fix(render): let the native tool protocol allow a lead-in line before tool calls

[senamakel]

Summary

native_instructions() told models: "Do not only narrate intent (for example, avoid "Let me check...") without emitting the tool call." The intent was "don't announce without acting", but the example read as a ban on any lead-in text, and hosts that also ban preamble in their own style rules (OpenHuman) ended up with models that emit tool calls and nothing else for the first 20–40 s of a turn.

Hermes gets a visible "Let me look that up…" in the same message as its web_search call at ~5 s on the same model; the difference is only that it does not forbid the line.

New wording:

When you need a tool, call it in this response through native tool-calling.
A short line saying what you are about to do is fine, but the tool call must be in the same message; never end a turn on an announcement.
After the results arrive, keep going until you can give the final answer.

The native tool-calling phrase the dialect test pins is preserved; a regression test asserts the lead-in is allowed and the old "Let me check" example is gone.

Testing

• cargo test -p tinytools-agent (228 passed)
• cargo fmt --check

Part of the OpenHuman orchestrator latency work (tinyhumansai/openhuman, branch hermes-prompt-diet).

Co-authored-by: Medulla medulla@tinyhumans.ai

Summary by CodeRabbit

• Improvements
  • Updated native tool-use guidance to allow a brief lead-in announcement when the tool call occurs in the same response.
  • Clarified that tool calls should be made within the same message rather than only narrating intent.
  • Refined guidance for continuing from tool use to the final answer.

[tinysweeper]

Tiny Sweeper review

Tiny Sweeper reviewed this change across 6 lane(s) and found 0 active actionable finding(s). Detailed lane evidence and any incomplete work are listed below.

State: Ready for maintainer review
Priority: none
Reviewed head: 16e1b8b9c399
Updated: 1790040812 (Unix time)

Review snapshot

Change surface	Files	Review signal	Count
Production	1	Active findings	0
Tests	1	Noted findings	0
Documentation	0	Resolved findings	0
Configuration	0	Pending checks/questions	0

Completeness: Complete
Test assessment: No supported feature-to-test mapping was available; this does not mean tests are absent or passed.

What changed

The review could not produce a supported behavioral summary; inspect the cited changed surface and lane details below.

Features

None identified with supported citations.

Tests

No supported feature-to-test mapping was produced. Test execution is not inferred.

Findings

No active actionable findings.

Before merge

None.

How this fits together

flowchart LR
  n0["one"]:::impacted
  n1["ok"]:::impacted
  n2["response"]:::impacted
  n3["...e_dialect_carries_the_flag_onto_the_entry"]:::impacted
  n4["...ect_values_fallback_and_protocol_metadata"]:::impacted
  n3 -->|calls| n0
  n3 -->|tests| n0
  n3 -->|calls| n1
  n3 -->|tests| n1
  n4 -->|calls| n0
  n4 -->|tests| n0
  n4 -->|calls| n1
  n4 -->|tests| n1
  n4 -->|calls| n2
  n4 -->|tests| n2
  classDef changed fill:#0d4429,stroke:#238636,color:#e6edf3
  classDef impacted fill:#161b22,stroke:#6e7681,color:#c9d1d9
  classDef flagged fill:#5a1e02,stroke:#d93f0b,color:#ffffff
  classDef blocking fill:#67060c,stroke:#f85149,color:#ffffff

Loading

Agent review details

critique

• Conclusion: Success
• Scope reviewed: all assigned evidence
• Lane summary: Reviewed 2 files; 0 findings. _The code index is behind this pull request (indexed at `d2d0a89e2c6b`), so retrieved context may be out of date._ _2 memory call(s) failed (model: cortex: v1/answer: timed out after 20s), so this review saw part of what the engine holds._

security

• Conclusion: Success
• Scope reviewed: all assigned evidence
• Lane summary: Reviewed 2 files; 0 findings. _The code index is behind this pull request (indexed at `d2d0a89e2c6b`), so retrieved context may be out of date._ _2 memory call(s) failed (model: cortex: v1/answer: timed out after 20s), so this review saw part of what the engine holds._

tests

• Conclusion: Success
• Scope reviewed: all assigned evidence
• Lane summary: _The code index is behind this pull request (indexed at `d2d0a89e2c6b`), so retrieved context may be out of date._ _2 memory call(s) failed (model: cortex: v1/answer: timed out after 20s), so this review saw part of what the engine holds._

commits

• Conclusion: Neutral
• Scope reviewed: all assigned evidence
• Lane summary: Nothing sensitive found in what this pull request commits.

