@linzumi/cli 1.0.170 → 1.0.172

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@linzumi/cli",
-  "version": "1.0.170",
+  "version": "1.0.172",
   "description": "Linzumi CLI \u2014 point a Codex agent at the real code on your laptop, with your team watching and steering from shared threads.",
   "type": "module",
   "bin": {

package/scripts/codex-wire-canary.sh

@@ -0,0 +1,56 @@
+#!/usr/bin/env bash
+# Codex wire canary launcher (spec/codex-wire-capture-replay.spec.md, "Live
+# canary lane"): resolves the wire corpus + the pinned @openai/codex version
+# and runs the ReScript canary bin (HarnessCodexWireCanaryMain.res).
+#
+#   codex-wire-canary.sh corpus            the scheduled corpus check
+#                                          (fidelity + shape baseline +
+#                                          freshness vs the pinned codex)
+#   codex-wire-canary.sh fresh <capture>   drift-diff a freshly recorded
+#                                          canary capture BEFORE promotion
+#
+# The canary corpus is every wire capture the replay lane treats as real
+# wire truth: test/fixtures/agent-raw-replay/wire-corpus/*.ndjson plus the
+# real-probe session (test/fixtures/agent-raw-replay/real-probe/codex.ndjson).
+# The committed shape baseline lives beside the corpus
+# (wire-corpus/shape-baseline.json); the vitest corpus gate pins the same
+# set, so this script and the test cannot disagree silently.
+#
+# Build the ReScript package first (`npx rescript build
+# packages/linzumi-cli-rescript`); the CLI test pipeline already does.
+set -euo pipefail
+
+here="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+cli_root="$(cd "$here/.." && pwd)"
+bin_module="$cli_root/../linzumi-cli-rescript/src/harness/HarnessCodexWireCanaryMain.res.mjs"
+baseline="$cli_root/test/fixtures/agent-raw-replay/wire-corpus/shape-baseline.json"
+
+if [[ ! -f "$bin_module" ]]; then
+  echo "codex-wire-canary: compiled module missing at $bin_module - run: npx rescript build <repo>/packages/linzumi-cli-rescript" >&2
+  exit 66
+fi
+
+mode="${1:-}"
+case "$mode" in
+  corpus)
+    pinned_version="$(node -p "require('$cli_root/package.json').dependencies['@openai/codex'].replace(/^[^0-9]*/, '')")"
+    captures=("$cli_root"/test/fixtures/agent-raw-replay/wire-corpus/*.ndjson)
+    captures+=("$cli_root/test/fixtures/agent-raw-replay/real-probe/codex.ndjson")
+    exec node "$bin_module" corpus \
+      --baseline "$baseline" \
+      --pinned-codex-version "$pinned_version" \
+      "${captures[@]}"
+    ;;
+  fresh)
+    shift
+    if [[ $# -lt 1 ]]; then
+      echo "codex-wire-canary: fresh mode needs the freshly recorded capture path" >&2
+      exit 64
+    fi
+    exec node "$bin_module" fresh --baseline "$baseline" "$@"
+    ;;
+  *)
+    echo "codex-wire-canary: unknown mode '${mode}' (use corpus or fresh)" >&2
+    exit 64
+    ;;
+esac

package/scripts/replay-agent-backend.ts

@@ -67,6 +67,15 @@ type BackendReplayOptions = {
   // machinery is exercised against the REAL backend.
   readonly wire: boolean;
   readonly wireStrictness: string | undefined;
+  // Replay-to-a-point (spec §Replay-to-a-point): --wire-breakpoint specs
+  // for the wire stub (`ordinal:<n>`, `method:<rpcMethod>[:<n>]`,
+  // `first-delta`, `turn-started[:<n>]`, `turn-completed[:<n>]`), in hold
+  // order. While the stub holds, the whole stack (runner -> backend ->
+  // client projection) is frozen at the recorded prefix; an external
+  // driver (Playwright DOM assertions, the film rig) observes and releases
+  // the hold over the stub's HTTP control routes (GET /replay/breakpoint,
+  // POST /replay/resume against the printed control URL).
+  readonly wireBreakpoints: readonly string[];
 };
 
 type WireReplayServerHandle = {
@@ -93,6 +102,7 @@ async function startWireReplayServerForFixture(
       readonly capturePath?: string;
       readonly strictness?: string | undefined;
       readonly clock?: string | undefined;
+      readonly breakpoints?: readonly string[];
     }) => Promise<WireReplayServerHandle>;
     wireReplayClockForBackendSpeed: (
       speed: string | undefined
@@ -105,13 +115,27 @@ async function startWireReplayServerForFixture(
   // test/agentReplayFixture.test.ts.
   const wireClock = wireModule.wireReplayClockForBackendSpeed(options.speed);
 
