[gaord]

What and why

A client that wants to show what a shell command changed has had nothing to
read. The engine records a command's writes as a command execution rather than
as a file_change item, so the only per-call attribution a client could find was
metadata.mutation, which only file tools publish — the turn's snapshot delta
covers the turn, not the call.

The engine already brackets every call that may write — a file tool, a shell
command, a program, a write-capable MCP tool — with a tool:<call_id> restore
point before it and a post-tool:<call_id> one after, and records both on the
turn. This adds the route that reads that pair:

GET /v1/threads/{id}/turns/{turn_id}/calls/{tool_call_id}/changes

It answers with the change kind, the line counts, the revision the span left and
the patch, per path, all read from the two trees rather than from the working
copy — so the answer is what the call did, not what the file holds now. Every
entry carries the restore point file-revert accepts, so a command's own write
is revertible like any other.

The span is a window of time, not attribution by cause, and the route says so
where it matters: a call the engine judged read-only took no receipts, one whose
closing snapshot was lost has half a pair, an older turn's trees may have been
pruned, and a path the snapshots do not track cannot appear at all. Each of those
is state: "unavailable" with its own reason, never an empty files list,
because "unknown" and "changed nothing" are different answers.

The tool call id is the model endpoint's own opaque string, echoed back verbatim,
so this compares it rather than parsing it: the charset a record id is held to
would refuse call_01_…|<uuid> and answer "no such call" for a call the turn
plainly recorded.

Issue

No-Issue: a new capability for Runtime API clients; nothing in the tracker
describes it.

How I tested it

• cargo test -p codewhale-tui --lib call_change_route — a route test over a real
  server against a real side-git repo: an id containing | resolves; every path,
  revision and patch comes from the span's own trees after the work tree moved on;
  a read-only call answers call_not_bounded; a half pair answers
  post_snapshot_missing; pruned trees, and a workspace with no store at all,
  answer snapshots_pruned; an unknown turn or a call the turn never ran is a 404;
  limit=0 is a 400.
• cargo test -p codewhale-tui --lib snapshot:: (101 passed), patch_between (2),
  has_tree (1), turn_artifact (7).
• cargo test -p codewhale-tui --test feature_registry — 6 passed. This adds a
  route and no flag, so no new row; docs/RUNTIME_API.md is the runtime-api
  row's doc and carries the contract.
• cargo clippy -p codewhale-tui --all-targets --all-features --locked -- -D warnings -A clippy::uninlined_format_args -A clippy::too_many_arguments -A clippy::unnecessary_map_or — clean.
• cargo fmt --all -- --check reports only crates/tui/src/child_env.rs, which
  arrived unformatted with the tip commit on main and is not touched here; every
  file this PR touches is formatted (running the formatter across it would put an
  unrelated hunk in this diff).
• Full lib suite on this branch: 14517 passed, 20 failed. Fifteen of those pass in
  isolation (load-sensitive tests), and all are in modules this change does not
  touch. On this machine core::engine::tests::yolo_mode_does_not_prompt_for_mcp_action
  overflows a test thread's stack on main as well; RUST_MIN_STACK=16M lets the
  suite run to completion.

Checklist

• One focused change, rebased on current main
• cargo fmt --all passes, and tests cover new or changed behavior
• If this adds or changes a user-facing feature, I updated its docs and its row in docs/features.toml (cargo test -p codewhale-tui --test feature_registry checks it)

[gaord]

hopefully this could land in 0.10.1. it help users know what have changed over a turn more accurately