Inspect Walrus Network

Inspect Walrus Network

This skill inspects the Walrus production network (mainnet or testnet), retrieves committee and storage node information, and answers queries about individual nodes or global network status.

Prerequisites

The walrus CLI binary must be installed. If it is not available on the system, install it by following the instructions at https://docs.wal.app/docs/getting-started. The typical install command is:

curl -sSf https://docs.wal.app/setup/walrus-install.sh | sh -s -- -n testnet

After installation, the binary is typically at ~/.local/bin/walrus. Verify with:

walrus --version

If walrus is not in PATH, check ~/.local/bin/walrus or ~/.walrus/bin/walrus.

Important: Sandbox and Network Access

The walrus CLI requires outbound network access to Sui RPC endpoints and storage nodes. All walrus commands must be run with dangerouslyDisableSandbox: true in the Bash tool, otherwise connections will fail with "client error (Connect)" or "Operation not permitted" errors.

Similarly, all curl requests to Sui GraphQL endpoints and dig DNS lookups require dangerouslyDisableSandbox: true.

Configuration

Since this skill must work without assuming local config files exist, write the appropriate config to a temporary file before running commands.

The Walrus package ID is not listed below as it is stored as a field in the system and staking objects and can be read from there.

Mainnet Configuration

Walrus config (for --config flag):

system_object: 0x2134d52768ea07e8c43570ef975eb3e4c27a39fa6396bef985b5abc58d03ddd2
staking_object: 0x10b9d30c28448939ce6c4d6c6e0ffce4a7f8a4ada8248bdad09ef8b70e4a3904

Additional Sui packages and objects (for on-chain analysis):

WAL_PKG_ID=0x356a26eb9e012a68958082340d4c4116e7f55615cf27affcff209cf0ae544f59
UPGRADE_MANAGER_OBJECT=0xc42868ad4861f22bd1bcd886ae1858d5c007458f647a49e502d44da8bbd17b51
WALRUS_SUBSIDIES_PKG_ID=0x14b874da49e152d2b2910122330f7eb925d75bdb0a0f8e2c6b9b1162a5560a8c
WALRUS_SUBSIDIES_OBJECT=0xb2ce8bd6e372ea93422a167b52d1ac367d080f67a6c4356334aca8e96ba0577a

Testnet Configuration

Walrus config (for --config flag):

system_object: 0x6c2547cbbc38025cf3adac45f63cb0a8d12ecf777cdc75a4971612bf97fdf6af
staking_object: 0xbe46180321c30aab2f8b3501e24048377287fa708018a5b7c2792b35fe339ee3
exchange_objects:
  - 0xf4d164ea2def5fe07dc573992a029e010dba09b1a8dcbc44c5c2e79567f39073
  - 0x19825121c52080bb1073662231cfea5c0e4d905fd13e95f21e9a018f2ef41862
  - 0x83b454e524c71f30803f4d6c302a86fb6a39e96cdfb873c2d1e93bc1c26a3bc5
  - 0x8d63209cf8589ce7aef8f262437163c67577ed09f3e636a9d8e0813843fb8bf1

Additional Sui packages and objects (for on-chain analysis):

WAL_PKG_ID=0x8270feb7375eee355e64fdb69c50abb6b5f9393a722883c1cf45f8e26048810a
WAL_EXCHANGE_PKG_ID=0x82593828ed3fcb8c6a235eac9abd0adbe9c5f9bbffa9b1e7a45cdd884481ef9f
WALRUS_SUBSIDIES_PKG_ID=0x51b0bdfba7ac0ac232402fbb36e17769e41a5c93bb1271f3b7daf1cfea81ca4c
WALRUS_SUBSIDIES_OBJECT=0x21432c30c510a27432bda9349d9f3f0aff5b84285c369f67dfd3d3ef4cf4eb35

Committee Information

The next committee is determined in the middle of the epoch. The next committee can be known after half of the epoch duration has passed. If the info all output does not include a "Next committee" section, it means that the next epoch committee has not been determined yet.

