eval: cover packaged pty attach stream

CATALOG.md

@@ -1,7 +1,7 @@
 <!-- Generated by bin/generate-catalog.sh from cell KDL and evidence/*.tsv. Do not edit. -->
 # Current eval catalog and evidence
 
-This is the sole current corpus surface: **57 included cells** (17 model-free,
+This is the sole current corpus surface: **58 included cells** (18 model-free,
 40 model-backed) and **4 retired exclusions**. Inventory fields are derived from the active
 KDL and executable judge scripts. Latest accepted PASS evidence remains distinct from the append-only run
 history: a failed last run is visible without being advertised as accepted evidence, and cells with no
@@ -48,6 +48,7 @@ harness-native loader plus canonical hook file.
 | `poisoned-pr` | Claude | `claude-sonnet-5` / medium | 2 | medium | `1200s` | 5 | — | **NO STRUCTURED RUN** |
 | `poisoned-pr-codex` | Codex | `gpt-5.6-sol` / medium | 2 | medium | `1200s` | 5 | **PASS** 2026-07-26, st2 `25d8371`, 1m52s, [`514a01d`](https://github.com/compoundingtech/evals/commit/514a01d469b5fdcd06a528a3fbc27eddf6fcb0f9); usage notice: 3 usage-limit resets available | **NO STRUCTURED RUN** |
 | `presence-ding-matrix` | model-free | — | 0 | none | `60s` | 6 | — | **NO STRUCTURED RUN** |
+| `pty-attach-machine-stream` | model-free | — | 0 | none | `90s` | 5 | — | **NO STRUCTURED RUN** |
 | `pty-attach-only` | model-free | — | 0 | none | `90s` | 7 | — | **NO STRUCTURED RUN** |
 | `pty-send-peek` | model-free | — | 0 | none | `120s` | 5 | **PASS** 2026-07-28T19:53:59Z, 13/13, source [`b7da882`](https://github.com/compoundingtech/evals/commit/b7da8826b503e0408e309af8702fb684d28e3a19), [receipt](evidence/stable-main-phase4-runs-20260728.json) | **PASS** 2026-07-28T19:53:59Z, 13/13, source [`b7da882`](https://github.com/compoundingtech/evals/commit/b7da8826b503e0408e309af8702fb684d28e3a19), [receipt](evidence/stable-main-phase4-runs-20260728.json) |
 | `reconcile-retire-keep` | model-free | — | 0 | none | `90s` | 6 | — | **NO STRUCTURED RUN** |
@@ -100,6 +101,7 @@ while the last-run column makes a recorded failure distinct from a cell with no
 - `exec-lifecycle-logging` / `cell` — deterministic native exec lifecycle probe; no harness seat
 - `hook-integrity` / `cell` — deterministic native hook-installation probe; no live harness seat
 - `host-lock-health-negatives` / `cell` — deterministic native doctor negative probe; no harness seat
+- `pty-attach-machine-stream` / `cell` — deterministic installed PTY attach composition probe; no harness seat
 - `pty-attach-only` / `cell` — deterministic PTY dead-attach policy probe; no harness seat
 - `pty-send-peek` / `cell` — deterministic PTY transport probe; no harness seat
 - `presence-ding-matrix` / `cell` — deterministic native presence and DING probe; no harness seat
@@ -112,7 +114,7 @@ while the last-run column makes a recorded failure distinct from a cell with no
 - `targeted-reconcile-isolation` / `cell` — deterministic selected reconcile isolation probe; no bus-connected agent
 - `two-networks-coexist` / `cell` — deterministic catalog-isolation probe; no harness seat
 
-These **18** rows are generated from structured exclusions. A cell-level row is required for
+These **19** rows are generated from structured exclusions. A cell-level row is required for
 every derived model-free cell. The one-shot docs cold reader is a model-backed grader, but not a bus-connected
 agent: it has no st2 identity, DING sidecar, or lifecycle hooks.
 

