For the complete documentation index, see llms.txt.

Documentation

Repository docs

This route renders the repository README and markdown under docs/ .

Source: docs/DEPLOYMENT_AGENT.md

Rendered document

docs/DEPLOYMENT_AGENT.md

Parsed server-side (markdown to HTML in the app). Same bytes you get from the checkout.

Deployment Agent

The Deployment Agent is a read-only AI inspector for your live deployments. Open it from the header Agent button.

It answers questions by fetching your deployment data — status, history, health — through tools, and can pull platform docs (via Moss search) when troubleshooting guidance is needed. It does not guess repos, services, or health states.

What it can do

CapabilityExample questions
List deployments"Show me my deployments"
Inspect current state"What's the status of my api service?"
Review history"Why did my last deployment fail?"
Check health"Is my service healthy right now?"
Search platform docs"What does an ALB unhealthy target mean?" (often paired with a deployment tool)
Search external docs"How does the ECS deployment circuit breaker work?"

What it cannot do

The agent is read-only. It cannot:

  • Trigger deploys or rollbacks
  • Change env vars, branch, or subdomain
  • Re-run Smart Analysis or Improve scan
  • Access another user's deployments

For actions, use the deploy workspace UI.

How to ask effective questions

Use repo and service names from your dashboard:

You sayAgent uses
acme/smart-deployrepoName: smart-deploy
service webserviceName: web

Good prompts:

  • "Show me my deployments"
  • "Why did smart-deploy web fail on the last deploy?"
  • "Is shop-api healthy right now?"

Ambiguous prompts: If you mention only api without a repo, the agent lists deployments first instead of guessing.

Tools the agent uses

ToolReturns
list_deploymentsUp to 25 deployments: status, branch, target, URL
get_deployment_detailsStatus, commit, revision, region, cloud resources, scan summary
get_deployment_historyRecent attempts: success/fail, failed step, log excerpts
get_runtime_healthRecent probes: app status, HTTP code, latency, ECS/ALB signals
search_docsRelevant Smart Deploy doc excerpts (Moss + deterministic search) for troubleshooting and how-to
search_external_docsOne public external technical-doc result, searched and converted to Markdown by Context.dev

External documentation setup

Set CONTEXT_DEV_API_KEY on the WebSocket worker environment. The key remains server-side: the agent sends it only to Context.dev and never to the browser, model prompt, tool result, or logs. The tool is optional; without the key it returns a clear unavailable result and the rest of the agent continues to work.

External results are limited to one HTTPS page per tool call, with article content trimmed before it is sent to the model. Prefer official sources such as docs.aws.amazon.com; the tool can restrict a query to a supplied public hostname.

Live status updates

While working, the agent streams progress:

  • Accepted → status updates → tool started/completed → final message

If the WebSocket worker is offline, you see: "The deployment agent is offline right now. Refresh the page and try again."

Limits

LimitValue
Tool calls per question6
Conversation memoryLast 6 turns
History/health samples per tool5
Write actionsNone

Complex root-cause analysis may hit the tool limit. Use Analyze failure on a specific history entry or read full logs in the History tab.

Starter prompts

Built-in shortcuts in the agent sheet:

  1. Show me my deployments
  2. Why did my last deployment fail?
  3. Is my service healthy right now?

When to escalate

NeedUse instead
Full step logsDeployment History tab
Deep failure analysisAnalyze failure on a history entry
Fix build planImprove scan
Platform how-toFAQ, Debugging Deployments

See AI Assistance.

Related