotters, current access and upstream-weather checks before visiting.
Jeokbyeok is a west-facing granite wall beside Biseondae, with steep cracks and overhangs. Solo is listed as 5.8/A1, Echo as 5.11c/A2 and Mura as 5.9/A4: the free grade alone is not a route-selection guide.
Munsusan offers upper and lower south-facing gneiss crags near Ulsan. Sunshine makes it a winter option but a warm-weather drawback. Choose your sector and route length before planning the observatory parking approach and return transport.
Musucheon North offers basalt sport climbing, with older guidebook records from 5.8 cracks to 5.12 overhangs. Shade does not remove flash-flood risk: establish water conditions, retreat and lawful access before entering.
Seoninbong is a granite wall in Dobongsan with cracks, slabs and multipitch climbing; Leopard (표범길), Bat (박쥐길) and Yodel Buttress are the recognised classics. Even the moderate lines need your own protection, and most teams climb to mid-height and rappel the route. Subway lines 1 and 7 run to Dobongsan Station, about 40 minutes' walk from the wall; the closure after the 2025 rockfall was lifted that May, but check the KNPS live notices before you go.
Seonunsan brings vertical faces, pockets and steep sport routes into neighbouring sectors. Soksal and Tugu offer demanding climbing, while Mun includes easier routes. Choose the sector for your grade and sunshine, and treat temple-side access separately.
Ulsanbawi in Seoraksan offers granite cracks, chimneys and multipitch routes. The complete ridge is a different, multiday objective. Choose the wall route or ridge first, then confirm park authorization and the approach.
Yuseondae combines ridge climbing and southwest-facing granite multipitch. MP lists Longing for Two as 5.8 and 11 pitches, and Yoosun AB as 5.10b and six pitches. Choose the style before planning the approach and descent.