@codyswann/lisa 4.1.6 → 4.3.1

package/plugins/lisa-wiki-agy/plugin.json

@@ -1,6 +1,6 @@
 {
   "name": "lisa-wiki",
-  "version": "4.1.6",
+  "version": "4.3.1",
   "description": "LLM Wiki — a distributable, git-native markdown knowledge base for Claude Code and Codex",
   "author": {
     "name": "Cody Swann"

package/plugins/lisa-wiki-copilot/.claude-plugin/plugin.json

@@ -1,6 +1,6 @@
 {
   "name": "lisa-wiki",
-  "version": "4.1.6",
+  "version": "4.3.1",
   "description": "LLM Wiki — a distributable, git-native markdown knowledge base for Claude Code and Codex",
   "author": {
     "name": "Cody Swann"

package/plugins/lisa-wiki-cursor/.claude-plugin/plugin.json

@@ -1,6 +1,6 @@
 {
   "name": "lisa-wiki",
-  "version": "4.1.6",
+  "version": "4.3.1",
   "description": "LLM Wiki — a distributable, git-native markdown knowledge base for Claude Code and Codex",
   "author": {
     "name": "Cody Swann"

package/scripts/build-plugins.sh

@@ -183,6 +183,34 @@ if [ -f "$ROOT_DIR/scripts/lib/invoked-as-script.mjs" ] &&
   done
 fi
 
+# The shared child-start deadline, into the same lanes, for the same reason.
+#
+# A guard that starts a child and never bounds it does not hang loudly — it
+# returns a VERDICT. `spawnSync` hands back empty streams and a null status when
+# it kills a child at its deadline, which is indistinguishable from a program
+# that ran and said no, so `status === 0 ? out : null` reads a busy machine as a
+# clean negative and the guard says "allow" (#2980).
+#
+# Materialized rather than hand-copied on exactly the `invoked-as-script`
+# reasoning above: downstream every lane lands flat in one `scripts/`, so a
+# single `scripts/lib/bounded-spawn.mjs` serves them all and `./lib/...`
+# resolves from each, while inside THIS repo the lanes are separate trees whose
+# unit tests import the lane copies directly. One canonical source, a
+# byte-equality test, and the copies cannot drift.
+#
+# Guarded on the GENERATOR as well as the source, again for the fixture reason:
+# this script runs against isolated fixtures that vendor `scripts/lib/` without
+# the repository's own `scripts/*.mjs`.
+if [ -f "$ROOT_DIR/scripts/lib/bounded-spawn.mjs" ] &&
+  [ -f "$ROOT_DIR/scripts/materialize-copy-overwrite.mjs" ]; then
+  for spawn_lane in all typescript expo; do
+    spawn_lib_dir="$ROOT_DIR/$spawn_lane/copy-overwrite/scripts/lib"
+    mkdir -p "$spawn_lib_dir"
+    materialize "$ROOT_DIR/scripts/lib/bounded-spawn.mjs" \
+      "$spawn_lib_dir/bounded-spawn.mjs"
+  done
+fi
+
 # Stack-specific plugins (NO base copy)
 STACKS=(typescript expo nestjs cdk harper-fabric phaser rails)
 for stack in "${STACKS[@]}"; do

