Skip to content

[new feature] Signadot Skills: agent-driven validation against live environments #129

@foxish

Description

@foxish

New feature or functionality

Signadot Skills would give coding agents the ability to validate their work against live infrastructure (Sandboxes and RouteGroups). The core idea:

  • Plans: A Plan is a named, versioned sequence of Actions (e.g. @request-http, @capture-logs, @check) generated from a natural-language prompt. Plans are tagged like Docker images (e.g. check-location-service-v1) for reproducibility.
  • Actions: Platform-governed primitives for interacting with live environments — HTTP/gRPC requests, log capture, assertions, screenshot capture, sandbox orchestration, etc. Backed by the signadot/actions repository.
  • Export to SKILL.md: Tagged plans can be exported as native coding agent skills (.claude/SKILL.md) or consumed via MCP/CLI, so the agent can re-run the exact same validation sequence while iterating.

Typical workflow: describe a validation task → Signadot generates a Plan → review and refine → tag → export → agent runs it autonomously during development.

Additional considerations

  • Runnergroups manage action dependencies and ship with a default image for built-in actions; custom actions should be supportable in the future.
  • Governance matters: platform teams need control over which actions are permitted.
  • Distribution across engineering teams (how skills are shared/discovered) is an open question.

Metadata

Metadata

Assignees

No one assigned

    Labels

    enhancementNew feature or request

    Projects

    Status

    In Progress

    Milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions