fix(computer-use): stop dropping tool name when action is present

[gzak]

Summary

_derive_connector_type(tool_name, action) in axonflow/adapters/computer_use.py discarded tool_name whenever action was present, returning only f"computer_use.{action}". Every actioned Computer Use tool call (e.g. computer with action="left_click") collapsed to the same connector_type shape regardless of which actual tool triggered it — connector_type couldn't distinguish tool identity for any actioned tool. The non-actioned branch correctly preserved tool_name; only the actioned branch had the bug.

Fixes #2910 (epic #2905). Depends on the same platform two-field (server, tool) identity contract as #2906 (platform sub-issue #2904, R3-approved/CI-green, not yet merged to axonflow-enterprise).

Design decision (as requested in the sub-issue)

The two-field contract has exactly two slots (connector_type, tool), but there were three meaningful pieces of information floating around: the "computer_use" domain marker (previously just a hardcoded string prefix), tool_name (e.g. "computer", "bash", "text_editor"), and action (e.g. "left_click", "screenshot", only present for some tools).

I mapped:

• connector_type = tool_name (bare, e.g. "computer") — matches the rest of the SDK's existing convention for connector_type values (bare category names like "postgres", "snowflake", no namespace prefix — see langgraph.py's MCPInterceptorOptions default of request.server_name). The "computer_use." prefix is dropped since it added no distinguishing information once tool_name is preserved in full.
• tool = action (when present, else unset) — this is the piece that genuinely varies within a single tool_name (e.g. "computer" can be "left_click" or "screenshot"), so it fits the "tool" slot's role as the finer-grained identity component.

This means tool_name is never dropped (previously it was, whenever action was set), and action is preserved as its own field instead of being folded into a single derived string. _derive_connector_type is removed entirely; ComputerUseGovernor.check_tool_use now passes connector_type=name, tool=tool directly to mcp_check_input. check_result() has no action available at its call site (its signature only takes tool_name), so tool is simply left unset there — no regression, since that call site never had action info to begin with.

Also (shared with #2906, since both fixes need the wire-level tool field and are independent/unordered PRs):

• axonflow.client.Client.mcp_check_input / mcp_check_output (async and sync) gain a tool: str | None = None parameter, sent as its own tool field only when non-empty.
• MCPCheckInputRequest / MCPCheckOutputRequest gain a matching optional tool field.
• tests/fixtures/wire_shape_baseline.json acknowledges tool as an SDK-superset field on both request models (mirrors the existing content_type entry) until #2904 merges to axonflow-enterprise. Justified per CONTRIBUTING.md's baseline burndown policy — blocked on the platform-side PR landing, not something this PR can burn down.

Module docstring at the top of computer_use.py was reviewed for staleness — it doesn't reference connector_type derivation at all, so no update was needed there.

Test plan

• Replaced TestDeriveConnectorType (tested the removed helper directly) with TestConnectorTypeToolSplit, asserting connector_type and tool are sent as two separate, correct values for actioned and non-actioned tools, including a test that two calls to the same tool with different actions produce the same connector_type but different tool values (exactly what the old bug broke).
• Updated existing TestCheckToolUse / TestCheckResult assertions that referenced the old "computer_use.*" concatenated shape.
• pytest tests/test_computer_use.py — 29 passed
• Full suite: pytest — 1047 passed, 29 skipped
• ruff check / ruff format --check on changed files — clean
• mypy axonflow — no new errors (one pre-existing unrelated error in heartbeat.py)

[saurabhjain1592]

R3 review (master + independent hostile pass). The core fix is verified correct — previously _derive_connector_type discarded the tool name whenever an action existed; now both survive, with good unit + live-wire coverage (actioned/non-actioned/same-tool-different-action/check_result). Must-fix:

BLOCKER — CI is red

lint fails (run 29540706653) with 12 NEW falsey-clobber findings, mostly in files this PR doesn't touch (client.py:1682/1723/1781/2299, computer_use.py:175/209) — line-number baseline drift. Regenerate .lint_baselines/falsey_clobber.json in this PR.

MEDIUM — semantic drift needs an epic-level ruling before merge

This PR sends tool = ACTION (left_click) with connector_type = tool name (computer, bash), while the four langgraph PRs send tool = tool name. On a new platform, audit rows here read tool_server="bash" / tool_name="left_click" — bash is not a server, and the meaning of the tool_name audit column now depends on which adapter emitted it. If intended, record the decision on epic #2905; if not, tool should carry the tool name and the action belongs elsewhere (e.g. operation/statement).

MEDIUM — stale contract claims + #216 collision

wire_shape_baseline.json note + types.py comment say #2904 is "not yet merged" — it merged 2026-07-16 (enterprise c8df2006b). Update, and reconcile with #216's edits to the same lines (guaranteed conflict; #216's text is the correct one).

MEDIUM — hard policy break with no escape hatch

connector_type loses the computer_use. namespace entirely (computer_use.bash → bash) on EVERY platform version, so existing policies/per-connector settings scoped to computer_use.* stop applying — and this adapter has no connector_type_fn escape hatch. The changelog migration note is the only mitigation; it's currently absent. Add it (Go #189 has the model text).