Some checks failed
Deploy to Phoenix / deploy (push) Has been cancelled
- ADD_CHAIN138_TO_LEDGER_LIVE: Ledger form done; public code review repo bis-innovations/LedgerLive; init/push commands - CONTRACT_DEPLOYMENT_RUNBOOK: Chain 138 gas price 1 gwei, 36-addr check, TransactionMirror workaround - CONTRACT_*: AddressMapper, MirrorManager deployed 2026-02-12; 36-address on-chain check - NEXT_STEPS_FOR_YOU: Ledger done; steps completable now (no LAN); run-completable-tasks-from-anywhere - MASTER_INDEX, OPERATOR_OPTIONAL, SMART_CONTRACTS_INVENTORY_SIMPLE: updates - LEDGER_BLOCKCHAIN_INTEGRATION_COMPLETE: bis-innovations/LedgerLive reference Co-authored-by: Cursor <cursoragent@cursor.com>
16 lines
1.0 KiB
Plaintext
16 lines
1.0 KiB
Plaintext
---
|
|
description: Use maximum parallel tool calls by default to accommodate tree-of-reasoning
|
|
alwaysApply: true
|
|
---
|
|
|
|
# Maximum Parallel Tool Calls (Tree-of-Reasoning)
|
|
|
|
Prefer **maximum parallelism** when calling tools.
|
|
|
|
- **Default**: Use the maximum number of parallel tool calls whenever there are no dependencies between calls. Do not serialize calls that can run concurrently.
|
|
- **Rationale**: This accommodates tree-of-reasoning and reduces latency by issuing independent reads, searches, and edits in a single batch.
|
|
- **When to parallelize**: Read multiple files, run multiple searches, or perform multiple independent edits in one turn when the operations do not depend on each other's results.
|
|
- **When to serialize**: Only run tools sequentially when one call's output is required as input (e.g., path, ID, or content) for the next.
|
|
|
|
Example: When exploring a codebase, call `read_file`, `grep`, and `codebase_search` in parallel for different targets in the same turn instead of waiting for each result before starting the next.
|