package/scripts/lib/bounded-spawn.mjs

@@ -0,0 +1,185 @@
+/**
+ * bounded-spawn — the one implementation of "start a child, and refuse to
+ * return a verdict from one that never finished", shared by every guarded
+ * `.mjs` entry point Lisa ships that starts a synchronous child.
+ *
+ * @remarks
+ * ## The defect this exists to remove
+ *
+ * A test with a hung child fails a gate. **A guard with a hung child returns a
+ * verdict.** That is the whole reason this module is separate from the test
+ * tree's equivalent: out here the consequence of a child that never answered
+ * is a check that says "allow".
+ *
+ * The mechanism is one line, and it is measured rather than reasoned. These are
+ * the three outcomes of a synchronous child start, on this platform:
+ *
+ * | outcome | `.code` | `.status` | `.signal` | streams |
+ * | --- | --- | --- | --- | --- |
+ * | killed at its deadline | `"ETIMEDOUT"` | `null` | `"SIGKILL"` | **empty** |
+ * | ran and exited non-zero | *(absent)* | `3` | `null` | real |
+ * | binary not found | `"ENOENT"` | `null` | `null` | empty |
+ *
+ * Now read the shape that appears all over this repository:
+ *
+ * ```js
+ * return result.status === 0 ? result.stdout : null;
+ * ```
+ *
+ * `status === 0` is false for a killed child **and** for a program that ran and
+ * said no, so the two are indistinguishable and both become `null`. The caller
+ * treats `null` as "the answer is no" and proceeds. **A busy machine therefore
+ * produces the same value a clean answer does**, and nothing anywhere says the
+ * word "time".
+ *
+ * That is why the fail-open spelling was never a careless choice. It is
+ * indistinguishable from correct handling unless you already know the kill case
+ * exists — which is precisely the argument for putting the discrimination in
+ * one shared place instead of asking 80 call sites to remember it.
+ *
+ * ## Why this THROWS
+ *
+ * Returning a distinguishable value would leave every call site free to ignore
+ * it, and the call sites that most need to check are the ones that already
+ * decided a falsy result means "no". Throwing inverts the default: a call site
+ * that does nothing gets fail-closed behaviour, and only a call site that
+ * *deliberately* catches can proceed.
+ *
+ * Measured on the tree this was written for: of 80 synchronous child starts in
+ * shipped scripts, **51 propagate** — a throw escapes and fails the process, so
+ * those are fixed by this module alone — and **29 sit inside a `try` with a
+ * `catch`**, where a throw would be swallowed. Those 29 are not a conversion
+ * job; each needs a human answer to "what should happen when this child is
+ * killed?", and the answer is written at the call site with
+ * {@link isChildTimeout}.
+ *
+ * ## Why the discriminator is `error.code`, not a custom error class
+ *
+ * A class cannot cross the boundary. Plugin payloads have no `./lib/` to import
+ * from — the accommodation `threshold-ratchet.mjs` and `preflight-secrets.mjs`
+ * already make — so a shared class would be unavailable in exactly the tree
+ * with the most call sites. `ETIMEDOUT` is set by Node itself on all three
+ * forms (`spawnSync` puts it on `result.error`; `execFileSync` and `execSync`
+ * throw an error carrying it), so a payload that cannot import this module
+ * writes the identical check inline and is stating the same **platform fact**
+ * rather than duplicating a Lisa convention.
+ *
+ * `ENOENT` is deliberately NOT treated as a timeout. A missing binary is a real
+ * answer about the environment, it is reported the same way on every run, and
+ * conflating it with a kill would hide a broken install behind a retryable-
+ * looking error.
+ * @module scripts/lib/bounded-spawn
+ */
+
+import { execFileSync, spawnSync } from "node:child_process";
+
+/**
+ * Deadline a child gets when the caller states none, in milliseconds.
+ *
+ * Deliberately generous. This is not a performance budget — it is the point
+ * past which a child is presumed hung, and the cost of setting it too low is a
+ * guard that fails on a slow machine while the cost of setting it too high is
+ * only a slower failure. `git` resolved through `PATH` on macOS goes through
+ * Apple's `xcrun` shim, which has been measured at over 20 seconds under load
+ * against 11ms for a real binary (CodySwannGT/lisa#2887).
+ *
+ * **30s is not padding, and tightening it is not a safe optimisation.** A