-  return wireModule.startWireReplayServerJs({
+  const server = await wireModule.startWireReplayServerJs({
     capturePath: options.fixturePath,
     ...(options.wireStrictness === undefined
       ? {}
       : { strictness: options.wireStrictness }),
     ...(wireClock === undefined ? {} : { clock: wireClock }),
+    ...(options.wireBreakpoints.length === 0
+      ? {}
+      : { breakpoints: options.wireBreakpoints }),
   });
+  if (options.wireBreakpoints.length > 0) {
+    // The external driver's handle on the frozen stack: poll GET
+    // /replay/breakpoint until held, assert on the frozen DOM/projection,
+    // then POST /replay/resume.
+    console.error(
+      `[wire-breakpoint] control http://127.0.0.1:${server.port} breakpoints=${options.wireBreakpoints.join(
+        ','
+      )}`
+    );
+  }
+  return server;
 }
 
 type BackendReplayBrowserAuth = {
@@ -403,6 +427,7 @@ function parseArgs(args: readonly string[]): BackendReplayOptions {
     oracleFeatureId: argValue(args, '--oracle-feature'),
     wire: args.includes('--wire'),
     wireStrictness: argValue(args, '--wire-strictness'),
+    wireBreakpoints: argValues(args, '--wire-breakpoint'),
   };
 }
 
@@ -1888,6 +1913,24 @@ function evaluateRequiredFeatureGate(
   }
 }
 
+// Every value of a repeatable flag, in argv order (--wire-breakpoint may
+// name several ordered holds).
+function argValues(args: readonly string[], name: string): readonly string[] {
+  const values: string[] = [];
+  for (let index = 0; index < args.length; index += 1) {
+    if (args[index] !== name) {
+      continue;
+    }
+    const value = args[index + 1];
+    if (value === undefined) {
+      throw new Error(`${name} requires a value`);
+    }
+    values.push(value);
+    index += 1;
+  }
+  return values;
+}
+
 function argValue(args: readonly string[], name: string): string | undefined {
   const index = args.indexOf(name);
 
@@ -1906,7 +1949,7 @@ function argValue(args: readonly string[], name: string): string | undefined {
 
