Skip to content

RFC 002: tool.call events vs LLM tool-use #3

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
brettin merged 2 commits into from
Jun 10, 2026
Merged

RFC 002: tool.call events vs LLM tool-use #3

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
53891ce
Add RFC 002: tool.call events vs LLM tool-use.
brettinanl Jun 4, 2026
6fb8199
Add Status section: RFC 002 implemented by PR #9
brettinanl Jun 10, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
19 changes: 19 additions & 0 deletions contracts/rfcs/002-tool-call-events-vs-llm-tool-use.md
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,19 @@
# RFC 002: tool.call events vs LLM tool-use

## Status

**Implemented.** The normative substance of this RFC has already been addressed in `main` by PR #9 ("Tighten RunInvocation companion contract"). This document is retained as a historical / design-rationale record.

- Event-semantics clarification (model turn vs executed tool) landed in `contracts/companion/run-invocation-interop.md`, which now states that a `RunInvocation` records MAG/model turns while executed-tool telemetry remains in the event stream, with tool results appearing as later `role:"tool"` messages linked by `tool_call_id`.
- Budget-field alignment (`maxToolCalls`) landed in the same file: the `## maxToolCalls semantics` section was rewritten from "model-returned `tool_calls` batches per MAG invoke" to "model-returned tool call items per MAG invoke", matching the metering behavior described below.

## Problem

- Spec `tool-call-event` is audit telemetry (`toolName`, `status`, `latencyMs`, `tokenCost`).
- Platform `run.outputs.tool_calls` is model-requested function calls.
- `maxToolCalls` on budgets: platform counts MAG-returned batches per run, not local agent execution.

## Proposal

- Clarify event semantics (model turn vs executed tool).
- Align budget field documentation with metering behavior.