Claude Skills by atretyak1985

Design, review, or debug .gitlab-ci.yml pipelines for the project's repos. Covers stage ordering, job configuration, artifact flow, deploy safety flags, and the 8-stage model. Not for GitHub Actions, Jenkins, GitOps controllers, or GCP auth stanzas.

Use this skill when the project uses a pull-based GitOps controller (Wave B) for cluster reconciliation via a version-pinning repo's desired state. Don't use it during Wave A imperative GitLab deploys (use gitlab-ci-cd) or for CI pipeline structure (use gitlab-ci-cd).

Use this skill when a task involves Helm chart templating, values file structure, subchart dependency management, or chart validation (lint/template/dry-run) for the project's charts. Don't use it for Helm deploy orchestration (use deployment), Keycloak config (use keycloak), IaC drift (use infrastructure-as-code), or Docker builds (use docker-build).

Use this skill when a task involves detecting config drift between live cluster state and code, capturing manual kubectl/psql/gcloud fixes into Helm values or Terraform, preparing populated values files, or verifying a fresh deploy would succeed from code alone. Don't use it for Helm template authoring (use helm-chart-expert) or migration safety checks (use migration-check).

Configure Keycloak realm/client settings, integrate Auth.js v5 OIDC in the web portal (project.json → mainApp), set Keycloak Helm values (init/full stage), recover bootstrap admin, or debug KC_HOSTNAME/issuer mismatches. Do not use for Helm chart templating (use helm-chart-expert), GCP firewall rules (use kubernetes-deployment), or database migrations (use migration-check).

Use this skill for CLUSTER-LEVEL Kubernetes operations: Minikube/k3s lifecycle, GCP firewall rules, minikube tunnel debugging, and bootstrap-secret patterns. NOT for app-service Helm deploys or upgrade orchestration (use deployment) or Helm chart template authoring (use helm-chart-expert).

Use this skill when promoting an image or chart version across environments (dev -> staging -> production), rolling back a promotion, or updating the version-pinning repo's digests. Don't use it for single-environment Helm deploys (use deployment) or code-level rollback (use refactor-plan).

Provider-agnostic Jira access verification -- the first unconditional step of every jira-pack run. Resolves the Atlassian MCP tools by NAME (never by channel prefix), pins the discovered prefix for the run, resolves cloudId, and smoke-reads the target ticket. Any failed step stops the run with a JIRA ACCESS: FAILED report before any write call is made. NOT for reading/writing ticket content itself (that's jira-tasks / the write tools) and NOT for validating the jira config block (that's jira-...

Read and validate the jira block in .claude/project.json for jira-pack: required keys, defaults, working-repo resolution. NOT for talking to Jira itself (that's jira-tasks / the Atlassian MCP tools) and NOT for Confluence.

Close the needs-fix branch for both ticket classes: isolated git worktree, a red test BEFORE any implementation on change tickets, delegation to a core executor, every publish action (push / PR / comment / QA transition) gated behind a green @verification-agent verdict, commit + push + PR, then writeback via jira-writeback. NOT for classifying the ticket (that's jira-triage) and NOT for turning an over-budget attempt into a plan (that's jira-escalation, which this skill hands off to when a tr...

Convert a needs-fix attempt that has outgrown /jira-fix's autonomy budget (or a ticket triage already classified too-large) into a phased plan saved to the private workspace, post the plan's full text to Jira via comment-too-large.md, leave the ticket's status untouched, and remove any branch/worktree jira-delivery created before this fired. NOT for classifying the ticket (that's jira-triage) and NOT for the fix attempt itself (that's jira-delivery, which hands off here on budget exhaustion).

Parse a Jira ticket past access preflight, classify it as a defect or a change, run the mandatory evidence step that class demands (reproduction / baseline + absence proof), and assign exactly one of five verdicts (already-fixed / cannot-reproduce / needs-fix / needs-info / too-large). NOT for posting the verdict back to Jira (that's jira-writeback) and NOT for resolving jira config or the working repo (that's jira-config).

Post the run's verdict comment (rendered from the plugins/jira-pack/templates/ template matching verdict + ticket class) and, when the verdict allows it, transition the ticket to jira.qaStatus by exact match. Idempotent via a hidden marker carrying verdict and class; the comment always precedes the transition attempt. NOT for classifying the ticket (that's jira-triage) and NOT for the board card (that's swarmery-board-card).

Create and drive a /board card as a mirror of a jira-task-runner run: born in triage, moved straight to in_progress, finished at in_review or done. The card MUST NEVER pass through the todo column -- see 'The load-bearing rule' below for why. Covers projectId resolution, the exact POST/PATCH bodies, idempotent reuse via the jira-ticket label, the daemon-unavailable fallback, and dry-run request-body printing. NOT for talking to Jira itself (that's jira-tasks / the Atlassian MCP tools) and NOT...

Use this skill when writing or reviewing Python code for Raspberry Pi 5 edge devices -- UART hardware drivers, camera streaming via picamera2, GPIO control, systemd services, or resource monitoring. Don't use it for MAVLink protocol parsing (use mavlink-integration), web portal server code, or Helm work.

Use this skill when a task involves MAVLink protocol communication via pymavlink, UART/UDP/TCP drone connections, ArduPilot SITL simulation, or telemetry message parsing in the edge service. Don't use it for raw UART hardware driver code without MAVLink framing (use embedded-systems), WebSocket transport to the web portal (use api-integration), or Docker builds (use docker-build).

Use this skill to create and drive a UAV mission end-to-end through the web UI with the Playwright MCP browser tools on LOCALDEV (https://<localdev-host>) — rebuild/redeploy, Keycloak login, the multi-step BY_ROUTE/REALTIME mission wizard, pre-flight, the Start-Mission FSM, AND the full per-drone flight-command lifecycle (arm → take-off → hover → start/fly → stop → observation divert → resume → home/land/disarm), per-tab verification, and cleanup. Use it for E2E UI verification (incl. verifyi...

Plugin skill y from toolpack — stored as "toolpack:y".

NOT auto-discovered — only skills/ is scanned for plugins (§2.2).

Fixture skill whose usage guide deliberately omits the Worked example subsection.

Fixture global skill (minimal user-scope shape, format doc §2.3).

A SKILL.md whose first line is not `---`. Per system-docs-format.md §5.5 (and system-config-format.md §1.1, the rule the agent and command scanners already follow) a file with no frontmatter is not a registrable item, so the scanner must skip this directory silently — no skill row, no parse_error, no docs_missing finding. The CI coverage gate skips it for the same reason; the two sides must agree or an undocumented item hides behind a green gate.

Too short.