internal dev cache-expert agent skill#11720
Merged
sipsma merged 1 commit intodagger:mainfrom Jan 22, 2026
Merged
Conversation
Been working with Claude/Codex to brain dump how caching works and how to debug caching issues into a set of docs loadable through a skill. This is still WIP but it's easier to keep iterating on it once merged as things change. Can't hurt in the meantime either. Signed-off-by: Erik Sipsma <erik@sipsma.dev>
sipsma
commented
Jan 22, 2026
Contributor
Author
There was a problem hiding this comment.
This doc is quite out of date and seemed to confuse the LLM more than help, so I just rm'd it here. It's largely covered in new docs added to the skill here.
Contributor
Author
There was a problem hiding this comment.
cc @vito since you wrote it way back when
shykes
approved these changes
Jan 22, 2026
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Had some promising but not quite fruitful trials w/ latest LLM coding models (codex 5.2-xhigh + opus 4.5) when it comes to debugging super gnarly engine issues around caching.
Seemed like one bottleneck was that it had to learn (and re-learn after context compaction) the whole codebase every time (which is often necessary for particularly tricky problems). So giving it detailed developer docs that focus it in on the parts that truly matter when debugging cache stuff may save a lot of file read/reasoning tokens.
Another issue was that it needed to be told all the time (or discover on its own) generally how to debug. Not just what commands to run but the techniques (i.e. println debugging of cache keys will get you very far).
Trying to package this all up in a skill. ATM the main utility of skill vs other packaging is just that LLMs have baked in knowledge on progressive disclosure of various skill docs, which is highly important here so we don't spend all our tokens just reading these docs.
This is still a WIP, it's basically become "the ultimate brain dump". I've been chipping away in the evenings, doing a mix of hand human writing and having LLMs interview me about a topic and then write everything it learned from me and from the code.
There's still a lot left to cover but probably reaching the point of utility. May give it a try on some caching related flakes and see how it does.