+ * deadline below the shim's measured worst case would make this module's own
+ * timeout the dominant failure mode — the tool would start manufacturing the
+ * failures it exists to detect, and every one of them would look exactly like
+ * the defect it was built to catch. If a number here ever needs changing, the
+ * thing to change first is `git` being resolved through `PATH` at all.
+ */
+export const DEFAULT_CHILD_BUDGET_MS = 30_000;
+
+/**
+ * Whether this error is a child that was killed at its deadline.
+ *
+ * The one question a `catch` around a bounded child start has to ask. A `catch`
+ * that does not ask it treats "the box was busy" as "the command said no",
+ * which is the fail-open this module exists to remove — so a catch block
+ * enclosing a bounded start must either re-raise on this or say in a comment
+ * why continuing is correct.
+ *
+ * Accepts `unknown` because that is what a `catch` binding is, and reads the
+ * property defensively for the same reason: a thrown non-object is legal
+ * JavaScript and must answer `false` rather than crash the guard that was
+ * trying to be careful.
+ * @param {unknown} error A caught value, or `result.error` from `spawnSync`.
+ * @returns {boolean} True when the child was killed at its deadline.
+ */
+export function isChildTimeout(error) {
+  return (
+    typeof error === "object" &&
+    error !== null &&
+    /** @type {{ code?: unknown }} */ (error).code === "ETIMEDOUT"
+  );
+}
+
+/**
+ * Re-raise a caught value when it is a killed child, otherwise do nothing.
+ *
+ * The one-line form for the 29 call sites whose `catch` predates this module.
+ * Written as a helper rather than left to each site because the correct spelling
+ * is `throw error` — NOT wrapping, NOT `throw new Error(...)` — and a wrapped
+ * error loses the `code` the next frame up needs to make the same decision.
+ * @param {unknown} error The caught value.
+ * @returns {void}
+ * @throws {unknown} The original value, when it is a killed child.
+ */
+export function rethrowIfChildTimeout(error) {
+  if (isChildTimeout(error)) throw error;
+}
+
+/**
+ * Options a bounded start always applies, whatever the caller asked for.
+ *
+ * `killSignal` is `SIGKILL` rather than the default `SIGTERM` deliberately: the
+ * hang this module exists for is a process that is not servicing signals, and a
+ * `SIGTERM` a child ignores turns the deadline into a suggestion.
+ * @param {object} options The caller's options.
+ * @returns {object} The caller's options with the deadline enforced.
+ */
+function bounded(options) {
+  return {
+    ...options,
+    killSignal: "SIGKILL",
+    timeout: options.timeout ?? DEFAULT_CHILD_BUDGET_MS,
+  };
+}
+
+/**
+ * `spawnSync` with a deadline, refusing to return a killed child's result.
+ *
+ * `spawnSync` does not throw on a timeout — it returns a result whose streams
+ * are EMPTY and whose `status` is `null`, which is why the fail-open shape in
+ * the module docblock cannot see it. This throws instead, so a caller that does
+ * nothing inherits fail-closed behaviour.
+ * @param {string} command The executable.
+ * @param {readonly string[]} args Arguments.
+ * @param {object} [options] `spawnSync` options; `timeout` overrides the default.
+ * @returns {import("node:child_process").SpawnSyncReturns<string>} The result.
+ * @throws {Error} When the child was killed at its deadline.
+ */
+export function boundedSpawnSync(command, args = [], options = {}) {
+  const result = spawnSync(command, [...args], bounded(options));
+  if (isChildTimeout(result.error)) throw result.error;
+  return result;
+}
+
+/**
+ * `execFileSync` with a deadline.
+ *
+ * `execFileSync` already throws on a timeout, so this changes the FAILURE MODE
+ * rather than adding one: without a `timeout` there is no deadline to reach and
+ * the call blocks for as long as the child lives. The throw it produces carries
+ * `code: "ETIMEDOUT"`, which is what {@link isChildTimeout} reads.
+ * @param {string} command The executable.
+ * @param {readonly string[]} args Arguments.
+ * @param {object} [options] `execFileSync` options; `timeout` overrides the default.
+ * @returns {string|Buffer} Whatever the child wrote to stdout.
+ * @throws {Error} When the child was killed, or exited non-zero.
+ */
+export function boundedExecFileSync(command, args = [], options = {}) {
+  return execFileSync(command, [...args], bounded(options));
+}

