Skip to content

README.md

Latest commit

 

History

History
51 lines (40 loc) · 3.28 KB

README.md

File metadata and controls

51 lines (40 loc) · 3.28 KB
Raw

Leaf agent references

One file per leaf agent. Each file is the end-to-end reference for that agent's feature surface: what it does, how it's deployed, what APIs it wraps, data model, known gotchas, and troubleshooting. Keep user-facing operational content here; keep platform-internals content in ../architecture.md.

Current agents

File Agent Summary
cost-operations.md cost-operations-agent AWS spend analysis via Cost Explorer, CUR / Athena, and Cost Optimization Hub. Two deploy modes, CUR setup guide, COH enrollment gate.
pricing.md pricing-agent AWS Pricing catalog lookups, Cost Anomaly Detection, AWS Budgets status. One deploy mode — no cross-account needed.
health-events.md health-events-agent AWS Health events ingestion + query. Four deploy modes (single-account / org-mgmt / org-delegated / cross-account), rules-based risk scoring, LLM enrichment, backfill.
network-resiliency.md network-resiliency-agent Direct Connect topology discovery + 22-rule resilience assessment + DX pricing. Shared rule engine across MCP and frontend REST Lambdas.
tag-governance.md tag-governance-agent Read-only org-wide tag compliance via Resource Explorer + Organizations Tag Policies, with remediation deep-links and cost-allocation activation health. Three deploy modes, requires Resource Explorer multi-account search.
cloudwatch.md cloudwatch-agent Read-only CloudWatch alarm assistant — recommends best-practice alarms from a vendored metric catalogue and tunes thresholds from real metric history, emitting CloudFormation the user applies. Single-account or one cross-account spoke per deploy.

When to add a file here

Add a new reference file whenever a leaf agent has any of:

  • Multiple deploy modes (single-account vs org-wide vs cross-account).
  • An AWS Support-plan dependency or other pre-deploy gate the user has to clear.
  • A non-trivial data model the user queries against (e.g. a DynamoDB table with GSIs, or an MCP-surfaced schema with tiering or scoping rules).
  • Known gotchas that a user following the happy path will likely hit.

If the agent is a thin wrapper over one AWS API with no special setup, its MCP tool entry in tools.json plus the agent prompt in hierarchy.json is usually enough — no file needed.

Template to follow

Both existing files share this section shape; mirror it when adding a new one:

  1. What the feature does — one paragraph + one representative prompt
    • ASCII sketch of the pipeline if non-trivial.
  2. Deploy modes — decision tree for which mode applies to the user.
  3. Prerequisites / bring-up — AWS CLI commands to enable the feature on a cold account if any extra setup is required.
  4. Data model — tool response shapes, DynamoDB rows with indexes (if applicable). Real JSON shapes, not prose descriptions.
  5. Report template — if the agent ships a report template, list its sections and their tool calls here.
  6. Known gotchas / troubleshooting — symptom → likely cause → fix, as a table.

Lead with the user-facing statement of what the feature does. Put internals second. Skip any section that doesn't apply.