@codyswann/lisa 4.29.12 → 4.31.0

package/typescript/copy-overwrite/scripts/check-skipped-required-checks.mjs

@@ -12,7 +12,7 @@
  * yours to edit).
  *
  * Usage:
- *   node scripts/check-skipped-required-checks.mjs [rootDir] [--remote] [--json]
+ *   node scripts/check-skipped-required-checks.mjs [rootDir] [--json]
  *   node scripts/check-skipped-required-checks.mjs --vacuity [--fail-on-vacuous]
  *   node scripts/check-skipped-required-checks.mjs --pr=1234 [--repo=OWNER/NAME]
  *
@@ -31,7 +31,7 @@
  *
  * Two of the three live here:
  *
- *  - **Skipped** (`--remote` / offline arm, below): GitHub counts a `skipped`
+ *  - **Skipped** (the offline arm, below): GitHub counts a `skipped`
  *    required check as SATISFIED, so a `skip_jobs` token makes the gate
  *    decorative. Static, offline, BLOCKING.
  *  - **Vacuous** (`--pr` arm): the check really ran and really reported
@@ -40,11 +40,15 @@
  *
  * ## Where this runs
  *
- * Lisa's `quality.yml` runs the OFFLINE arm on every pull request, and the
- * shipped `required-checks-drift.yml` runs `--remote` on a weekly schedule.
- * That split is deliberate and is argued under "Why this is a DECLARATION
- * guard" below: the enforced PR path may not depend on network or `gh` auth,
- * and the snapshot it enforces may not be allowed to rot unwatched.
+ * Lisa's `quality.yml` runs this on every pull request. EVERY arm is offline:
+ * nothing here reads a live ruleset, needs a token, or touches the network for
+ * the required-context rules. That is a deliberate reduction (#3599) — the
+ * scheduled arm that used to re-read the ruleset was removed along with the
+ * `administration:read` credential it required, which sat as a standing repo
+ * secret in every consumer to detect a rare event. If the committed snapshot
+ * and the live ruleset drift apart, that is now discovered by consequence
+ * rather than by check, and that is the accepted trade. Do not reintroduce a
+ * cheaper detector for it.
  *
  * ## `enforcement`
  *
@@ -73,13 +77,26 @@
  * assert.
  *
  * So `required_contexts` is treated as a cache of a live fetch, not as
- * testimony. It is trusted only while `ruleset.baseline_fetched_at` carries a
- * parseable timestamp no older than SNAPSHOT_MAX_AGE_DAYS. Untranscribed or
- * expired, the guard REFUSES to answer: every rule that depends on
- * `required_contexts` is skipped, no ✅ is printed, and the report says NOT
- * CHECKED and why. The rules that do NOT read `required_contexts` — an
- * undeclared token, a token written with whitespace — still run, because they
- * are true regardless of what the ruleset requires.
+ * testimony. Trust is a function of PRESENCE: it is believed while
+ * `ruleset.baseline_fetched_at` carries a parseable timestamp, and refused when
+ * that stamp is empty or unreadable. Untranscribed, the guard REFUSES to
+ * answer: every rule that depends on `required_contexts` is skipped, no ✅ is
+ * printed, and the report says NOT CHECKED and why. The rules that do NOT read
+ * `required_contexts` — an undeclared token, a token written with whitespace —
+ * still run, because they are true regardless of what the ruleset requires.
+ *
+ * NOTHING EXPIRES, and that is a decided constraint (#3599), not an oversight.
+ * A ninety-day ceiling used to sit here, and it was coherent only while a
+ * scheduled arm existed that could re-read the ruleset and re-stamp the file.
+ * With that arm removed, a deadline is an obligation with no way to discharge
+ * it: somebody would have to re-transcribe every consumer's snapshot every
+ * quarter forever, or the guard silently goes NOT CHECKED. A guard that
+ * reliably expires into "not checked" is worse than one answering from a stamp
+ * somebody wrote once, because the first looks healthy right up until it has
+ * been quietly inert for a quarter. A softer form — believe it, but warn past N
+ * days — is the same substitution and is refused for the same reason: it
+ * recreates the recurring manual obligation and adds a signal nobody can act
+ * on. Non-empty believed. Empty refused. Nothing expiring.
  *
  * Refusal is exit 1, so an explicit `npm run check:skipped-required-checks`
  * cannot be mistaken for a pass. Under `"enforcement": "warn"` — which is what
@@ -114,16 +131,18 @@
  * authoritative alone; the coherence rules below are what stop either rotting
  * into decoration.
  *
- * That mutual policing has one blind spot, and `--remote` exists for it: two
- * snapshots in one repo can only catch each other drifting from the CODE.
+ * That mutual policing has one blind spot, and it is now UNCOVERED ON PURPOSE:
+ * two snapshots in one repo can only catch each other drifting from the CODE.
  * Neither can see the ruleset itself change in the admin console — which is
- * exactly how acmeorgd's list silently went from ten contexts to eleven, with every
- * test still green, because the "independent" transcription was made from the
- * same reading at the same moment.
+ * exactly how one repository's list silently went from ten contexts to eleven,
+ * with every test still green, because the "independent" transcription was made
+ * from the same reading at the same moment.
  *
- * `--remote` is opt-in so the ENFORCED path stays offline. A guard that needed
- * network and `gh` auth on every run would flake, and a flaky guard gets
- * skipped — which reintroduces exactly the false-green class this file refuses.
+ * A scheduled arm used to close that gap by reading the live ruleset, and it
+ * was removed (#3599) because closing it cost a permanent `administration:read`
+ * token in every consumer. The blind spot is the price of not holding that
+ * credential. It is stated here rather than hidden, and it is not an invitation
+ * to build a lighter-weight version of the same thing.
  *
  * ## `--pr` — the VACUOUS arm, and why it only ever reports
  *
@@ -148,8 +167,8 @@
  * ## `--vacuity` — why a flag nobody passes is the same defect
  *
  * MEASURED (CodySwannGT/lisa#2928): for the whole life of this arm, NOTHING
- * INVOKED IT. `quality.yml` ran the offline arm with no `--pr`, the shipped
- * `required-checks-drift.yml` ran `--remote`, and the package script named
+ * INVOKED IT. `quality.yml` ran the offline arm with no `--pr`, the then-shipped
+ * scheduled drift workflow ran its own arm, and the package script named
  * `check:vacuous-required-checks` was a bare invocation — so the command named
  * for the vacuous check reported on SKIPS and said nothing about vacuity unless
  * a caller happened to remember `-- --pr=1234`. A rule that runs only when
@@ -270,18 +289,6 @@ import { invokedAsScript } from "./lib/invoked-as-script.mjs";
 /** Repo-relative path of the per-repo declaration file. */
 export const DECLARATION_PATH = ".github/required-checks.json";
 
