
Description
Maintain and improve existing k6 test scripts. Covers threshold tightening based on trend data, version migration between k6 releases, auto-fixing tests when the underlying service changes, refactoring for cleanliness, and auditing scripts against current best practices from docs. Use when the user asks to fix a failing k6 test, tighten thresholds, migrate a script to a new k6 version, refactor a test, update a script after a service change, or improve a script with best practices. Trigger on phrases like "fix my k6 test", "tighten my thresholds", "migrate to k6 v2", "update my test script", "refactor this k6 test", "my test is failing after a deploy", "apply best practices to my script", "modernize my k6 test", or "the service changed and my test broke". Also trigger when another skill (k6-trend-analysis or k6-cloud-investigate-test) hands off with a recommendation to edit a script.
SKILL.md
k6 Test Maintenance
Maintain, fix, and improve existing k6 test scripts. Five maintenance tasks,
each with a step-by-step procedure in references/workflows.md:
- Threshold tightening -- adjust threshold values based on observed metrics
- Version migration -- update scripts for new k6 releases
- Service change adaptation -- fix tests when the underlying service changes
- Refactoring -- clean up and modernize test code
- Best practices audit -- check scripts against current k6 best practices
Core principle: behavior-aware change control
Classify every proposed change by whether it alters the test's runtime behavior:
- Syntactic (behavior unchanged): the k6 runtime produces identical metrics,
pass/fail results, and endpoints. Examples: rename a variable,
let→const, remove unused imports, update comments, reformat. Apply directly. - Behavioral (behavior differs): anything affecting metrics, pass/fail,
timing, request targets, or load shape. Examples: threshold value changes,
adding
sleep(), endpoint URL updates, check rewrites, scenario changes, new thresholds. Always present as a diff with rationale and require confirmation.
The threshold for "behavioral" is deliberately low. If in doubt, treat it as behavioral and ask -- a trivial-looking threshold change can cascade to CI gates, SLO calculations, and alerting.
Dependencies
k6-manage-- fetch and edit GCk6-hosted scripts safely (§5: GET, backup, edit, validate, PUT, verify by sha256). Read it before touching any cloud-hosted script.gcx-- sole tool for Grafana Cloud API access.- mcp-k6 tools --
validate_scriptandget_documentation. Check availability first; fall back tok6 x docsif absent. k6 x docsCLI -- documentation lookup when mcp-k6 isn't configured.k6CLI -- local validation (k6 inspect,k6 run).
Validation loop (every edit)
Every workflow produces a modified script. Never present or PUT an unvalidated script -- run this loop, fixing and re-running until it passes:
- Parse-check:
k6 inspect <script>-- catches syntax errors, invalid options, broken imports. Works on all types including browser tests (no browser needed). If mcp-k6 is available, also runvalidate_script. - Local smoke (non-browser, service reachable):
k6 run --vus 1 --iterations 1 <script>. - Classify the change (below) and verify per the matrix -- recipes in
references/verification.md. - Cloud-hosted scripts: apply via the k6-manage §5 safe-edit recipe
(GET → backup → edit → validate → PUT as
application/octet-stream→ sha256-verify).
Change classification
- Class A -- declarative-config only. The diff is confined to
options.thresholdsor similar declarative fields that don't alter what the k6 runtime executes; the bytes insidedefault function, imported modules, and check predicates are byte-identical. Example:p(95)<500→p(95)<420. - Class B -- runtime logic changes. Any change to
default function, imports, helper modules, request URLs, check predicates, or toscenarios.*.vus/iterations/duration/executor(which alter load shape and metric distributions). Example: changing a URL, adding a check, rewriting auth, switching executors.
When in doubt, treat as Class B.
Verification matrix
| Class | Test duration | Verification |
|---|---|---|
| A | any | sha256 + k6 inspect + historical pass/fail prediction. No cloud run needed. |
| B | short (< 5 min) | sha256 + k6 inspect + full cloud run (k6-manage §11). |
| B | long (≥ 5 min) | sha256 + k6 inspect + local 1-iteration smoke + k6 cloud run of a local copy with --vus 1 --iterations 1. PUT to the saved test only after the cloud smoke passes. |
Verification depth depends on the change class, not the test's duration -- most
edits don't need a full run, and production tests may run for hours. Per-class
recipes (Class A prediction table, Class B short/long, edge cases like scenario
changes and loosening) are in
references/verification.md.
Documentation lookup
Before proposing any change that touches k6 APIs, imports, or patterns, confirm it against current docs and cite the source in your report -- this grounds recommendations in the real API, not stale model knowledge. Look up in order:
- mcp-k6 (preferred):
get_documentation("best_practices"),get_documentation("javascript-api/k6-browser"),validate_script(...). k6 x docsCLI (always available):2-call strategy: try the direct path first; if it returns a topic list, pick the subtopic and call again. Full parent paths required (k6 x docs using-k6 thresholds k6 x docs javascript-api k6-http k6 x docs search "websocket migration"using-k6 thresholds, notthresholds).k6 x docsserves docs for the installed k6 version -- it may lag the target version when migrating.- Web fetch (last resort):
https://grafana.com/docs/k6/latest/.
Async check pattern
A common browser-test bug: using check() from k6 with async predicates. The
built-in check() does not await Promises, so
check(page, { 'title': p => p.locator('h1').textContent() === 'Foo' }) silently
passes because the Promise object is truthy. Two valid fixes:
- Async-aware check from jslib:
import { check } from 'https://jslib.k6.io/k6-utils/1.5.0/index.js'-- then predicates can beasyncandawaitinside them works. - Resolve the value before the check:
const text = await page.locator('h1').textContent(); check(text, { ... })-- keeps the standard synccheckfromk6.
When you hit this during any workflow (migration, refactor, audit), flag it as a behavioral bug and propose one of these fixes.
Script sources
- GCk6-hosted -- fetched and pushed via
k6-manage§5 (GET → backup → edit → validate → PUT → verify sha256). - Local on disk -- read and edit directly. Validate before presenting.
Determine the source before starting: a GCk6 test URL or ID is cloud-hosted; a file path is local.
Workflows
Full procedures are in references/workflows.md:
- Threshold tightening -- propose values with observed-metric justification, diff, apply, Class A verify.
- Version migration -- find deprecated/renamed APIs, classify syntactic vs behavioral, apply, Class B verify.
- Service change adaptation -- map each service change to a script change, propose fixes, Class B verify.
- Refactoring -- find issues, auto-apply syntactic, propose behavioral, Class B verify after confirmation.
- Best practices audit -- doc-driven audit across thresholds, load design, resource management, code quality, and browser specifics.
All five follow behavior-aware change control: auto-apply syntactic changes, present behavioral ones as diffs for confirmation.
Gotchas
| Issue | Detail |
|---|---|
| Cloud script format | GCk6 scripts can be single files or tar archives. Detect with file(1) before editing (see k6-manage §5). |
| Zero-observation thresholds | A threshold on a metric with no observations passes by default. When adding new thresholds, ensure the metric is actually emitted by the test. |
| abortOnFail cascades | If a threshold has abortOnFail: true, tightening it means runs abort earlier. Warn the user. |
| Browser script validation | Browser scripts can't be validated with k6 run --iterations 1 without a browser. Use k6 inspect for parse-only validation, or validate_script via mcp-k6. |
| k6 x docs version alignment | k6 x docs serves docs for the installed k6 version; when migrating to a newer version, local docs may not reflect the target API. Note this in migration lookups. |
| Script drift after edit | After pushing a cloud-hosted script, the next run uses the new version, but historical runs keep their bundled snapshot. To investigate a past failure, compare the run-bundled script (read-only), not the current one. |
References
references/workflows.md-- step-by-step procedures for the five maintenance tasksreferences/verification.md-- per-class post-edit verification recipes
More skills from the skills repository
View all 48 skillsadaptive-metrics
optimize Grafana Cloud metrics costs
Jul 12Cost OptimizationGrafanaMetricsObservabilityadmin
manage Grafana Cloud accounts and RBAC
Jul 12Access ControlAuthGrafanaOperationsadmission-control
implement admission control webhooks
Jul 12ArchitectureSecurityValidationalerting-irm
configure Grafana Alerting and Incident Management
Jul 12AlertingGrafanaIncident ResponseMonitoringalloy
build unified telemetry pipelines with Grafana Alloy
Jul 12GrafanaLogsMetricsObservability +1app-observability
monitor application performance in Grafana Cloud
Jul 12APMDistributed TracingGrafanaLLM +1
More from Grafana
View publisherfaro-setup-web
instrument web apps with Grafana Faro
faro-web-sdk
Jul 12Distributed TracingFrontendMonitoringObservabilityconfiguring-yesoreyeram-infinity-datasource
configure Grafana Infinity data source
grafana-infinity-datasource
Jul 12API DevelopmentAuthenticationConfigurationGrafanaquerying-yesoreyeram-infinity-datasource
query data with Infinity datasource
grafana-infinity-datasource
Jul 15CSVData AnalysisGrafanaGraphQL +1agento11y
manage Grafana Agent Observability resources
gcx
Jul 18AgentsGrafanaMonitoringObservabilityagento11y-eval-starter
scaffold evaluation experiments for agents
gcx
Jul 21AgentsEvalsExperimentsGrafanaagento11y-instrument
instrument LLM apps for agent observability
gcx
Jul 21AgentsGrafanaInstrumentationLLM +1