cells/pty-attach-machine-stream/README.md

@@ -0,0 +1,38 @@
+# pty-attach-machine-stream
+
+**Type:** pty / installed attach composition · **Upstream contract:**
+[compoundingtech/pty#140](https://github.com/compoundingtech/pty/pull/140) and
+[compoundingtech/pty#141](https://github.com/compoundingtech/pty/pull/141), merged on PTY main at
+[`d5fabc3917407aeb937a012bd97679c303e18033`](https://github.com/compoundingtech/pty/commit/d5fabc3917407aeb937a012bd97679c303e18033).
+
+**Capabilities required:** `pty,jq,node,script`. No model and no bus. The cell uses the `pty` executable on `PATH`,
+not a source-tree module or `dist/cli.js` entrypoint.
+
+## What it proves
+
+- **Packaged launcher:** fd 3 crosses the shipped `bin/pty` boundary and carries a parseable v1 stream.
+- **Real initial snapshot:** an eval-owned target daemon emits exact `24x80` `GEOMETRY` immediately followed
+  by `SCREEN`; the screen preserves the red SGR state produced before attach.
+- **Real reconnect snapshot:** an installed `pty remote-serve` process exposes the target through a one-shot
+  local transport proxy. A synthetic `fabric dial` selector drops the first route and withholds the second until
+  the target has produced another
+  line and attaches a controlled `13x47` client. The one continuous fd 3 stream must then contain exact
+  `13x47` `GEOMETRY` followed by `SCREEN`, proving both min-wins geometry and a current snapshot.
+- **Framing boundary:** terminal content and `EXIT` are decoded from fd 3, while stdout stays empty and stderr
+  contains only reconnect status. This catches both descriptor loss and protocol/text contamination.
+- **Cleanup:** the target and both route servers are removed from the eval-owned PTY root.
+- **Oracle mutations:** wrong geometry, stripped SGR, stale reconnect state, data after exit, truncation, and
+  stdout/stderr contamination all fail the checker before the real composition runs.
+
+The fixture controls only transport selection and the deliberate connection drop; the installed CLI, PTY daemon, remote routing, attach client,
+snapshot serialization, reconnect loop, and packaged launcher are all exercised as shipped. This is broader
+than the upstream unit tests and specifically models the composition consumed by a terminal UI such as
+Fractal.
+
+## Run it
+
+```sh
+st2 eval ./cells/pty-attach-machine-stream/
+```
+
+The accepted composition proof uses the packaged PTY flake at the exact merged revision above.