-/**
- * How long a transcribed `required_contexts` snapshot stays trustworthy.
- *
- * A SOURCE CONSTANT, not an input: a ceiling somebody can widen from a config
- * file fails open on exactly the runs nobody tests. Rulesets are edited in an
- * admin console with no signal in the repository, so an old transcription is
- * not evidence — it is a memory of evidence. Ninety days is long enough that
- * the weekly `--remote` run refreshes it many times over, and short enough that
- * an abandoned repository stops being told its skips are fine.
- */
-export const SNAPSHOT_MAX_AGE_DAYS = 90;
-
 /**
  * Matches a `skip_jobs:` key and captures everything after the colon.
  *
@@ -331,7 +338,6 @@ export const VIOLATIONS = Object.freeze({
   stale: "declaration_overstates_requirement",
   orphaned: "orphaned_exemption",
   badExemption: "exemption_without_valid_ticket",
-  remoteDrift: "ruleset_snapshot_drift",
   whitespace: "whitespace_in_skip_token",
   vacuous: "vacuous_required_check",
   unproven: "unproven_required_check",
@@ -399,15 +405,8 @@ const ENFORCEMENT_MODES = Object.freeze(["error", "warn"]);
  * both that a context is ruleset-required AND that a token it actually skips
  * silences it; that is a reviewed state, and shipping past it is the exact
  * defect this file exists to refuse.
- *
- * `ruleset_snapshot_drift` blocks for the same reason: it only fires under an
- * explicit `--remote` on a repository that filled in its ruleset ids, and it is
- * measured against live truth rather than inferred from a seed.
  */
-const ALWAYS_BLOCKING = Object.freeze([
-  VIOLATIONS.suppressesRequired,
-  VIOLATIONS.remoteDrift,
-]);
+const ALWAYS_BLOCKING = Object.freeze([VIOLATIONS.suppressesRequired]);
 
