Analysis Agent

The Analysis Agent is a pure-computation engine for routing analysis, What-If simulation, and topology visualization. It runs entirely in an isolated sandbox — no network access, no device connections.

Feature Claims

ID	Claim	Status
C-NE-22	What-If simulation: disabling a link recalculates connectivity and blast radius	🔶 Env-Blocked
C-NE-23	Analysis Agent uses only run_python_simulation — no live device access	⬜ Pending

---

One Tool, Full Power

Analysis has a single tool: run_python_simulation. Inside this sandbox, the agent writes Python code with access to:

Global	Type	Purpose
db	DatabaseProxy	db.query(sql) — read-only access to production snapshot data
sim	SimulationProxy	sim.clone(tables) then sim.execute(sql) — writable in-memory copy for mutations
nx	networkx	Graph engine: DiGraph, shortest_path, connected_components, has_path
netutils	netutils	IP math: subnet overlap, interface normalization, prefix calculations
json, math, itertools, collections	stdlib	Standard utilities

The sandbox enforces network_isolation=True — no HTTP calls, no socket connections.

---

Use Cases

Routing Analysis

"Which BGP neighbors on R1 are not established?"

The agent queries v_bgp_neighbors_auto via db.query() and returns a filtered table.

What-If Simulation

"What happens if R2 loses all links?"

The agent:

1. Clones topology data: sim.clone(['topology_links', 'ospf_neighbors', 'bgp_neighbors'])
2. Mutates: sim.execute("UPDATE sim_topology_links SET link_status = 'down' WHERE source_device = 'R2'")
3. Builds a networkx graph from the mutated state
4. Computes blast radius via nx.connected_components()
5. Returns affected devices, lost paths, and protocol impact

Path Analysis

"What is the shortest OSPF path from R1 to R4?"

The agent builds a weighted graph from OSPF costs and runs nx.shortest_path() with cost weights.

Topology Visualization

"Draw the network topology"

The agent queries netops.topology_links, builds a networkx graph, and renders a Mermaid diagram with device states and interface labels.

---

Simulation Workflow

Every simulation follows this pattern:

# Step 1 — Check data coverage
coverage = db.query("""
    SELECT 'topology_links' tbl, count(*) rows FROM netops.topology_links
    UNION ALL SELECT 'bgp_neighbors', count(*) FROM v_bgp_neighbors_auto
""")

# Step 2 — Clone needed tables
sim.clone(['topology_links', 'ospf_neighbors', 'routes', 'bgp_neighbors'])

# Step 3 — Mutate for What-If scenario
sim.execute("UPDATE sim_topology_links SET link_status = 'down' WHERE source_device = ?", ['R2'])

# Step 4 — Build networkx graph
links = sim.execute("SELECT source_device, destination_device, link_status FROM sim_topology_links").fetchall()
G = nx.DiGraph()
for src, dst, status in links:
    G.add_edge(src, dst, link_status=status)

# Step 5 — Analyze with active-only subgraph
active = [(u, v) for u, v, d in G.edges(data=True) if d['link_status'] == 'active']
G_active = G.edge_subgraph(active).copy()

# Step 6 — Return results
_result = {
    "blast_radius": [list(c) for c in nx.connected_components(G_active.to_undirected())],
    "isolated_devices": [n for n in G.nodes() if n not in G_active],
}

The agent always produces a Markdown report with: Current State, Simulation Scope, Impact Analysis, Change Plan, Verification Commands, and Risk Classification.