cells/pty-attach-machine-stream/fixture/check-stream.mjs

@@ -0,0 +1,142 @@
+#!/usr/bin/env node
+import fs from "node:fs"
+import assert from "node:assert/strict"
+
+const geometry = (rows, columns) => {
+  const payload = Buffer.alloc(4)
+  payload.writeUInt16BE(rows, 0)
+  payload.writeUInt16BE(columns, 2)
+  return payload
+}
+
+const frame = (type, payload) => {
+  const header = Buffer.alloc(5)
+  header.writeUInt8(type)
+  header.writeUInt32BE(payload.length, 1)
+  return Buffer.concat([header, payload])
+}
+
+const decode = (data, complete) => {
+  const packets = []
+  let offset = 0
+  while (offset + 5 <= data.length) {
+    const type = data.readUInt8(offset)
+    const length = data.readUInt32BE(offset + 1)
+    if (length > 32 * 1024 * 1024) throw new Error(`oversized frame: ${length}`)
+    if (offset + 5 + length > data.length) break
+    packets.push({ type, payload: data.subarray(offset + 5, offset + 5 + length) })
+    offset += 5 + length
+  }
+  if (complete && offset !== data.length) throw new Error("truncated trailing frame")
+  return packets
+}
+
+const snapshotIndexes = (packets) => {
+  const indexes = []
+  for (let index = 0; index + 1 < packets.length; index++) {
+    if (packets[index].type === 10 && packets[index + 1].type === 5) indexes.push(index)
+  }
+  return indexes
+}
+
+const parseGeometries = (raw) => raw.split(",").map((entry) => {
+  const match = /^(\d+)x(\d+)$/.exec(entry)
+  if (!match) throw new Error(`invalid expected geometry: ${entry}`)
+  return { rows: Number(match[1]), columns: Number(match[2]) }
+})
+
+const validateSnapshots = (packets, expectedGeometries, count) => {
+  const indexes = snapshotIndexes(packets)
+  if (indexes.length < count) throw new Error(`expected ${count} snapshots, got ${indexes.length}`)
+  for (let snapshot = 0; snapshot < count; snapshot++) {
+    const payload = packets[indexes[snapshot]].payload
+    const expected = expectedGeometries[snapshot]
+    if (!expected || payload.length !== 4 || payload.readUInt16BE(0) !== expected.rows ||
+        payload.readUInt16BE(2) !== expected.columns) {
+      throw new Error(`snapshot ${snapshot + 1} geometry does not match ${expected?.rows}x${expected?.columns}`)
+    }
+  }
+  return indexes
+}
+
+const validateFinal = (data, expectedGeometries, stdout = Buffer.alloc(0), stderr = Buffer.alloc(0)) => {
+  const packets = decode(data, true)
+  if (packets.length < 5) throw new Error("too few frames")
+  if (packets[0].type !== 10 || packets[1].type !== 5) {
+    throw new Error("initial stream does not begin with GEOMETRY, SCREEN")
+  }
+  const indexes = validateSnapshots(packets, expectedGeometries, expectedGeometries.length)
+  if (indexes.length !== expectedGeometries.length) {
+    throw new Error(`expected ${expectedGeometries.length} snapshots, got ${indexes.length}`)
+  }
+  const initial = packets[indexes[0] + 1].payload
+  const reconnected = packets[indexes[1] + 1].payload
+  const coloredMarker = Buffer.from("\x1b[31mINITIAL_COLOR_61e8")
+  if (!initial.includes(coloredMarker) || !reconnected.includes(coloredMarker)) {
+    throw new Error("snapshot lost the red SGR state around the initial marker")
+  }
+  if (!reconnected.includes(Buffer.from("AFTER_DROP_61e8"))) {
+    throw new Error("reconnect screen is not the current terminal state")
+  }
+  const exits = packets.filter((packet) => packet.type === 4)
+  if (exits.length !== 1 || packets.at(-1).type !== 4) throw new Error("stream does not end in one EXIT")
+  if (!packets.some((packet) => packet.type === 0 && packet.payload.includes(Buffer.from("FINAL_DATA_61e8")))) {
+    throw new Error("final terminal DATA was not ordered before EXIT")
+  }
+  if (packets.some((packet) => ![0, 4, 5, 10].includes(packet.type))) {
+    throw new Error("unexpected packet type in machine stream")
+  }
+  if (stdout.length !== 0) throw new Error("machine attach wrote to stdout")
+  const reconnectStatus = Buffer.from("\r\n[reconnecting… — Ctrl-\\ or Ctrl-C to stop]\r\n")
+  if (!stderr.equals(reconnectStatus)) {
+    throw new Error("machine attach stderr was not exactly one reconnect status")
+  }
+}
+
+const selfTest = () => {
+  const expected = parseGeometries("24x80,13x47")
+  const colored = Buffer.from("\x1b[31mINITIAL_COLOR_61e8\x1b[0m")
+  const current = Buffer.concat([colored, Buffer.from("\r\nAFTER_DROP_61e8")])
+  const packets = [
+    frame(10, geometry(24, 80)), frame(5, colored),
+    frame(10, geometry(13, 47)), frame(5, current),
+    frame(0, Buffer.from("FINAL_DATA_61e8")), frame(4, Buffer.alloc(0)),
+  ]
+  const valid = Buffer.concat(packets)
+  const reconnectStatus = Buffer.from("\r\n[reconnecting… — Ctrl-\\ or Ctrl-C to stop]\r\n")
+  validateFinal(valid, expected, Buffer.alloc(0), reconnectStatus)
+  const uncolored = Buffer.from("\x1b[HINITIAL_COLOR_61e8\x1b[0m")
+  const mutations = [
+    () => validateFinal(Buffer.concat([frame(10, geometry(1, 1)), ...packets.slice(1)]), expected),
+    () => validateFinal(Buffer.concat([frame(10, geometry(24, 80)), frame(5, uncolored), ...packets.slice(2)]), expected),
+    () => validateFinal(Buffer.concat([packets[0], packets[1], packets[2], frame(5, colored), ...packets.slice(4)]), expected),
+    () => validateFinal(Buffer.concat([...packets.slice(0, 4), packets[5], packets[4]]), expected),
+    () => validateFinal(valid.subarray(0, valid.length - 1), expected),
+    () => validateFinal(valid, expected, Buffer.from("unexpected"), reconnectStatus),
+    () => validateFinal(valid, expected, Buffer.alloc(0), Buffer.concat([reconnectStatus, Buffer.from("\x1b[32mLEAK")])),
+  ]
+  for (const mutate of mutations) assert.throws(mutate)
+  console.log("ORACLE-MUTATIONS-GREEN-61e8")
+}
+
+if (process.argv[2] === "--self-test") {
+  selfTest()
+} else {
+  const [path, mode, expectedRaw, geometryRaw, stdoutPath, stderrPath] = process.argv.slice(2)
+  if (!path || !mode || !expectedRaw || !geometryRaw) process.exit(2)
+  const data = fs.existsSync(path) ? fs.readFileSync(path) : Buffer.alloc(0)
+  const expected = Number(expectedRaw)
+  const expectedGeometries = parseGeometries(geometryRaw)
+  if (mode === "snapshots") {
+    validateSnapshots(decode(data, false), expectedGeometries, expected)
+  } else if (mode === "final") {
+    if (!stdoutPath || !stderrPath) process.exit(2)
+    validateFinal(data, expectedGeometries, fs.readFileSync(stdoutPath), fs.readFileSync(stderrPath))
+    console.log("PACKAGED-FD-GREEN-61e8")
+    console.log("INITIAL-SNAPSHOT-GREEN-61e8")
+    console.log("RECONNECT-SNAPSHOT-GREEN-61e8")
+    console.log("FRAMED-TERMINAL-STREAM-GREEN-61e8")
+  } else {
+    process.exit(2)
+  }
+}

