Skills are the new Code by Guy Podjarny
89
90%
Does it follow best practices?
Impact
87%
1.38xAverage score across 4 eval scenarios
Passed
No known issues
| # | Quote | Topic | Lines |
|---|---|---|---|
| 1 | "software will transform from revolving around code and implementation to revolving around intent and instructions" | Thesis / Tessl premise | 19–44 |
| 2 | "Tools... are pieces of software that we call to be able to do something deterministically, and they really are what turns a model into an agent." | Definition of tools | 45–68 |
| 3 | "context is really all about giving the agent information that it either cannot know, like opinions, or that it can find out but it is very inefficient or error prone to do so" | Definition of context | 69–98 |
| 4 | "We have skills that are loaded on demand by the user or with some hints." | Definition of skills | 69–98 |
| 5 | "a harness... is again deterministic software that wraps the probabilistic model. For instance, Claude code is a hardness." | Definition of harness | 99–138 |
| 6 | "you cannot run GHPR open unless you loaded that skill" (Intercom example of hooks) | Hooks as constraint | 99–138 |
| 7 | "skills are the most sort of dominant reusable unit of context that we've built" | Why skills are the focus | 159–172 |
| 8 | "Especially in the OpenClaw world. We've seen over 30% of skills being malicious." | Skill security | 173–204 |
| 9 | "negligent skills, which are skills that lack safety instructions" | Negligent skills | 173–204 |
| 10 | "At the moment, skills are more like copying something down from Stack Overflow." | Lack of dependency management | 205–232 |
| 11 | "skills, as they get executed by agents, over time, will rot just like software rots" | Skill rot | 233–262 |
| 12 | "if you don't have a good way to maintain and update them over time, they will become not very effective, and eventually they will become harmful. And misleading." | Skill rot consequences | 233–262 |
| 13 | "the solution [to] all those problems is to think about skills as code" | Central thesis | 263–278 |
| 14 | "static analysis, dynamic tests, dependency management, security tooling, and observability" | The five tools | 263–278 |
| 15 | "you cannot scale software without tests... the same is true for skills" | Evals are mandatory | 299–334 |
| 16 | "You can write very useful ones, or you can write... ones that just waste your time." (on evals) | Eval quality | 299–334 |
| 17 | "we think that the right future is for humans to operate mostly on the CDLC... while the agents should handle the SDLC end to end" | CDLC vs SDLC division | 411–428 |
| 18 | "skills which hopefully I've made some case is the new code" | Title thesis restated | 411–428 |
| 19 | "harnesses, which I think of more as frameworks. These are not something that you'd see every developer on the team building." | Harnesses as frameworks | 411–428 |
| 20 | "creating sort of the new development methodology is creating our new craft, and that is a community activity" | Closing | 429–470 |
evals
scenario-1
scenario-2
scenario-3
scenario-4