fix(honcho): skip dialectic char cap for explicit honcho_reasoning tool calls

[LavyaTandel]

What

Skip the dialecticMaxChars truncation cap when honcho_reasoning is called explicitly by the model. Previously, tool results were silently clipped to 600 chars mid-word even when the model spent a turn requesting a full synthesized answer.

Why

dialectic_query() shared one truncation path for two call sites:

1. Auto-injection (system prompt supplement) — tight cap correct
2. honcho_reasoning tool — model explicitly requested a full answer, no reason to clip

The server returns untruncated results; the clip is purely client-side.

Fix

Add apply_injection_cap: bool = True parameter to dialectic_query(). The honcho_reasoning tool handler passes False. Auto-injection path unchanged (default True).

Files changed:

• plugins/memory/honcho/session.py: new parameter + guard on truncation
• plugins/memory/honcho/__init__.py: tool handler passes apply_injection_cap=False

Runtime Proof

Before: honcho_reasoning result truncated at 600 chars with trailing …
After: full server response returned, no truncation

Duplicate Scan

No existing PRs for #59469.

Closes #59469

[alt-glitch]

This was generated by AI during triage.

Duplicate of #59471 (earlier, same fix). Both add apply_injection_cap to dialectic_query() in plugins/memory/honcho/session.py and pass apply_injection_cap=False from the honcho_reasoning tool handler — same mechanism, same code sites. #59471 was opened first, so it's canonical.

[teknium1]

Thanks for isolating the explicit-tool path from the auto-injection path. Current main still has the shared unconditional cap at plugins/memory/honcho/session.py:658-660, and honcho_reasoning reaches it through plugins/memory/honcho/__init__.py:1358-1362.

Problems

• tests/honcho_plugin/test_session.py:501-506 asserts the exact dialectic_query() call without apply_injection_cap; this PR adds that keyword at plugins/memory/honcho/__init__.py:1362, so the existing test will fail as written.
• The diff has no behavioral regression test covering an over-cap response for both the default injection path and the explicit tool path.

Suggested changes

• Update the exact-call assertion to include apply_injection_cap=False.
• Add a focused long-response test proving the default still caps results and the explicit-tool flag does not.

This is an automated hermes-sweeper review.

[teknium1]

Please update tests/honcho_plugin/test_session.py:501-506: it uses assert_called_once_with(...) for this call and currently omits this new keyword, so the existing test will fail after this change.

[teknium1]

This is now fixed on main via #62290 (commit 8d1c96f), which adopted the earliest submission for #59469 (#59471 by @vizi0uz, submitted ~45 min before this one). Explicit honcho_reasoning tool calls no longer get clipped to dialecticMaxChars — the cap is scoped to automatic context injection only. Thanks for the fix, sorry it collided with an earlier duplicate.