| name | big-return-skill |
| description | Parse oversized MCP tool output. When a tool returns "Output has been saved to <file>.txt" because the result exceeded Claude Code's token limit, run this skill's parser instead of re-reading the file in chunks. Covers `get_reachability` on large topologies and `service_snapshot` on labs with many published hostnames. |
Oversized Tool Outputs
When a tool reports:
Error: result (N characters) exceeds maximum allowed tokens.
Output has been saved to <file_path>.txt
Run the parser via the launcher:
python h.py parse_large "<file_path>"
Auto-detection picks one of three formatters based on payload shape:
- Reachability: payload carries
hosts + results (covers both get_reachability() and safe_reachability --json). The reachability formatter groups failures by source/destination, surfaces clustered suspect destinations, and preserves the host/IP inventory.
- Service snapshot: text payload starts with
=== SERVICE SNAPSHOT ===. The formatter preserves the header (resolver groups, DNS outcome groups, HTTP outcome groups, suspect clients, coverage warnings) and all per-service-device blocks in full, and condenses each per-client block to its nameservers + aggregate flags. The DNS/HTTP outcome groups already aggregate the per-host/per-URL outcomes across clients, so per-zone failure patterns (e.g. one pod's DNS resolves while peers return no_addresses) stay visible. To see raw per-row addresses for a single client, rerun service_snapshot --client <name> instead of re-reading the saved file.
- Generic: any other oversized JSON falls through to a key/structure summary; plain-text payloads are emitted truncated.
Override with --type reachability, --type service_snapshot, or --type generic only if auto-detection picks the wrong formatter.
Reading the parser output
- Rows missing
tx, rx, loss_percent, or rtt_avg_ms are NOT healthy — treat as investigation leads.
- If many sources show suspicious rows toward a small destination set, that's a shared-path symptom: check the path or the shared upstream router before blaming the source hosts.
- The parser output is triage guidance, not final proof. Confirm locally before naming a faulty device.
