Blast Radius Check

Who the node is. The id is the key the DOT pins.

id

blast-radius-check

name

Blast Radius Check

type

validation

phases

Testing

What it does, and the prose the agent is handed when the graph runs.

action32 words

Expand the drafted mitigation into the concrete set of hosts and services it would touch, hold back anything wider than the declared limit, and emit the risk verdict the escalation path reads.

spec133 words

Read the proposed mitigation on draft and expand every target expression in it, globs, tags, service names, into the concrete list of hosts and services it would touch, resolving them against live inventory rather than trusting the count the draft claims for itself. Measure that resolved list against two limits: at most twelve affected hosts, and nothing that belongs to tier-0 or to payments. Emit your verdict on risk, and put the mitigation on mitigation with the resolved target list attached only when it sits inside both limits. Treat a target you cannot resolve as over the limit, not as safe, and never widen or narrow the mitigation to make it fit, you judge it, you do not edit it. You are done when a verdict is on risk, whichever way it went.in full above

model

claude-opus-5

agent

Blast-radius QA

skill

skills/blast-radius-check.md

tools

HTTP fetch

mcp

fetch

params

protected_tiers: ["tier-0","payments"], max_affected_hosts: 12

What arrives, what leaves, which nodes it expects to hear from, and what may not.

inputs

draft : json

outputs

mitigation : json, risk : status

dependencies

resolution-composer

cannot

no type is refused

will_not

edit the mitigation it judges, clear a target it could not resolve

The keys the static analysis reads. Nothing here instructs the agent.

risk_markers

none

notes80 words

It reads live inventory over HTTP to expand the target set, which is the only reason this node reaches the network at all. The core vocabulary has no metrics or observability capability, so http-fetch stands in for the metrics API. That tool is what makes the analyzer infer unvalidated-external-access here: this node is itself the validation step, and doc 3 §4.1 will not let a node vouch for its own fetch, so the inference is correct and is left standing.

The card's own version, and who wrote it.

version

1.0.0

author

autogen

provenance

not stated