skills I created/used - #2
Open
JorgeG94 wants to merge 2 commits into
Open
Conversation
Purely additive: nothing in the existing skill (SKILL.md, references/,
README.md) is moved, renamed or edited.
skills/
mom6-gpu-architect map a routine's shape and call graph, then decide
refactor-first vs port-in-place. The triage input is
the loop-interior call list, not the line count -- a
callee at loop depth > 0 must become device-callable or
be hoisted first. Ships routine-map.sh and callees.sh.
gpu-data-residency where enter/exit data belongs, map(to:) vs map(alloc:),
and where update from/to is needed for host-only
consumers. Engine is a host/device coherence state
machine walked in execution order. Ships
residency-scan.sh.
mom6-gpu-programmer the driver: sequences the other skills and owns the
stop-gates between phases.
knowledge/
KNOWLEDGE.md and the 15 deep-dive docs. The skills are not self-contained --
they make 59 references into these -- so the knowledge travels with them.
knowledge/README.md records provenance: the file:line anchors are only valid
against dev/gpu @ c82e1254a, and there is no MOM6 source in this repo to
contradict a stale one. It also flags that KNOWLEDGE.md's §8a/§8b confidence
markers (verified from source / needs a run / open maintainer decision) are
load-bearing and should not be flattened to assertion.
These skills sit alongside the existing one rather than replacing it: that one
covers how to write a correct MOM6 GPU loop; these cover what to port, in what
order, and where the data lives. A follow-up should reconcile the overlaps and
agree a multi-skill layout -- both are better raised with the author than done
unilaterally.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
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
1 participant
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.
To make these I first had Claude super explore the codebase and create KNOWLEDGE.md plus the other little files that are inside the repo. Based on that I distilled the information down and created three skills:
gpu-data-residencywhich knows about enter/exit data and maps back where and what is being allocated/copied;mom6-gpu-architectwhich is a planner skills, looks at a module you point it at and uses the knowledge and stuff to propose a porting plan, suggests if refactor is needed first;mom6-gpu-programmerwhich uses both skills and is an end-to-end skill that will design, propose (ask you about it), and go ahead and implement. It knows to keep ocean.stats for bitwise repro. It WILL commit once it has proven something is bitwise repro, if it can't get bitwise reproducibility it will stop and wait for you.I am still reading through all the knowledge docs it created but this skill produced this PR