 function printHelp(): void {
   process.stdout.write(
-    `Usage: pnpm run replay:agent-backend -- --fixture <path> [options]\n\nOptions:\n  --dry-run                         Validate fixture/manifest/diagnostics without backend connection\n  --probe-backend                   Also join the real Phoenix local_runner channel, then exit\n  --provider auto|codex|claude-code Replay provider. Default: auto\n  --speed immediate|realtime|<n>x|<n>ms\n                                      Replay speed. Default: immediate\n  --backend-url <url>               Local Phoenix backend URL; required by --probe-backend/replay\n  --setup-fixture                   Create a throwaway fixture user/workspace/channel first; requires local fixture registration\n  --setup-label <label>             Prefix for throwaway fixture setup. Default: agent-backend-replay\n  --workspace <slug>                Target workspace slug for session-token exchange and future replay\n  --channel <slug>                  Target channel slug for token scoping and future replay\n  --thread-id <id>                  Target thread id for the future backend replay slice\n  --runner-id <id>                  Local runner id for backend probing. Default: generated\n  --cwd <path>                      Cwd advertised in backend probe/replay. Default: current directory\n  --work-description <text>         Work description sent through the normal start mutation\n  --token <token>                   Local-runner token; also reads LINZUMI_AGENT_BACKEND_REPLAY_TOKEN\n  --session-token <token>           Browser/session token to exchange for a local-runner token; also reads LINZUMI_AGENT_BACKEND_REPLAY_SESSION_TOKEN\n  --out-dir <path>                  Artifact directory for summary.json. Default: agent-backend-replay-output\n  --wire                            Serve the fixture from the stub codex app-server websocket (spec/codex-wire-capture-replay.spec.md) so the runner's real socket machinery is exercised\n  --wire-strictness shape|exact     Wire mode validation of runner-sent packets against the recording. Default: shape\n  --require-feature <featureId>     Fail before backend setup unless manifest readiness says the feature is ready\n  --oracle-feature <featureId>      Run feature-specific backend oracles without using readiness as a preflight gate\n  --no-summary                      Do not write summary.json; stdout JSON is still printed\n\nExamples:\n  pnpm --filter @linzumi/cli run replay:agent-backend -- --fixture test/fixtures/agent-raw-replay/real-probe/claude.ndjson --provider claude-code --dry-run\n  pnpm --filter @linzumi/cli run replay:agent-backend -- --fixture test/fixtures/agent-raw-replay/real-probe/claude.ndjson --provider claude-code --probe-backend --backend-url http://127.0.0.1:4140 --workspace default --session-token <browser-session-token>\n  pnpm --filter @linzumi/cli run replay:agent-backend -- --fixture test/fixtures/agent-raw-replay/real-probe/claude.ndjson --provider claude-code --backend-url http://127.0.0.1:4140 --setup-fixture\n`
+    `Usage: pnpm run replay:agent-backend -- --fixture <path> [options]\n\nOptions:\n  --dry-run                         Validate fixture/manifest/diagnostics without backend connection\n  --probe-backend                   Also join the real Phoenix local_runner channel, then exit\n  --provider auto|codex|claude-code Replay provider. Default: auto\n  --speed immediate|realtime|<n>x|<n>ms\n                                      Replay speed. Default: immediate\n  --backend-url <url>               Local Phoenix backend URL; required by --probe-backend/replay\n  --setup-fixture                   Create a throwaway fixture user/workspace/channel first; requires local fixture registration\n  --setup-label <label>             Prefix for throwaway fixture setup. Default: agent-backend-replay\n  --workspace <slug>                Target workspace slug for session-token exchange and future replay\n  --channel <slug>                  Target channel slug for token scoping and future replay\n  --thread-id <id>                  Target thread id for the future backend replay slice\n  --runner-id <id>                  Local runner id for backend probing. Default: generated\n  --cwd <path>                      Cwd advertised in backend probe/replay. Default: current directory\n  --work-description <text>         Work description sent through the normal start mutation\n  --token <token>                   Local-runner token; also reads LINZUMI_AGENT_BACKEND_REPLAY_TOKEN\n  --session-token <token>           Browser/session token to exchange for a local-runner token; also reads LINZUMI_AGENT_BACKEND_REPLAY_SESSION_TOKEN\n  --out-dir <path>                  Artifact directory for summary.json. Default: agent-backend-replay-output\n  --wire                            Serve the fixture from the stub codex app-server websocket (spec/codex-wire-capture-replay.spec.md) so the runner's real socket machinery is exercised\n  --wire-strictness shape|exact     Wire mode validation of runner-sent packets against the recording. Default: shape\n  --wire-breakpoint <spec>          Replay-to-a-point hold (repeatable, ordered): ordinal:<n> | method:<rpcMethod>[:<n>] | first-delta | turn-started[:<n>] | turn-completed[:<n>]. The stub freezes the stack at the hold; observe/release via GET /replay/breakpoint + POST /replay/resume on the printed control URL\n  --require-feature <featureId>     Fail before backend setup unless manifest readiness says the feature is ready\n  --oracle-feature <featureId>      Run feature-specific backend oracles without using readiness as a preflight gate\n  --no-summary                      Do not write summary.json; stdout JSON is still printed\n\nExamples:\n  pnpm --filter @linzumi/cli run replay:agent-backend -- --fixture test/fixtures/agent-raw-replay/real-probe/claude.ndjson --provider claude-code --dry-run\n  pnpm --filter @linzumi/cli run replay:agent-backend -- --fixture test/fixtures/agent-raw-replay/real-probe/claude.ndjson --provider claude-code --probe-backend --backend-url http://127.0.0.1:4140 --workspace default --session-token <browser-session-token>\n  pnpm --filter @linzumi/cli run replay:agent-backend -- --fixture test/fixtures/agent-raw-replay/real-probe/claude.ndjson --provider claude-code --backend-url http://127.0.0.1:4140 --setup-fixture\n`
   );
 }