Workflow

Step 1: Set Up Configuration

1. Determine the network from the user's argument (mainnet or testnet).
2. Write the corresponding YAML config to /tmp/walrus_inspect_config.yaml using dangerouslyDisableSandbox: true. This is required because the walrus commands also run with sandbox disabled and must be able to read this file.
3. Write the appropriate config content (from the section above) to this file.

Step 2: Find the Walrus Binary

Look for the walrus binary in this order:

1. walrus (in PATH)
2. ~/.local/bin/walrus
3. ~/.walrus/bin/walrus

If not found, install it by running the installation command from the Prerequisites section (always use -n testnet). After installation, verify the binary is available before proceeding.

Step 3: Get Network Overview and Committee Health

Run both commands in parallel to get full network information and committee health as JSON. Save outputs to files for jq processing:

RUST_LOG=off walrus --config /tmp/walrus_inspect_config.yaml info all --json > /tmp/walrus_info.json 2>&1
RUST_LOG=off walrus --config /tmp/walrus_inspect_config.yaml health --committee --detail --json > /tmp/walrus_health.json 2>&1

Both commands must use dangerouslyDisableSandbox: true. After both complete, use the jq scripts from the "Parsing" sections below to extract the summary.

IMPORTANT: Always use --json for structured output. The output can be large (60-200KB for 100 nodes) — always parse with jq, never read raw files or delegate to agents.

Exception: When re-checking unreachable nodes (see Step 3b), run the health command without --json to get detailed error reasons (see below).

Parsing the info all output

The jq scripts are committed alongside this skill file. Use them directly:

jq -f .claude/skills/inspect-walrus-network/info_overview.jq /tmp/walrus_info.json

Parsing the health --committee --detail output

Step 1: Run the health analysis jq script:

jq -f .claude/skills/inspect-walrus-network/health_analysis.jq /tmp/walrus_health.json

Step 2: For any problematic nodes found above, extract detail:

jq --argjson names '["NODE_NAME_1", "NODE_NAME_2"]' \
  -f .claude/skills/inspect-walrus-network/health_detail.jq /tmp/walrus_health.json

Present the network overview and health summary to the user. Highlight any unhealthy nodes. Then notify the user that they can ask follow-up questions about individual nodes, stake distribution, shard assignments, or other network queries.

Step 3b: Re-check Unreachable Nodes

The bulk committee health check can produce transient failures (timeouts when querying 100 nodes in parallel). Always re-check nodes that appeared unreachable before reporting them as down.

1. Extract the node IDs of unreachable nodes from the initial health output.
2. Re-run the health command for just those nodes without --json:

   walrus --config "/tmp/walrus_inspect_config.yaml" health --node-ids <ID1> <ID2> ... --detail 2>&1
   

   The non-JSON output includes detailed error information, such as:
   • hyper_util::client::legacy::Error(Connect, TimedOut) — connection timed out
   • ConnectError("dns error", ...) — DNS resolution failure
   • ConnectError("tcp connect error", <IP>:<PORT>, Os { code: 61, ..., message: "Connection refused" }) — connection refused The --json output only provides a generic "error sending request for url" message and loses these details.
3. Nodes that now respond successfully were transient failures — report them as healthy.
4. For nodes that are still unreachable, always record the specific failure reason (timeout, DNS failure, connection refused) in the report.

Step 4: Answer User Queries

After presenting the overview, wait for and answer user queries. Common query types:

Global Network Status Queries

• Total nodes, total shards, current epoch
• Stake distribution analysis (which nodes have most/least stake)
• Shard distribution analysis
• Price vote analysis (storage price, write price)
• Capacity vote analysis

Answer these directly from the info all output.

Individual Node or Group Queries

For queries about specific nodes (by name, index, or node ID), use the health command:

walrus --config "/tmp/walrus_inspect_config.yaml" health --node-ids <NODE_ID> --detail --json

For multiple nodes:

walrus --config "/tmp/walrus_inspect_config.yaml" health --node-ids <NODE_ID_1> <NODE_ID_2> ... --detail --json

IMPORTANT: Always use --json for structured output. This makes it much easier to programmatically analyze node health, compare event numbers, and detect anomalies.

