Skip to content

dx-resilience-auditor

View on GitHub

⚠️ Sample code: Not intended for production use without additional review and testing. Validate in a non-production environment first.

Custom agent that audits Layer-3 failover resilience across redundant Direct Connect VIFs — on demand via chat, or on a schedule (e.g. weekly). It verifies that the prefixes exchanged over redundant VIFs actually cover each other in both directions, and reports every prefix that would be blackholed if a single VIF failed. Strictly read-only.

Setup

Two steps are the same for everyone; the rest offers three equivalent installation options.

Always first:

  1. Deploy the MCP server — dx-resilience-mcp (SAM template) in the account owning the VIFs.
  2. Register it at the account level — console → Capability Providers → Register MCP Server. Endpoint = the Function URL; auth = AWS SigV4 with the access role, Region of the stack, Service Name lambda. (Registration here does not yet attach it to any Agent Space.)
  3. Attach it to your Agent Space — this is a separate, required step: select your Agent Space → Capabilities tab → MCP Servers → Add → choose dx-resilience-mcp → Select specific tools → enable get_dx_topology, analyze_vif_resilience, compare_prefix_sets, each classified read-only. You should then see "3 connected".

Critical — MCP tool identifiers are prefixed with the server name. In the Agent Space toolbox the tools appear as dx-resilience-mcp_get_dx_topology (etc.), i.e. <registered-server-name>_<tool>. Two places must use these prefixed IDs: the skill's enable_tools (activates them when the skill loads) and the custom agent's tool list. Register the MCP server as exactly dx-resilience-mcp so the shipped IDs match; if you register it under a different name, replace the dx-resilience-mcp_ prefix accordingly in both places. Bare names (no prefix) are silently accepted but never bind at runtime — the agent then aborts every run with "analysis tools not available".

Then create the skill, agent, memory store, and trigger — pick ONE option:

Requires a GitHub account connected to your Agent Space. Keeps the skill and agent in sync with this repository: when we ship fixes, click Sync.

  1. Import the skill — dx-vif-resilience via Knowledge → Skills → Add skill → Import from repository (directory URL).
  2. Import this agent — Agents → Create agent → Import, pointing at this directory's SYSTEM_PROMPT.md. Assign the skill during import.
  3. Assign tools via Chat (tools can't be set in the form): Add the get_dx_topology, analyze_vif_resilience and compare_prefix_sets tools to my dx-resilience-auditor agent.
  4. Attach a memory store (recommended) — create a custom store (e.g. dx-resilience-intent) and attach it, so confirmed redundant sets and acknowledged asymmetries persist across runs: Attach the dx-resilience-intent memory store to dx-resilience-auditor.
  5. Add a schedule trigger (optional) — agent detail page → Triggers → Schedule, e.g. rate(7 days) or cron(0 9 ? * MON *). Optionally add a per-trigger prompt such as "Full audit, both directions, all redundant sets."

Option B — Copy & paste (no prerequisites)

Same as Option A, but instead of importing: upload the skill directory as a zip (Add skill → Upload skill), and create the agent via Create agent → Form, pasting the contents of SYSTEM_PROMPT.md into the system prompt field. Fully editable locally, but you won't receive upstream updates — watch this repo and re-paste to update.

Option C — CloudFormation (for IaC-managed Agent Spaces)

Provision the agent, memory store, and trigger declaratively with assets-template.yaml (AWS::DevOpsAgent::Asset + AWS::DevOpsAgent::Trigger): it creates the custom agent, the dx-resilience-intent memory store, and the weekly trigger in one stack.

aws cloudformation deploy \
  --template-file assets-template.yaml \
  --stack-name dx-resilience-auditor-assets \
  --parameter-overrides \
      AgentSpaceId=<your-agent-space-id> \
      SkillAssetId=<assetId of the imported dx-vif-resilience skill> \
      McpToolIds=<tool ids of the three registered MCP tools>

Notes: - Import the skill first (Option A step 1 — even the CFN path imports the skill from GitHub rather than duplicating its content). Get SkillAssetId and the tool IDs from aws devops-agent list-assets / your Agent Space toolbox. - The agent prompt is embedded in the template as AGENT.md; SYSTEM_PROMPT.md in this directory is the source of truth — keep them in sync.

Usage

  • Scheduled: runs unattended, produces an assessment artifact + recommendations; low-confidence set groupings are labeled for one-time confirmation.
  • Chat: Run dx-resilience-auditor and focus on dxvif-aaa and dxvif-bbb.
  • First run tip: confirm the proposed redundant sets in chat once — the agent writes them to the memory store and every later scheduled run uses them.

Safety

All assigned tools are non-mutative; the MCP server's IAM role is structurally read-only (directconnect:Describe*/List*/Get* only). The agent's prompt and skill additionally forbid mutating actions, including BGP failover tests.