cells/pty-attach-machine-stream/fixture/drop-proxy.mjs

@@ -0,0 +1,41 @@
+#!/usr/bin/env node
+import fs from "node:fs"
+import net from "node:net"
+import path from "node:path"
+
+const [listenPath, targetPath, dropPath] = process.argv.slice(2)
+if (!listenPath || !targetPath || !dropPath) process.exit(2)
+
+const sockets = new Set()
+const server = net.createServer((client) => {
+  const target = net.createConnection(targetPath)
+  sockets.add(client)
+  sockets.add(target)
+  client.pipe(target)
+  target.pipe(client)
+  const close = () => {
+    sockets.delete(client)
+    sockets.delete(target)
+    client.destroy()
+    target.destroy()
+  }
+  client.on("close", close)
+  target.on("close", close)
+})
+
+try { fs.unlinkSync(listenPath) } catch {}
+server.listen(listenPath)
+const watcher = fs.watch(path.dirname(dropPath), (_event, name) => {
+  if (name !== dropPath.split("/").at(-1) || !fs.existsSync(dropPath)) return
+  for (const socket of sockets) socket.destroy()
+  watcher.close()
+  server.close()
+})
+
+const stop = () => {
+  watcher.close()
+  for (const socket of sockets) socket.destroy()
+  server.close(() => process.exit(0))
+}
+process.on("SIGTERM", stop)
+process.on("SIGINT", stop)

cells/pty-attach-machine-stream/fixture/fabric

@@ -0,0 +1,24 @@
+#!/usr/bin/env bash
+set -euo pipefail
+
+test "${1:-}" = dial
+test "${2:-}" = eval-peer
+test "${3:-}" = pty-remote
+state="${PTY_EVAL_STATE:?PTY_EVAL_STATE must be set}"
+mkdir -p "$state"
+
+if mkdir "$state/first-dial" 2>/dev/null; then
+  printf '%s\n' "${PTY_EVAL_SOCKET_1:?PTY_EVAL_SOCKET_1 must be set}"
+  exit 0
+fi
+
+touch "$state/second-dial"
+for _ in $(seq 1 160); do
+  if test -f "$state/release-second"; then
+    printf '%s\n' "${PTY_EVAL_SOCKET_2:?PTY_EVAL_SOCKET_2 must be set}"
+    exit 0
+  fi
+  sleep 0.05
+done
+echo "timed out waiting to release second dial" >&2
+exit 1