package/typescript/copy-overwrite/scripts/lib/bounded-spawn.mjs

@@ -0,0 +1,188 @@
+// This file is managed by Lisa and IS replaced on each `lisa` run.
+// Do not edit directly — durable changes belong upstream in Lisa.
+
+/**
+ * bounded-spawn — the one implementation of "start a child, and refuse to
+ * return a verdict from one that never finished", shared by every guarded
+ * `.mjs` entry point Lisa ships that starts a synchronous child.
+ *
+ * @remarks
+ * ## The defect this exists to remove
+ *
+ * A test with a hung child fails a gate. **A guard with a hung child returns a
+ * verdict.** That is the whole reason this module is separate from the test
+ * tree's equivalent: out here the consequence of a child that never answered
+ * is a check that says "allow".
+ *
+ * The mechanism is one line, and it is measured rather than reasoned. These are
+ * the three outcomes of a synchronous child start, on this platform:
+ *
+ * | outcome | `.code` | `.status` | `.signal` | streams |
+ * | --- | --- | --- | --- | --- |
+ * | killed at its deadline | `"ETIMEDOUT"` | `null` | `"SIGKILL"` | **empty** |
+ * | ran and exited non-zero | *(absent)* | `3` | `null` | real |
+ * | binary not found | `"ENOENT"` | `null` | `null` | empty |
+ *
+ * Now read the shape that appears all over this repository:
+ *
+ * ```js
+ * return result.status === 0 ? result.stdout : null;
+ * ```
+ *
+ * `status === 0` is false for a killed child **and** for a program that ran and
+ * said no, so the two are indistinguishable and both become `null`. The caller
+ * treats `null` as "the answer is no" and proceeds. **A busy machine therefore
+ * produces the same value a clean answer does**, and nothing anywhere says the
+ * word "time".
+ *
+ * That is why the fail-open spelling was never a careless choice. It is
+ * indistinguishable from correct handling unless you already know the kill case
+ * exists — which is precisely the argument for putting the discrimination in
+ * one shared place instead of asking 80 call sites to remember it.
+ *
+ * ## Why this THROWS
+ *
+ * Returning a distinguishable value would leave every call site free to ignore
+ * it, and the call sites that most need to check are the ones that already
+ * decided a falsy result means "no". Throwing inverts the default: a call site
+ * that does nothing gets fail-closed behaviour, and only a call site that
+ * *deliberately* catches can proceed.
+ *
+ * Measured on the tree this was written for: of 80 synchronous child starts in
+ * shipped scripts, **51 propagate** — a throw escapes and fails the process, so
+ * those are fixed by this module alone — and **29 sit inside a `try` with a
+ * `catch`**, where a throw would be swallowed. Those 29 are not a conversion
+ * job; each needs a human answer to "what should happen when this child is
+ * killed?", and the answer is written at the call site with
+ * {@link isChildTimeout}.
+ *
+ * ## Why the discriminator is `error.code`, not a custom error class
+ *
+ * A class cannot cross the boundary. Plugin payloads have no `./lib/` to import
+ * from — the accommodation `threshold-ratchet.mjs` and `preflight-secrets.mjs`
+ * already make — so a shared class would be unavailable in exactly the tree
+ * with the most call sites. `ETIMEDOUT` is set by Node itself on all three
+ * forms (`spawnSync` puts it on `result.error`; `execFileSync` and `execSync`
+ * throw an error carrying it), so a payload that cannot import this module
+ * writes the identical check inline and is stating the same **platform fact**
+ * rather than duplicating a Lisa convention.
+ *
+ * `ENOENT` is deliberately NOT treated as a timeout. A missing binary is a real
+ * answer about the environment, it is reported the same way on every run, and
+ * conflating it with a kill would hide a broken install behind a retryable-
+ * looking error.
+ * @module scripts/lib/bounded-spawn
+ */
+
+import { execFileSync, spawnSync } from "node:child_process";
+
+/**
+ * Deadline a child gets when the caller states none, in milliseconds.
+ *
+ * Deliberately generous. This is not a performance budget — it is the point
+ * past which a child is presumed hung, and the cost of setting it too low is a
+ * guard that fails on a slow machine while the cost of setting it too high is
+ * only a slower failure. `git` resolved through `PATH` on macOS goes through
+ * Apple's `xcrun` shim, which has been measured at over 20 seconds under load
+ * against 11ms for a real binary (CodySwannGT/lisa#2887).
