fix(agent): honor max_len in build_tool_preview for values <= 3 (#9439)

[LarHope]

Closes #9439

What does this PR do?

agent.display.build_tool_preview() unconditionally truncated with preview[:max_len - 3] + \"...\". When max_len is 1, 2, or 3, max_len - 3 is ≤ 0, so the slice returns nearly the entire source string and the result ends up far longer than the caller asked for — violating the documented cap.

This PR special-cases max_len <= 3 to emit \".\" * max_len, so the return value is always bounded by max_len.

Related Issue

Fixes #9439

Type of Change

• 🐛 Bug fix (non-breaking change that fixes an issue)

Changes Made

• agent/display.py — add max_len <= 3 branch in build_tool_preview() (+4/-1 lines).
• tests/agent/test_display.py — add three regression tests (max_len ∈ {1, 2, 3}, exactly-4 boundary, no-truncation path) and tighten the existing test_long_value_truncated assertion from <= 43 to <= 40 so the contract is actually enforced.

Total diff: 2 files, +27/-3.

How to Test

Reproduction from the issue:

```python
from agent.display import build_tool_preview
for ml in (1, 2, 3, 4, 10):
r = build_tool_preview('terminal', {'command': 'abcdefghijklmnopqrstuvwxyz'}, max_len=ml)
print(f'max_len={ml}: {r!r} len={len(r)}')
```

Before this PR:
```
max_len=1: 'abcdefghijklmnopqrstuvwx...' len=27
max_len=2: 'abcdefghijklmnopqrstuvwxy...' len=28
max_len=3: 'abcdefghijklmnopqrstuvwxyz...' len=29
```

After:
```
max_len=1: '.' len=1
max_len=2: '..' len=2
max_len=3: '...' len=3
max_len=4: 'a...' len=4
max_len=10: 'abcdefg...' len=10
```

Run the full display test module:

```bash
pytest tests/agent/test_display.py -v
```

All 27 tests pass (24 existing + 3 new regressions).

Checklist

Code

• I've read the Contributing Guide
• My commit messages follow Conventional Commits
• I searched for existing PRs — none touching build_tool_preview
• My PR contains only changes related to this fix
• I've run the display test module and all tests pass
• I've added regression tests for the bug
• Tested on Ubuntu 24.04 (Python 3.11)

Documentation & Housekeeping

• No doc changes needed — N/A
• No config keys changed — N/A
• No architecture/workflow changes — N/A
• Pure-Python string logic, no cross-platform concerns — N/A
• No tool behavior changes — N/A

[alt-glitch]

Likely duplicate of #9497 — same fix for build_tool_preview() max_len<=3 edge case (#9439). Competing PRs.

[teknium1]

Thanks for the focused regression fix. The issue remains on current main, but the implementation needs to be retargeted after later display refactoring.

Problems

• agent/display.py:476-482 now sends terminal and execute_code previews through _truncate_preview, which already handles limits <= 3. Therefore the submitted terminal tests would pass while the remaining generic branch at agent/display.py:545-546 still performs the unsafe preview[:max_len - 3] + "..." truncation.

Suggested changes

• Salvage the change at agent/display.py:545-546 by calling _truncate_preview(preview, max_len).
• Add the tiny-limit regression against a generic path such as web_search, which exercises that remaining branch.

Automated hermes-sweeper review.

[teknium1]

Current main now routes terminal through _truncate_preview, so this branch no longer covers the PR's terminal repro. Please apply the shared helper to the current generic fallback branch (agent/display.py:545-546) and add a generic-tool regression such as web_search.

[teknium1]

Salvaged onto current main as #109134 with your commit authorship preserved where the diff still applied (the file moved/was restructured since April, so some of it is a hand-port with credit in the commit message). Once #109134 merges this PR will be closed with a link to the landed SHA. Thanks for the fix.

[teknium1]

Landed via #109134 (merge 9d09598) with your commit cherry-picked so authorship is preserved — thank you @LarHope. Closing this PR as superseded by the merged salvage; the fix is on main now.