 /**
  * Violation kinds that NEVER fail the build, in any enforcement mode.
@@ -488,6 +487,27 @@ export const SETTLE_INTERVAL_SECONDS = 15;
 /** Enables the blocking review-evidence policy. */
 const REQUIRE_REVIEW_EVIDENCE_FLAG = "--require-review-evidence";
 
+/**
+ * The flag that used to select the live-ruleset drift mode, retired with it.
+ *
+ * Kept as an explicit rejection rather than left to the argument parser. Every
+ * unrecognised `--*` argument is discarded before the positional read, so a
+ * caller that still selects the retired mode would otherwise get the ordinary
+ * offline run and an exit code of zero — a caller asking for a live comparison
+ * and being told "pass" by something that never looked. Retiring a mode
+ * silently is how a removed control keeps reporting success.
+ */
+const RETIRED_REMOTE_FLAG = "--remote";
+
+/**
+ * What a caller that still selects the retired mode is told.
+ *
+ * Names the flag only to say it is gone: the caller has to be able to find the
+ * thing it must delete. It is deliberately not a suggestion to run anything —
+ * there is no live-ruleset mode left to point at.
+ */
+const RETIRED_REMOTE_MESSAGE = `check-skipped-required-checks: \`${RETIRED_REMOTE_FLAG}\` was retired together with the live-ruleset drift arm and the standing \`administration:read\` token it required (CodySwannGT/lisa#3599). There is no live comparison left to run, so this invocation would otherwise have reported success without performing the check that was asked for. Drop the flag, and drop the \`check:skipped-required-checks:remote\` npm script that passes it.`;
+
 /** Matches the `GITHUB_REF` a `pull_request` event run carries. */
 const REF_PULL = /^refs\/pull\/(\d+)\/(?:merge|head)$/u;
 
@@ -829,11 +849,19 @@ export function collectSkipJobTokens(rootDir, workflows) {
  * the state Lisa's seeds ship in, and the seeds are a GUESS — a guess that was
  * measured wrong (#2476). No stamp, no answer.
  *
+ * Trust is PRESENCE, not freshness (#3599). There is no age ceiling here and
+ * there must not be one: the scheduled arm that could re-read a live ruleset
+ * and re-stamp this file is gone, so an expiry would be a deadline nobody can
+ * meet, and the guard would go quietly inert rather than loudly wrong. A
+ * warning past N days is the same mistake wearing a softer hat — it recreates
+ * the obligation and emits a signal no operator can discharge. The protection
+ * that matters is the REFUSAL on an untranscribed snapshot, and that is what
+ * the two arms below preserve.
+ *
  * @param {object} declaration - The per-repo declaration
- * @param {number} [now] - Current epoch milliseconds, injectable for tests
  * @returns {{trusted: boolean, reason: string}} Whether to believe the snapshot, and why not
  */
-export function snapshotTrust(declaration, now = Date.now()) {
+export function snapshotTrust(declaration) {
   const stamp = declaration.ruleset?.baseline_fetched_at;
   if (typeof stamp !== "string" || stamp.trim() === "") {
     return {
@@ -841,20 +869,12 @@ export function snapshotTrust(declaration, now = Date.now()) {
       reason: `\`ruleset.baseline_fetched_at\` is empty, so \`required_contexts\` has never been transcribed from a live ruleset. Lisa's seed ships a GUESS, and the guess was measured WRONG once in this fleet (#2476): in that repository it named a context no ruleset required and omitted six that were. It names no context here on purpose — what is required is a per-repository fact. Transcribe the real list, stamp the date, and this guard starts answering.`,
     };
   }
-  const fetchedAt = Date.parse(stamp);
-  if (Number.isNaN(fetchedAt)) {
+  if (Number.isNaN(Date.parse(stamp))) {
     return {
       trusted: false,
       reason: `\`ruleset.baseline_fetched_at\` is ${JSON.stringify(stamp)}, which is not a date this can read. Use an ISO-8601 timestamp, e.g. "2026-08-13".`,
     };
   }
-  const ageDays = (now - fetchedAt) / 86_400_000;
-  if (ageDays > SNAPSHOT_MAX_AGE_DAYS) {
-    return {
-      trusted: false,
-      reason: `\`required_contexts\` was last transcribed ${Math.floor(ageDays)} days ago, past the ${SNAPSHOT_MAX_AGE_DAYS}-day ceiling. Rulesets are edited in an admin console with no signal in this repository, so a stale transcription is a memory of evidence rather than evidence.`,
-    };
-  }
   return { trusted: true, reason: "" };
 }
 
