Skip to content

docs(skills): scale-semantics guidance and value-level LWS eval - #195

Closed
rogirun wants to merge 4 commits into
mainfrom
feat/skill-scale-semantics
Closed

rogirun wants to merge 4 commits into
mainfrom
feat/skill-scale-semantics

Conversation

@rogirun

@rogirun rogirun commented Jul 28, 2026

Copy link
Copy Markdown
Contributor

What does this PR do?

Skill improvements for add-workload-type, surfaced by testing it against the
built-in catalog:

  • Adds scale-semantics guidance: a component's replicasPath is the count of its
    own instances, not the pods beneath it. Grouped/replicated workloads expose two
    numbers (how many groups/replicas vs members-per-group); a group-level
    component scales on the group count. The LeaderWorkerSet trap
    (group -> .spec.replicas, not .spec.leaderWorkerTemplate.size) is spelled
    out, with a sibling-consistency self-check.
  • Adds a value-level eval (id 4) that asserts extracted scale numbers against a
    concrete manifest, grounded in the real API, so a valid-but-wrong scale path is
    catchable rather than invisible to a match-the-definition comparison.

Stacks on #187.

Related issue(s)

Fixes #194

Checklist

  • All commits are signed off with DCO (git commit -s)
  • New/modified files have SPDX license and copyright headers
  • Documentation updated (if applicable)
  • Tests pass (make check) - docs/evals only, no Go impact
  • No proprietary or internal information included

rogirun added 4 commits July 27, 2026 12:53
Add an Agent Skill that guides authoring and validating a Karta definition
for a new Kubernetes workload type. Bundles the authoring workflow, a
technical-guide cheatsheet, a sample index, a troubleshooting catalog, and
three eval prompts. Content is grounded in the API types and validator.

Signed-off-by: Roee Gil <roee.gil@run.ai>
Replace the prose eval-prompts.md with evals/evals.json in the skill-creator
schema (id, prompt, expected_output, expectations) so the add-workload-type
skill is consumable by the skill-creator eval and benchmark tooling. Each eval
carries verifiable expectation assertions rather than a golden output.

Signed-off-by: Roee Gil <roee.gil@run.ai>
Close the gaps surfaced by testing the skill against all committed samples:
role-label keys are operator-specific (replica-type vs job-role), disambiguate
roles that share a label via key existence, use byExpression when a controller
has no phase/conditions, note a single matcher may AND its kinds, and allow
intentional additionalChildKinds duplication for RBAC/owner traversal.

Signed-off-by: Roee Gil <roee.gil@run.ai>
A component's replicasPath is the count of its own instances, not the pods
beneath it. Add guidance (with the LeaderWorkerSet group-vs-size trap and a
sibling-consistency self-check) and a value-level eval that asserts extracted
scale numbers against a concrete manifest, so this class of valid-but-wrong
scale error is catchable.

Relates to #186.

Signed-off-by: Roee Gil <roee.gil@run.ai>
Base automatically changed from feat/add-workload-type-skill to main August 17, 2026 12:55
@rogirun rogirun closed this Sep 16, 2026
@rogirun
rogirun deleted the feat/skill-scale-semantics branch September 16, 2026 09:12
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

add-workload-type skill: scale-semantics guidance and value-level evals

1 participant