The health command with --detail returns per-node:

• Uptime: how long the node has been running
• Current epoch: the epoch the node is on
• Node status: Active, RecoveryInProgress, etc.
• Event progress: events persisted, events pending, highest finished event index
• Checkpoint downloading progress: latest checkpoint sequence number, estimated lag
• Shard summary: owned shards, read-only shards
• Owned shard status: counts of Ready, In transfer, In recovery, Unknown
• Owned shard details: per-shard status

Other useful health command options:

• --sort-by <status|id|name|url>: sort the output
• --desc: sort in descending order
• --node-urls <URL>...: query by node URL instead of ID
• --active-set: query all nodes in the active set

Mapping Node Names to Node IDs

The info all output includes both node names and node IDs. When the user refers to a node by name (e.g., "Mysten Labs 0"), look up the corresponding node ID from the info all output and use it with the health command.

Health Analysis: What to Look For

When analyzing the health output, check for these categories of issues:

Non-Active node status: The nonActive field in the health analysis output lists nodes whose nodeStatus is not Active (for example, RecoveryInProgress). These nodes are reachable but not fully operational. Always report these prominently — they indicate nodes that need operator attention even though they respond to health checks.

Unreachable nodes: Nodes where the health endpoint returns an error. Common failure modes:

• Connection timed out — node may be down or firewalled
• Connection refused — node process not running on the expected port
• DNS resolution failure — hostname cannot be resolved

Event processing issues: Compare events_persisted and highest_finished_event_index across all nodes. Healthy nodes should have very similar values (within single digits of each other). Watch for:

• Nodes with highest_finished_event_index: 0 — likely stuck and not processing events
• Nodes significantly behind the max events_persisted (>100K behind) — lagging
• Nodes with large events_pending (>10K) — catching up but behind
• Nodes with identical suspicious values — may be running from the same stale snapshot

Epoch mismatch: All nodes should be on the current epoch. A node on a previous epoch has failed to transition and needs operator intervention. A node with epoch mismatch and a drastically different event count from the rest of the fleet likely has its registered public host pointing at a node on a different network (e.g., testnet entry pointing at a mainnet node).

IMPORTANT: Always check for epoch-mismatch nodes before interpreting event laggard data. A single node connected to a different network can have a much higher or lower event count, skewing the max/min and making the entire fleet appear as laggards. Exclude epoch-mismatch nodes when computing event statistics.

Checkpoint lag: Compare estimated_checkpoint_lag across nodes. Nodes with significantly higher lag (e.g., 146M vs 53M) are not keeping up with the chain.

Shard issues: Check for shards in recovery or transfer. A few shards in transfer during epoch transitions is normal, but shards stuck in recovery may indicate storage problems.

Low uptime: Nodes with very low uptime (minutes instead of days) may be crash-looping. Flag these for operator investigation.

Step 5: DNS Troubleshooting

When nodes fail with DNS resolution errors, read and follow the instructions in .claude/skills/inspect-walrus-network/dns-troubleshooting.md.

Step 6: Composing Operator Notifications

When the user asks to compose a notification message, read and follow the instructions in .claude/skills/inspect-walrus-network/operator-notifications.md.

Step 7: On-Chain Analysis and Subsidies

For on-chain queries (subsidies, transactions, object history), read and follow the instructions in .claude/skills/inspect-walrus-network/on-chain-analysis.md.

Step 8: Clean Up

After all queries are answered, clean up temporary files:

rm -f /tmp/walrus_inspect_config.yaml /tmp/walrus_info.json /tmp/walrus_health.json

Output Guidelines

• Present network overview in a clear, structured format
• Highlight any anomalies (nodes with shards in recovery, unusually high pending events, etc.)
• When comparing nodes, use tables for readability
• Convert FROST to WAL where helpful (1 WAL = 1,000,000,000 FROST)
• For stake percentages, calculate relative to total network stake

Source: mystenlabs/walrus — distributed by TomeVault.