@@ -1897,67 +1917,6 @@ export function vacuityRefusal(input) {
   return null;
 }
 
-/**
- * Fetches the live required contexts for every declared ruleset.
- *
- * @param {object} ruleset - `{ repo, ids }` from the declaration
- * @returns {string[]} Live contexts across all declared rulesets
- * @throws {Error} When `gh` is unavailable or the API cannot be read
- */
-export function fetchLiveRequiredContexts(ruleset) {
-  if (
-    !ruleset?.repo ||
-    !Array.isArray(ruleset.ids) ||
-    ruleset.ids.length === 0
-  ) {
-    throw new Error(
-      `check-skipped-required-checks: --remote needs \`ruleset.repo\` and \`ruleset.ids\` in ${DECLARATION_PATH}.`
-    );
-  }
-  const contexts = [];
-  for (const id of ruleset.ids) {
-    const raw = boundedExecFileSync(
-      "gh",
-      [
-        "api",
-        `repos/${ruleset.repo}/rulesets/${id}`,
-        "--jq",
-        '.rules[] | select(.type=="required_status_checks") | .parameters.required_status_checks[].context',
-      ],
-      { encoding: "utf8" }
-    );
-    for (const line of raw.split("\n").map(value => value.trim())) {
-      if (line !== "" && !contexts.includes(line)) contexts.push(line);
-    }
-  }
-  return contexts;
-}
-
-/**
- * Diffs the committed snapshot against the live ruleset, in BOTH directions.
- *
- * Both directions matter. A context added in the admin console makes the
- * snapshot UNDER-detect (acmeorgd's ten-to-eleven drift, unnoticed for a day with
- * every test green). A context removed there makes it OVER-detect, and the
- * obvious fix for a false alarm is to weaken the guard.
- *
- * @param {ReadonlyArray<string>} snapshot - Committed contexts
- * @param {ReadonlyArray<string>} live - Contexts from the API
- * @returns {object[]} Drift violations
- */
-export function compareRulesetBaseline(snapshot, live) {
-  const added = live.filter(context => !snapshot.includes(context));
-  const removed = snapshot.filter(context => !live.includes(context));
-  if (added.length === 0 && removed.length === 0) return [];
-  return [
-    {
-      kind: VIOLATIONS.remoteDrift,
-      token: null,
-      message: `\`required_contexts\` has drifted from the live ruleset.${added.length ? `\n  Live but not committed (the snapshot UNDER-detects): ${added.map(name => `"${name}"`).join(", ")}` : ""}${removed.length ? `\n  Committed but not live (the snapshot OVER-detects): ${removed.map(name => `"${name}"`).join(", ")}` : ""}\n  Update the snapshot and re-read what it now implies about the skip declarations.`,
-    },
-  ];
-}
-
 /**
  * Reads a `--name=<seconds>` flag, falling back when absent or unreadable.
  *
@@ -2071,17 +2030,18 @@ export function inspectVacuity(argv, declaration, options = {}) {
 /**
  * Runs the guard.
  *
- * `--remote` reads the ruleset live, so it does not need the cache to be
- * trustworthy — it is the thing that MAKES it trustworthy. Under `--remote` the
- * required-context rules therefore run regardless of the stamp, and any
- * disagreement surfaces as `ruleset_snapshot_drift` rather than as a verdict
- * about skips.
+ * Every arm is OFFLINE. There is no network read and no token anywhere in this
+ * path: the required-context rules answer from the committed snapshot when
+ * {@link snapshotTrust} believes it, and refuse when it does not.
  *
  * @param {ReadonlyArray<string>} argv - CLI arguments
  * @param {object} [options] - Injection seams forwarded to {@link inspectVacuity}
  * @returns {{violations: object[], checked: number, tokens: string[], enforcement: string, trust: {trusted: boolean, reason: string}, recipe: string, pr: string|undefined, evidenceChecked: number, vacuity: object|undefined}} The result
  */
 export function runGuard(argv, options = {}) {
+  if (argv.includes(RETIRED_REMOTE_FLAG)) {
+    throw new Error(RETIRED_REMOTE_MESSAGE);
+  }
   if (
     argv.includes(REQUIRE_REVIEW_EVIDENCE_FLAG) &&
     !argv.includes("--vacuity") &&
@@ -2095,26 +2055,11 @@ export function runGuard(argv, options = {}) {
   const rootDir = positional[0] ?? process.cwd();
   const declaration = loadDeclaration(rootDir);
   const collected = collectSkipJobTokens(rootDir, declaration.workflows);
-  const remote = argv.includes("--remote");
-  const live = remote
-    ? fetchLiveRequiredContexts(declaration.ruleset)
-    : undefined;
-  const trust = remote
-    ? { trusted: true, reason: "" }
-    : snapshotTrust(declaration);
-  const result = evaluateSkippedRequiredChecks(
-    live === undefined
-      ? declaration
-      : { ...declaration, required_contexts: live },
-    collected.tokens,
-    { trustRequiredContexts: trust.trusted }
-  );
+  const trust = snapshotTrust(declaration);
+  const result = evaluateSkippedRequiredChecks(declaration, collected.tokens, {
+    trustRequiredContexts: trust.trusted,
+  });
   const violations = [...collected.violations, ...result.violations];
-  if (live !== undefined) {
-    violations.push(
-      ...compareRulesetBaseline(declaration.required_contexts, live)
-    );
-  }
 
