chaos-impact

name	chaos-impact
description	Synthesize an Azure Monitor impact report for a Chaos Studio v2 ScenarioRun: pulls the run, walks targeted resources, queries metrics/logs/activity-log/alerts over the run window ± buffer, and renders a Markdown + JSON impact card.

ChaosImpact — Post-Run Azure Monitor Synthesis

⛔ ABSOLUTE RULE: Do NOT improvise, skip, or substitute any step. On ANY error, STOP and wait for the user.

When to use this skill (vs. the MCP server)

This skill is the human-interactive path: it resolves run context from the shared startchaos-state.json, prompts for missing inputs via exit codes, and renders a Markdown card plus a JSON sidecar. Use it when there is a user in the loop — typically right after /start-chaos (or as a post-step of it).

If you are an autonomous agent with no user to prompt, call the three Monitor tools on the chaos-studio MCP server directly (monitor_query_metrics, monitor_query_logs, monitor_search_activity_log) and assemble your own report. The PowerShell skill and the MCP tools are two independent surfaces over the same Azure endpoints — neither calls the other.

Both surfaces target Microsoft.Chaos 2026-05-01-preview plus the pinned Azure Monitor API versions in scripts/Constants.ps1, and use the local az login session for auth.

How It Works

All synthesis logic lives in scripts/Invoke-ChaosImpact.ps1. The script resolves the ScenarioRun, discovers Log Analytics workspaces via diagnostic settings, runs the Monitor query fan-out, classifies signals, and writes both artifacts.

The AI orchestrator's only job is:

1. Set $env:STARTCHAOS_STATE_PATH to ${SESSION_DIR}/startchaos-state.json (the skill reads but does not write this file).
2. Run the script with -ScenarioRunId <id> (positional accepted).
3. Handle exit codes that require user input (see below).
4. Re-run the script with the user's answers as parameters.

Running the Script

$env:STARTCHAOS_STATE_PATH = "<session-dir>/startchaos-state.json"
& "<skill-dir>/scripts/Invoke-ChaosImpact.ps1" -ScenarioRunId <runId> @extraArgs

When invoked standalone (no prior start-chaos state file), the orchestrator must supply -SubscriptionId, -ResourceGroup, -WorkspaceName, and -ScenarioName — otherwise the script exits 2.

Exit Codes → AI Actions

Exit	Meaning	AI Action
0	Report generated	Done — print the Markdown card from script output.
1	Hard error	STOP. Render the error from script output. Wait for user.
2	Missing run context	ask_user for any of: -SubscriptionId, -ResourceGroup, -WorkspaceName, -ScenarioName, -ScenarioRunId. Re-run with the supplied parameters.
3	Log Analytics workspace not discoverable for ≥ 1 resource, AND -LogAnalyticsWorkspaceId not supplied	ask_user whether to (a) supply a workspace ID (re-run with -LogAnalyticsWorkspaceId <id>) or (b) proceed without log signals (re-run with -LogAnalyticsWorkspaceId none).
4	Permission gap (caller lacks Monitoring Reader on one or more target resource groups) (Epic 3 — not yet implemented)	ask_user to confirm best-effort partial report (re-run with -AllowPartial) or abort.

Script Parameters

Parameter	Description
-ScenarioRunId	Required (positional). The ScenarioRun resource name.
-SubscriptionId / -ResourceGroup / -WorkspaceName / -ScenarioName	Run context. Read from startchaos-state.json when omitted; required otherwise.
-Buffer	ISO-8601 duration added before/after the run window. Default PT5M.
-LogAnalyticsWorkspaceId	Optional override when diag-settings discovery fails for some resources. Pass none to explicitly skip log signals.
-MaxResources	Per-run resource cap. Default 50. Instance-level targets (VMSS / AKS) are collapsed to parent before counting.
-OutputDir	Where to drop artifacts. Default $env:STARTCHAOS_SESSION_DIR or current directory.
-IncludeBaseline	When set, also computes baseline samples (the -Buffer × 2 window before run start). Default on.
-Format	markdown, json, or both. Default both.
-AllowPartial	Continue past permission gaps and emit a partial report. (Epic 3 — not yet implemented; currently accepted but has no effect.)

What the Script Handles (no AI logic needed)

• Run resolution and targeted-resource flattening (collapses instance-level → parent)
• MaxResources cap with explicit coverage caveats
• Diagnostic-settings discovery + per-resource workspace caching
• Azure Monitor metrics / Log Analytics / Activity Log / Alerts / Service Health fan-out
• Baseline subtraction and signal classification
• Markdown report card + JSON sidecar artifact
• All error cards with remediation commands