Skip to content

Commit e2a203d

Browse files
committed
docs(devlog): record what shipped for 010 and 020, and why 030 did not
1 parent 75d965b commit e2a203d

1 file changed

Lines changed: 46 additions & 0 deletions

File tree

‎devlog/_plan/260817_cursor_toolcall_decode/000_index.md‎

Lines changed: 46 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -217,3 +217,49 @@ down the middle. `010` is now smaller than it was three rounds ago.
217217
Added to the open follow-ups: the `NGHTTP2_CANCEL` zero-terminal path
218218
(`live-transport.ts:619` throws, `cursor.ts:181` swallows), confirmed by a fresh
219219
probe to produce zero adapter events. Separate defect, separate unit.
220+
221+
## Implementation status (2026-08-17)
222+
223+
| Phase | Outcome | Evidence |
224+
|-------|---------|----------|
225+
| `010` clean-EOF terminal | **SHIPPED** `54f68daf5` | 7 audit rounds; red-before-green proven; lidge 608 pass/0 fail |
226+
| `020` tool-result images | **SHIPPED** `878b067e8..cc906b0fc` | 5 review rounds; byte-equality with `4e167fd38` verified twice; lidge 624 pass/0 fail |
227+
| `030` xai apply_patch | **NOT REPRODUCED** | live probe: both `xai/grok-4.6` and `cursor/grok-4.6` used `apply_patch` successfully |
228+
229+
### What shipped for 010
230+
231+
A framed Cursor stream that ends with no terminal while a client tool call is open
232+
now fails with `CursorStreamTruncatedError` instead of settling as success. Before
233+
this, the deferred tool call emitted nothing at all: streaming degraded to
234+
`response.incomplete`, and the non-streaming path returned `"completed"` for a turn
235+
whose tool call had silently vanished. `expectedClose` (client-tool suspend) and an
236+
already-emitted terminal stay graceful.
237+
238+
### What shipped for 020
239+
240+
Tool-result images reach Cursor as real `McpImageContent`. The final design differs
241+
from the original plan in three ways, each forced by a review that proved the plan
242+
would have broken a working request:
243+
244+
1. **Bounding is post-serialization, not a byte budget.** A step is one blob shared
245+
with the call's arguments, text, and framing, so `toolCallStep` serializes and
246+
re-serializes with fewer images (oldest dropped first) until it fits the live
247+
`cursorBlobMaxEntryBytes()`.
248+
2. **Placeholders are capped to the legacy string length.** A longer replacement text
249+
could itself push a previously admissible step past the ceiling.
250+
3. **Consecutive text is newline-joined into one item**, as the legacy encoding did.
251+
Emitting one protobuf item per part added framing that overflowed at the boundary.
252+
253+
The net invariant, verified by two independent reviewers across 300 adversarial
254+
probes: **a tool result with no images serializes byte-identically to the
255+
pre-feature encoding.**
256+
257+
### Why 030 did not ship
258+
259+
A live probe spawned `xai/grok-4.6` and `cursor/grok-4.6` as subagents on the same
260+
edit task. Both edited the file and both reported using `apply_patch`. The symptom
261+
this phase was written to fix did not reproduce, so no code was written. The
262+
measurement cycle in `030` stands, and the phase closes NOOP unless the user
263+
supplies a failing case. Note the probe did not isolate the top-level freeform
264+
surface — the cursor agent reached `apply_patch` through code mode — so this is
265+
"not reproduced", not "proven absent".

0 commit comments

Comments
 (0)