   // The vacuity arm is layered ON TOP of the offline run rather than replacing
   // it: it is a third variant of one family, so it belongs in one report. Its
@@ -2237,8 +2182,6 @@ function main(argv) {
       "```",
       result.recipe,
       "```",
-      "",
-      `Meanwhile \`--remote\` answers WITHOUT the cache, because it reads the ruleset live: \`npm run check:skipped-required-checks:remote\`.`,
       ""
     );
     process.stderr.write(

package/typescript/create-only/.github/required-checks.json

@@ -7,7 +7,7 @@
     "To arm it: gh api repos/OWNER/NAME/rulesets --jq '.[] | \"\\(.id) \\(.name)\"' to find the ids, then gh api repos/OWNER/NAME/rulesets/ID --jq '.rules[] | select(.type==\"required_status_checks\") | .parameters.required_status_checks[].context' for the contexts. Paste them into `required_contexts` BYTE FOR BYTE (emoji and the ' / ' separator included), fill in `ruleset.repo` / `ruleset.ids`, and set `baseline_fetched_at` to today.",
     "The guard compares with exact string equality on purpose: a repo routinely carries confusable pairs (an external app's required `GitGuardian Security Checks` beside a skippable, not-required `🔐 Credential Leakage` proving the same property; `🧹 Lint` beside `🐢 Slow Lint Rules`, whose skip tokens `lint` and `lint_slow` are a strict prefix pair), and a fuzzy match would raise a false alarm whose obvious fix is to delete the guard.",
     "The worst confusable pair Lisa ever shipped was a NOT-required `🧪 Run Tests` beside the required `🧪 Run Unit Tests` — it merged red on two PRs because 'Run Tests failed' reads like the required test gate failing. It was DELETED in #2485, not renamed, because it only re-ran what the two coverage-carrying required contexts already run. If you find a near-duplicate advisory check in your own roster, prefer deleting it over trusting a reader to tell three similar names apart.",
-    "A transcription expires after 90 days, because a ruleset can be edited with no signal in this repository. The shipped `.github/workflows/required-checks-drift.yml` runs `--remote` weekly to catch that; `--remote` reads the ruleset live and so answers even when the cache is untrusted.",
+    "A transcription NEVER EXPIRES. Trust is presence-based: a non-empty, parseable `baseline_fetched_at` means somebody read the list off a live ruleset, and it is believed; an empty or unreadable one means it was never transcribed, and the guard refuses to answer. A ninety-day ceiling used to sit here alongside a scheduled workflow that re-read the live ruleset and re-stamped this file. Both were removed (#3599) because that workflow needed a standing `administration:read` token in every repository. A deadline with no refresh mechanism is an obligation nobody can discharge, and a guard that reliably expires into NOT CHECKED looks healthy right up until it has been quietly inert for a quarter. The consequence is stated plainly: if this list and your live ruleset drift apart, you find out by consequence rather than by check. Re-transcribe when you change your ruleset.",
     "Lisa's quality.yml runs the offline arm on every pull request. This seed ships `\"enforcement\": \"warn\"`, which downgrades findings AND the NOT-CHECKED refusal to reports so a fresh install does not go red on arrival. Delete the key once you have transcribed the list — then it blocks.",
     "`skip_jobs` is matched as an exact comma-delimited token and GitHub Actions expression syntax has no string-replace, so write the list with no spaces: `skip_jobs: 'a,b'`, never `skip_jobs: 'a, b'`. A spaced token matches nothing and the job runs — reported as `whitespace_in_skip_token`.",
     "`_example_required_contexts` below is a STARTING POINT FOR TYPING, never read by the guard. Verify every line against your own ruleset before promoting any of it.",

