Add Node-RED design, debugging, and MCP workflow skills #32

New issue

Open

opened 2026-08-25 09:24:40 +00:00 by lost-rob0t · 0 comments

lost-rob0t commented

2026-08-25 09:24:40 +00:00

(Migrated from github.com)

Goal

Add portable agent skills for designing, editing, validating, debugging, and safely deploying Node-RED flows through the StarIntel Node-RED MCP service.

Dependencies

starintel-infra #61 Node-RED service
starintel-infra #62 custom Node-RED MCP

Skill set

Split the work into concise reusable skills, at minimum:

Node-RED flow design from a user requirement
node/subflow selection and composition
safe secrets/credentials handling
validate/lint/test before deploy
inspect/debug a failing flow using bounded runtime evidence
change an existing flow using revision-aware MCP operations
export/import flows without leaking credentials
production deployment checklist and rollback

Requirements

Skills must teach agents to use the MCP tools rather than inventing direct admin API calls or shell access.
Prefer read/validate/diff before mutation.
Preserve revision/hash preconditions on edits to avoid stale overwrite.
Never request, echo, or serialize Node-RED credential secret values.
Explain which operations are read-only vs mutating/deploying.
Include examples/fixtures that are useful but contain no real infrastructure secrets.
Keep skills framework-portable and dependency-aware according to this repo's conventions.
Document the dependency on the StarIntel MCP endpoint without hard-coding private credentials or one developer machine.

Acceptance

A fresh agent can use the skills to create a fixture flow, validate it, deploy it through the MCP, diagnose an intentional failure, patch it, and verify the resulting revision.
Skill docs do not encourage unsafe direct editor/admin access.
Repo-native tests/docs checks pass.

Execution mode — AUTO-RAGE

Run the full ADARD/ADADR loop autonomously: inspect current skill conventions and the MCP contract; design the smallest coherent skill set; adversarial review for unsafe agent behavior/secret leakage/stale writes; autonomous decision gate; test-first skill fixtures where supported; exact-head verification; push a focused branch and open/update a PR targeting main. Stop only for a genuinely unresolved requirement/security blocker, inaccessible dependency, or failed verification gate. Do not ask for approval between phases.

## Goal Add portable agent skills for designing, editing, validating, debugging, and safely deploying Node-RED flows through the StarIntel Node-RED MCP service. ## Dependencies - `starintel-infra` #61 Node-RED service - `starintel-infra` #62 custom Node-RED MCP ## Skill set Split the work into concise reusable skills, at minimum: - Node-RED flow design from a user requirement - node/subflow selection and composition - safe secrets/credentials handling - validate/lint/test before deploy - inspect/debug a failing flow using bounded runtime evidence - change an existing flow using revision-aware MCP operations - export/import flows without leaking credentials - production deployment checklist and rollback ## Requirements - Skills must teach agents to use the MCP tools rather than inventing direct admin API calls or shell access. - Prefer read/validate/diff before mutation. - Preserve revision/hash preconditions on edits to avoid stale overwrite. - Never request, echo, or serialize Node-RED credential secret values. - Explain which operations are read-only vs mutating/deploying. - Include examples/fixtures that are useful but contain no real infrastructure secrets. - Keep skills framework-portable and dependency-aware according to this repo's conventions. - Document the dependency on the StarIntel MCP endpoint without hard-coding private credentials or one developer machine. ## Acceptance - A fresh agent can use the skills to create a fixture flow, validate it, deploy it through the MCP, diagnose an intentional failure, patch it, and verify the resulting revision. - Skill docs do not encourage unsafe direct editor/admin access. - Repo-native tests/docs checks pass. ## Execution mode — AUTO-RAGE Run the full ADARD/ADADR loop autonomously: inspect current skill conventions and the MCP contract; design the smallest coherent skill set; adversarial review for unsafe agent behavior/secret leakage/stale writes; autonomous decision gate; test-first skill fixtures where supported; exact-head verification; push a focused branch and open/update a PR targeting `main`. Stop only for a genuinely unresolved requirement/security blocker, inaccessible dependency, or failed verification gate. Do not ask for approval between phases.