Skip to content

supercheck-io/supercheck / .agents/skills/platform-features

supercheck-platform-features

Verified

Work on the Supercheck CLI (commands, device login, JSON output), AI test repair and AI providers, realtime SSE streams, web playground, requirements, status pages and their uptime, tags, or their APIs and user workflows. For the MCP server use supercheck-mcp-server.

Supercheck platform features

flowchart TD
  USER --> CLI[CLI and monitoring as code]
  USER --> WEB[Web app]
  WEB --> PLAY[Playground]
  WEB --> REQ[Requirements]
  WEB --> STATUS[Status pages]
  WEB --> AI[AI test repair]
  RUN[Executions] --> SSE[Realtime events]
  TAG[Tags] --> REQ
  TAG --> TEST[Test and monitor resources]

CLI

  • Source: cli; package: @supercheck/cli; binary: supercheck; license: Apache-2.0 (from 0.3.1); supported Node runtime begins at 20 unless package metadata changes.
  • Config is declarative monitoring-as-code. Config loading validates schema and rejects embedded live/test/trigger token patterns, including legacy trigger formats.
  • Secret variable values become environment references; file variables are excluded from generated config and use supported multipart/dashboard paths.
  • Authentication storage uses the local conf store (static obfuscation key—not encryption); prefer SUPERCHECK_TOKEN from a secret store in CI, and never print tokens.
  • Preserve interactive and non-interactive behavior for every registered command. The source of truth is the addCommand list in cli/src/bin/supercheck.ts (commands live in cli/src/commands/); read it rather than relying on a copied list. Aliases include incident/incidents, service/services, alert/alerts, notification/notifications.
  • --json mode is enforced centrally in cli/src/bin/supercheck.ts (exitOverride, a postAction hook that emits a success object when a command printed nothing); new commands must not bypass it.
  • login --json prints the device URL and code and returns immediately; the approval UI lives in app/src/app/(device)/device/.
  • Resource kinds in config: monitors (with tags), tests (@title/@priority/@tags file annotations), jobs (no tags), requirements[], statusPages[] (settings, branding, notifications, components), notification providers, variables, tags. pull, diff, deploy and destroy cover requirements and status pages; requirement (alias requirements) is the imperative command group. Keep schema (cli/src/config/schema.ts), cli/src/utils/resources.ts, reconcile.ts, status-pages.ts, docs (docs/content/docs/cli), cli/CHANGELOG.md and the bundled guides in cli/src/commands/skills.ts in step.
  • Config semantics: omitted optional fields (requirement tags/tests, status page notifications flags, test annotations) leave the remote value alone and [] clears; tag names compare case-insensitively; status page subdomain is read-only (pulled, never sent); remote-only status page components are removed only with deploy --delete. A server without the write APIs must fail before any write with a clear message.
  • AI SRE CLI contracts are incident list/get/timeline/assign/unassign/ack/resolve (list --assignee me|unassigned|<id>; assign --user <id|me> [--commander]), sre triage/investigate/status/ask/brief, and read-only service list/get/health/dependencies. Use /api/sre/* routes with unified CLI-token/session authentication; never call browser-only server actions from the CLI.
  • Keep AI SRE tenant scope and RBAC server-side: incident reads need sre_incident:view, assign, acknowledge and resolve need sre_incident:update, investigations need both incident and investigation investigate, service reads need sre_service:view, and live connector tools additionally need sre_connector:investigate.
  • Treat deep investigation as asynchronous HTTP 202 acceptance. Do not describe it as streaming; stream Copilot chat and evidence briefs only. In JSON mode emit NDJSON for streams, enforce bounded event buffers and idle timeout, support Ctrl-C, and never retry side-effecting POST requests after uncertain completion.
  • Require confirmation and a non-empty audited summary for incident resolution (--summary, --comment is an alias); --force is the automation escape hatch. Do not add runbook execution or private-agent startup until versioned server protocols, authorization, approvals, and operational safety contracts exist.
  • Notification provider JSON uses --payload (--data alias); retain legacy --config compatibility and reject multiple simultaneous payload options.
  • Human output goes to readable terminal streams; --json output must remain machine-parseable without progress/noise on stdout.
  • Subprocess signal termination and command failures return nonzero status. Retry only idempotent network requests.
  • Reusable CI distinguishes a job trigger key from the CLI token required for polling/waiting. Bind workflow inputs through environment variables/quoted arrays, not shell interpolation.
  • Verify with typecheck, lint, all Jest suites, build, and npm pack --dry-run; inspect packaged files and executable entrypoint.