+ *
+ * **30s is not padding, and tightening it is not a safe optimisation.** A
+ * deadline below the shim's measured worst case would make this module's own
+ * timeout the dominant failure mode — the tool would start manufacturing the
+ * failures it exists to detect, and every one of them would look exactly like
+ * the defect it was built to catch. If a number here ever needs changing, the
+ * thing to change first is `git` being resolved through `PATH` at all.
+ */
+export const DEFAULT_CHILD_BUDGET_MS = 30_000;
+
+/**
+ * Whether this error is a child that was killed at its deadline.
+ *
+ * The one question a `catch` around a bounded child start has to ask. A `catch`
+ * that does not ask it treats "the box was busy" as "the command said no",
+ * which is the fail-open this module exists to remove — so a catch block
+ * enclosing a bounded start must either re-raise on this or say in a comment
+ * why continuing is correct.
+ *
+ * Accepts `unknown` because that is what a `catch` binding is, and reads the
+ * property defensively for the same reason: a thrown non-object is legal
+ * JavaScript and must answer `false` rather than crash the guard that was
+ * trying to be careful.
+ * @param {unknown} error A caught value, or `result.error` from `spawnSync`.
+ * @returns {boolean} True when the child was killed at its deadline.
+ */
+export function isChildTimeout(error) {
+  return (
+    typeof error === "object" &&
+    error !== null &&
+    /** @type {{ code?: unknown }} */ (error).code === "ETIMEDOUT"
+  );
+}
+
+/**
+ * Re-raise a caught value when it is a killed child, otherwise do nothing.
+ *
+ * The one-line form for the 29 call sites whose `catch` predates this module.
+ * Written as a helper rather than left to each site because the correct spelling
+ * is `throw error` — NOT wrapping, NOT `throw new Error(...)` — and a wrapped
+ * error loses the `code` the next frame up needs to make the same decision.
+ * @param {unknown} error The caught value.
+ * @returns {void}
+ * @throws {unknown} The original value, when it is a killed child.
+ */
+export function rethrowIfChildTimeout(error) {
+  if (isChildTimeout(error)) throw error;
+}
+
+/**
+ * Options a bounded start always applies, whatever the caller asked for.
+ *
+ * `killSignal` is `SIGKILL` rather than the default `SIGTERM` deliberately: the
+ * hang this module exists for is a process that is not servicing signals, and a
+ * `SIGTERM` a child ignores turns the deadline into a suggestion.
+ * @param {object} options The caller's options.
+ * @returns {object} The caller's options with the deadline enforced.
+ */
+function bounded(options) {
+  return {
+    ...options,
+    killSignal: "SIGKILL",
+    timeout: options.timeout ?? DEFAULT_CHILD_BUDGET_MS,
+  };
+}
+
+/**
+ * `spawnSync` with a deadline, refusing to return a killed child's result.
+ *
+ * `spawnSync` does not throw on a timeout — it returns a result whose streams
+ * are EMPTY and whose `status` is `null`, which is why the fail-open shape in
+ * the module docblock cannot see it. This throws instead, so a caller that does
+ * nothing inherits fail-closed behaviour.
+ * @param {string} command The executable.
+ * @param {readonly string[]} args Arguments.
+ * @param {object} [options] `spawnSync` options; `timeout` overrides the default.
+ * @returns {import("node:child_process").SpawnSyncReturns<string>} The result.
+ * @throws {Error} When the child was killed at its deadline.
+ */
+export function boundedSpawnSync(command, args = [], options = {}) {
+  const result = spawnSync(command, [...args], bounded(options));
+  if (isChildTimeout(result.error)) throw result.error;
+  return result;
+}
+
+/**
+ * `execFileSync` with a deadline.
+ *
+ * `execFileSync` already throws on a timeout, so this changes the FAILURE MODE
+ * rather than adding one: without a `timeout` there is no deadline to reach and
+ * the call blocks for as long as the child lives. The throw it produces carries
+ * `code: "ETIMEDOUT"`, which is what {@link isChildTimeout} reads.
+ * @param {string} command The executable.
+ * @param {readonly string[]} args Arguments.
+ * @param {object} [options] `execFileSync` options; `timeout` overrides the default.
+ * @returns {string|Buffer} Whatever the child wrote to stdout.
+ * @throws {Error} When the child was killed, or exited non-zero.
+ */
+export function boundedExecFileSync(command, args = [], options = {}) {
+  return execFileSync(command, [...args], bounded(options));
+}