Skip to content

fix(tsuga): correct field names, pagination, and portability across skills - #25

Merged
ArthurVerrez merged 2 commits into
mainfrom
sync-skill-updates
Sep 3, 2026
Merged

ArthurVerrez merged 2 commits into
mainfrom
sync-skill-updates

Conversation

@ArthurVerrez

@ArthurVerrez ArthurVerrez commented Sep 3, 2026 •

Copy link
Copy Markdown
Contributor

Skills read service fields the API does not return (sources[], *Count24h), so gates built on them could never pass. They now use traceRequestRate / traceErrorRate, and treat an absent rate as "query failed" rather than zero. Verified against CLI 1.28.0.

  • Resource lists default to 100 rows: monitor-coverage audits now page to totalCount instead of judging from page one. notification-silences list is not paginated and is excluded.
  • Cluster-scoped calls all take the same --cluster, not just the aggregations. An omitted environment no longer scopes an incident sweep to one env, since the registry reports only a service's busiest one.
  • check-skill-health scripts: fix an empty-array crash under macOS bash 3.2, a sed locale abort on binary assets that silently skipped a file's violations, prefix-only ISO timestamp validation, an ignored du status, and PASS lines that claimed sections were present after warning they were missing.
  • CSV reads in knowledge-technology use a quote-aware parser; awk -F, mis-parsed rows with commas inside quoted fields.
  • gh: finding PRs by path goes through the commits endpoint, since issue search does not index changed files and gh pr list caps at 30.
  • Dashboard and service-health skills pick up reference tables and gotchas that previously existed in only one of the two variants.
  • Plugin and marketplace bumped to 0.10.0.

Every tsuga command, flag, default, and limit referenced across the skills was checked against CLI 1.28.0; lint-all.sh reports 71 PASS / 0 WARN / 0 FAIL.

…kills

Several skills read service fields that the API does not return, so the gates
built on them could never pass. They now read traceRequestRate / traceErrorRate
and distinguish an absent rate (query failed) from zero (service quiet).

Resource lists page at 100 rows by default, so monitor-coverage audits built
from one page were wrong; they now page to totalCount. Cluster-scoped calls all
take the same --cluster, not just the aggregations, and an omitted env no longer
scopes an incident sweep to one environment.

The health-check scripts gain macOS fixes (empty-array expansion under bash 3.2,
a locale abort on binary assets), full ISO timestamp validation, and honest PASS
messages. Dashboard and service-health skills pick up the reference tables and
gotchas that were only in one place before.
Also brings two skill descriptions into the recommended length: gh was below it,
and knowledge-technology sat 34 characters under the 1024 limit where a loader
silently drops the skill.
@ArthurVerrez
ArthurVerrez marked this pull request as ready for review September 3, 2026 13:14
@ArthurVerrez
ArthurVerrez merged commit 81633d0 into main Sep 3, 2026
1 check passed
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.

1 participant