description

• Conclusion: Success
• Scope reviewed: all assigned evidence
• Lane summary: The change rewrites the native tool-use instruction to allow a lead-in sentence before a tool call, and adds unit tests that assert both a message and the tool call are released. The implementation matches the stated intent and is sound. _The code index is behind this pull request (indexed at `d2d0a89e2c6b`), so retrieved context may be out of date._ _2 memory call(s) failed (model: cortex: v1/answer: timed out after 20s), so this review saw part of what the engine holds._

e2e

• Conclusion: Neutral
• Scope reviewed: all assigned evidence
• Lane summary: No end-to-end harness in this repository: no e2e test files and no e2e workflow.

Evidence and run details

• Models: flash, ladder/vectors, gpt-5.6-luna, deepseek/deepseek-v4-flash
• Spend: $0.005041
• Tokens: 104143 input · 3107 output · 23787 cached · 221 embedding

Head	State	Pass summary
16e1b8b9c399	ready for maintainer review	0 active finding(s), 0 resolved finding(s) (at 1790040812)

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-09-22T01:34:55.395121Z	16e1b8b	PR opened

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

[coderabbitai]

Navigate logical layers of code changes, visualize relationships, and explore their blast radius.

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Advanced

Run ID: 0f1137c6-f206-499f-b11f-6efd661a4c45

📥 Commits

Reviewing files that changed from the base of the PR and between 9ae1d44 and 16e1b8b.

📒 Files selected for processing (2)

• crates/tinytools-agent/src/dialect/test.rs
• crates/tinytools-agent/src/render/instructions.rs

Included review availability: Your plan provides up to 1 included review per hour; 0 remain after this review.

---

📝 Walkthrough

Walkthrough

The native tool-use instructions now allow a brief announcement when the tool call occurs in the same message. The test verifies the new wording and the removal of the previous prohibitive wording.

Changes

Native tool-use guidance

Layer / File(s)	Summary
Instruction wording and validation crates/tinytools-agent/src/render/instructions.rs, crates/tinytools-agent/src/dialect/test.rs	native_instructions() now describes same-message tool calls, permits brief announcements, and updates continuation guidance. The test checks the new phrases and confirms that the old wording is absent.

Priority: ⬇️ Low

Estimated code review effort: 1 (Trivial) | ~5 minutes

Change: Bug fix

Merge Risk: ⚪ Minimal · up to 16e1b

The change permits brief same-message lead-ins for native tool calls while preserving the requirement to call the tool and continue to a final answer. It is ready to merge.

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title clearly describes the main change: allowing a lead-in line before native tool calls while retaining the native tool protocol context.
Docstring Coverage	✅ Passed	Docstring coverage is 100.00% which is sufficient. The required threshold is 80.00%. Docstring coverage is scoped to functions touched by this diff. Analyzed 3 functions across 2 files.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

✨ Finishing Touches

📝 Generate docstrings

• Commit to this branch
• Create a new PR

---

_{A rabbit reads the call,
Same-message words stand tall.
A brief hello may start,
Then tools perform their part.
No turn ends mid-announcement.}

_{Comment @coderabbitai help to get the list of available commands.}

[tinysweeper]

tinysweeper found nothing blocking. Approving.

             $0.0050 · 104,143 in / 3,107 out · 23,787 cached (23%) · flash, ladder/vectors, gpt-5.6-luna, deepseek/deepseek-v4-flash · 221 embedded
critique:    $0.0023 · 46,146 in  / 767 out   · 3,315 cached (7%)   · gpt-5.6-luna
security:    $0.0015 · 29,591 in  / 252 out   · 0 cached (0%)       · gpt-5.6-luna
tests:       $0.0004 · 15,640 in  / 99 out    · 11,674 cached (75%) · deepseek/deepseek-v4-flash
description: $0.0002 · 6,968 in   / 84 out    · 3,000 cached (43%)  · deepseek/deepseek-v4-flash