Realtime SSE

  • Authenticate and authorize private subscriptions before opening streams; bind each stream to the permitted tenant/resource/run.
  • Use dedicated blocking Redis connections where QueueEvents is involved.
  • Emit typed, bounded events and heartbeats; preserve event ordering/identity expected by clients.
  • Close subscriptions, timers, listeners, and Redis resources on abort, completion, timeout, navigation, and errors.
  • Reconnection/resubscription must not leak other resources or create duplicate UI state.
  • /api/test-status/events/[testId] reconciles from persisted state: a run that ends blocked, failed or error with no report row closes the stream as failed (cancel, capacity rejection, orphan sweep emit no queue event); queued or running without a report keeps waiting. The event no longer carries reportPath.

Requirements

  • Requirements are traceability and coverage objects. They are never executed and do not own an independent pass/fail state.
  • Coverage is derived from linked test executions and must remain explainable.
  • Coverage is computed per test from runs.metadata.testStatuses, falling back to the run status when a run has none. Runs tables show Triggered by as the job schedule (scheduled runs) or the key name and its owner (API runs).
  • Requirement, document, test, and tag links are tenant-scoped and protected against cross-project IDOR.
  • Browser requirements follow the recorder-first authoring path where applicable.
  • REST: app/src/app/api/requirements/** (list/create, [id], [id]/tests, [id]/tags, export). Shared logic lives in app/src/lib/requirements/service.ts, which dashboard server actions and REST both call; add behavior there, not in a route or action alone.
  • AI requirement processing uses sanitized bounded content and validated structured output.

Tags

  • Tags are scoped by organization and project and use consistent normalization/color/uniqueness rules.
  • Resource joins preserve tenant scope. Filtering semantics remain aligned across API, UI, CLI, and reports.
  • Deletion checks all supported relationships and returns conflict when a tag is still in use if that is the current contract.

Status pages

  • Public status endpoints expose an explicit safe field allowlist; admin APIs require RBAC and ownership.
  • Support public/private status, service associations, incidents/maintenance, branding policy, and verified custom domains according to current schema.
  • Custom-domain verification and any outbound DNS/HTTP checks use hardened resolution and SSRF controls.
  • Preserve cloud and self-hosted routing, CNAME guidance, TLS behavior, and branding configuration.
  • Public subscribe (app/src/actions/subscribe-to-status-page.ts) is anonymous, so it is hardened: a 5-minute cooldown (one atomic UPDATE) on re-sending verification and on webhook re-subscribe, a never-verified webhook row gets a rotated secret while a verified or quarantined one keeps its own, and Slack endpoints must be https://hooks.slack.com. Component status ignores pending and paused monitors while others report (calculateFromMonitors); with none reporting, paused keeps the last known status. A component whose running monitors all missed three checks (at least 15 minutes; never-checked monitors age from creation) shows "No recent data" on the public page and badge (app/src/lib/status-pages/stale-components.ts), based on check times only, never on billing state.
  • REST writes: POST /api/status-pages, PUT /api/status-pages/[id], and [id]/components (+ [componentId]). Shared logic lives in app/src/lib/status-pages/service.ts, called by both dashboard actions and REST. Monitor tags: app/src/app/api/monitors/[id]/tags/route.ts. Public/CLI contract changes also need docs/openapi.json and the CLI updated.

AI test repair and providers

  • Sanitize scripts, errors, logs, and user instructions before model use; defend against prompt injection and secret inclusion.
  • Keep provider credentials server-side and separate provider namespaces from storage credentials (notably Bedrock AWS versus S3/R2).
  • Validate generated code/patch structure and run it only through normal script validation and isolated execution.
  • Apply timeout, cancellation, retry, usage accounting, configured provider/model selection, and safe error behavior across supported providers.

Web playground

  • Playground execution is untrusted and ephemeral: enforce strict resource/time limits, isolation, rate limits, result bounds, and short retention.
  • It must not bypass normal script validation, SSRF policy, capacity controls, or secret boundaries.
  • Preserve zero-install UX while keeping durable project resources separate from temporary runs.

Verify

  • Test tenant/RBAC denial and malformed input for every feature API.
  • Test SSE disconnect cleanup and cross-tenant isolation.
  • Test CLI compatibility and JSON output as public contracts.
  • Use browser tests for recorder-first, status-domain, and realtime UI behavior that unit tests cannot prove.
finishing-a-development-branchobra/superpowersUse when implementation is complete, all tests pass, and you need to decide how to integrate the workStars: 319kUpdated Claude CodeCodexVerifiedInstall
systematic-debuggingobra/superpowersUse when encountering any bug, test failure, or unexpected behavior, before proposing fixesStars: 319kUpdated Claude CodeCodexVerifiedInstall
test-driven-developmentobra/superpowersUse when implementing any feature or bugfix, before writing implementation codeStars: 319kUpdated Claude CodeCodexVerifiedInstall
skill-creatoranthropics/skillsCreate new skills, modify and improve existing skills, and measure skill performance. Use when users want to create a skill from scratch, edit, or optimize an existing skill, run evals to test a skill, benchmark skill performance with variance analysis, or optimize a skill's description for better triggering accuracy.Stars: 196kUpdated Claude CodeCodexVerifiedInstall