fix(tool_search): accept sanitized keys in deferred tool_call probe

[Adolanium]

What does this PR do?

The Tool Search blind-call probe (validate_deferred_call_args) checked registry wire required names against the nested tool_call arguments. After schema sanitization, the model and tool_describe use renamed keys (for example issue_class_neq for Cloudflare-style issue_class~neq). A complete call with those model-facing keys was treated as missing required fields and never reached unrename_tool_args / dispatch.

That contradicts the probe's contract that valid calls are never blocked, and it breaks deferred MCP tools whose property keys need sanitizer renames (the Cloudflare path #67034 already handles at dispatch).

This PR reverse-maps args with the same unrename_tool_args path coerce_tool_args uses before the key-absence check. Probe error payloads list model-facing missing names and the sanitized parameters object so a repair round-trip matches tool_describe. Fail-open on unrename/validator errors is unchanged. Clean-key tools are unaffected.

Related Issue

Fixes #72648

Type of Change

• Bug fix (non-breaking change that fixes an issue)
• New feature (non-breaking change that adds functionality)
• Security fix
• Documentation update
• Tests (adding or improving test coverage)
• Refactor (no behavior change)
• New skill (bundled or hub)

Changes Made

• tools/tool_search.py: unrename deferred tool_call args before required-key checks in validate_deferred_call_args. Build model-facing parameters for probe errors via the schema sanitizer so error text and parameters match tool_describe.
• tests/tools/test_tool_search.py: add TestDeferredCallSchemaProbe cases for sanitized required keys (accept, still-block missing, end-to-end dispatch, blind-call error naming), using the flat MCP schema shape.

How to Test

1. python -m pytest tests/tools/test_tool_search.py::TestDeferredCallSchemaProbe -q - 10 passed
2. python -m pytest tests/tools/test_tool_search.py tests/tools/test_schema_sanitizer.py -q - 112 passed
3. Proof of the bug on main: register an MCP-shaped tool with required issue_class~neq, call handle_function_call("tool_call", arguments={"issue_class_neq": "...", "zone_id": "..."}). On main the probe returns missing issue_class~neq. With this branch the tool dispatches.
4. Existing clean-key cases still block empty blind calls and still dispatch when document_id is present.

Checklist

Code

• I've read the Contributing Guide
• My commit messages follow Conventional Commits (fix(scope):, feat(scope):, etc.)
• I searched for existing PRs to make sure this isn't a duplicate
• My PR contains only changes related to this fix/feature
• I've run pytest tests/tools/test_tool_search.py tests/tools/test_schema_sanitizer.py -q and all tests pass (112). Full suite not re-run here. Change is isolated to tool_search probe + tests.
• I've added tests for my changes
• I've tested on my platform: Windows 11

Documentation & Housekeeping

• I've updated relevant documentation (README, docs/, docstrings) - or N/A (docstring update only in the probe)
• I've updated cli-config.yaml.example if I added/changed config keys - or N/A
• I've updated CONTRIBUTING.md or AGENTS.md if I changed architecture or workflows - or N/A
• I've considered cross-platform impact (Windows, macOS) per the compatibility guide - or N/A (pure Python key mapping)
• I've updated tool descriptions/schemas if I changed tool behavior - or N/A (dispatch shape unchanged for successful calls)

Screenshots / Logs

# Before (main): sanitized keys blocked
{"error": "tool_call to 'mcp_cf_probe_filter' is missing required argument(s): issue_class~neq. The tool was NOT invoked.", ...}

# After: same call dispatches
{"ok": true, "issue_class_neq": "bug", "zone_id": "z1"}

[teknium1]

Thanks for the focused regression fix. Current main still has the key-namespace mismatch: tools/tool_search.py:973 checks registry-wire required names against model arguments, while schemas are sanitized before Tool Search assembly (model_tools.py:546-547) and are reverse-mapped only at dispatch (model_tools.py:734-740). The shared validator is also used by both agent executor paths (agent/tool_executor.py:713 and agent/tool_executor.py:1387).

Suggested changes

• During salvage, retain current main's tool_error(...) envelope at tools/tool_search.py:976-984 (introduced by 1a7f73b8ea) while applying the proposed model-facing parameter/error mapping.
• Restore the added regression cases beside TestDeferredCallSchemaProbe at tests/tools/test_tool_search.py:450-518; 39975613b1 pruned the former insertion context.

This is an automated hermes-sweeper review.

[Adolanium]

Salvaged onto current main in dcbb69c. The probe unrenames model args with the same unrename_tool_args path coerce_tool_args uses before the required-key check, so a call that matches tool_describe no longer gets rejected as missing required fields. The error keeps the tool_error envelope from 1a7f73b, now carrying the model-facing parameter schema and sanitized missing names so a repair round-trip matches what tool_describe shows.

The regression cases are restored beside TestDeferredCallSchemaProbe. tests/tools/test_tool_search.py passes 31/31.

[GottZ]

This was generated by AI during triage.

Summary

One PR addresses #72648. #72650 fixes the reported namespace mismatch by reverse-mapping sanitized model-facing keys before required-field validation and returning model-facing names and parameters in repair errors.

Related pull requests

• fix(tool_search): accept sanitized keys in deferred tool_call probe #72650 best fix — (+186/-4) — keep open with a salvage path: The diff directly corrects the probe/schema mismatch with unrename_tool_args, preserves rejection of genuinely missing fields, retains the current tool_error(...) envelope, and adds focused validation, dispatch, and error-schema regressions. Consistent with the automated keep_open review on fix(tool_search): accept sanitized keys in deferred tool_call probe #72650, the salvaged diff now contains both requested elements: the preserved error envelope and restored tests beside TestDeferredCallSchemaProbe.

Suggested consolidation

Keep #72650 open with a salvage path: retain the reverse-mapping and model-facing error-schema changes in tools/tool_search.py together with the four regression cases in tests/tools/test_tool_search.py. It is the only PR in this complex, directly addresses #72648, and has no duplicate PRs to close.

Complex graph

flowchart LR
    classDef open fill:#dbeafe,stroke:#1d4ed8,color:#1e3a8a
    classDef merged fill:#dcfce7,stroke:#15803d,color:#14532d
    classDef closed fill:#e5e7eb,stroke:#6b7280,color:#1f2937
    classDef unverified fill:#f3f4f6,stroke:#9ca3af,color:#374151
    classDef best stroke-width:3px,stroke:#b45309
    classDef target stroke-width:3px,stroke:#4338ca
    I72648(["issue #72648 (open)"])
    P72650["PR #72650 (open)"]
    P72650 -->|best fix| I72648
    class I72648 open
    class P72650 open
    class P72650 best
    class P72650 target
    click I72648 "https://github.com/NousResearch/hermes-agent/issues/72648"
    click P72650 "https://github.com/NousResearch/hermes-agent/pull/72650"

Loading

Graph: solid arrow = fixes / best fix, dashed arrow = partial or unverified (see edge label); boxed group = PRs duplicating each other; amber border = best fix; indigo border = target; gray node = closed (state tag in the node label).

Cross-PR triage: Reviewed 1 pull request and 1 issue in this complex. Each diff was read against this issue; Assessment working set: 10 kB of PR diffs, 9 kB of issue/PR text, <1 kB of discussion (1 comments), 2 verify verdicts. verdicts reflect diff content, not PR titles. Part of an automated triage batch.