package/typescript/create-only/.github/workflows/review-evidence.yml

@@ -11,7 +11,7 @@
 #
 # WHY THIS WORKFLOW EXISTS. The rule that detects it shipped long before
 # anything invoked it: `quality.yml` runs the guard's OFFLINE arm with no `--pr`,
-# `required-checks-drift.yml` runs `--remote` on a schedule, and the package
+# a scheduled drift workflow ran its own arm, and the package
 # script named for the vacuous check was a bare invocation. So the detection was
 # `declared-but-uncallable` — the very family the guard exists to describe,
 # one level up inside its own shipping. Measured on the repository that owns it:

package/typescript/deletions.json

@@ -26,6 +26,9 @@
     ".github/workflows/create-jira-issue-on-failure.yml",
     ".github/workflows/create-sentry-issue-on-failure.yml",
     ".github/workflows/reusable-auto-update-pr-branches.yml",
+    ".github/workflows/reusable-auto-update-pr-branches-dispatch.yml",
+    ".github/workflows/auto-update-pr-branches.yml",
+    ".github/workflows/auto-update-pr-branches-dispatch.yml",
     ".github/workflows/reusable-claude.yml",
     ".github/workflows/reusable-claude-ci-auto-fix.yml",
     ".github/workflows/reusable-claude-code-review-response.yml",
@@ -46,7 +49,8 @@
     ".github/workflows/reusable-claude-nightly-code-complexity-rails.yml",
     ".github/workflows/reusable-claude-nightly-jira-triage.yml",
     ".github/workflows/reusable-claude-nightly-test-coverage-rails.yml",
-    ".github/workflows/reusable-claude-nightly-test-improvement-rails.yml"
+    ".github/workflows/reusable-claude-nightly-test-improvement-rails.yml",
+    ".github/workflows/required-checks-drift.yml"
   ],
   "keep": []
 }

package/typescript/package-lisa/package.lisa.json

@@ -22,7 +22,6 @@
       "prepare": "node -e \"if (process.env.INIT_CWD?.includes('.serverless')) process.exit(0); process.exit(1);\" || husky install || true",
       "nightly:health": "node scripts/check-nightly-e2e-health.mjs",
       "check:skipped-required-checks": "node scripts/check-skipped-required-checks.mjs",
-      "check:skipped-required-checks:remote": "node scripts/check-skipped-required-checks.mjs --remote",
       "check:vacuous-required-checks": "node scripts/check-skipped-required-checks.mjs --vacuity",
       "test:node": "LISA_TEST_SCRATCH_PREFIXES='[\"node-\"]' lisa-test-run --profile typescript --adapter direct -- node scripts/lisa-test-node.mjs",
       "check:work-item": "node scripts/lisa-work-item.mjs validate-pr",
@@ -110,7 +109,8 @@
     ],
     "scripts": [
       "knip",
-      "knip:lisa"
+      "knip:lisa",
+      "check:skipped-required-checks:remote"
     ]
   },
   "adopt": {

package/ui/index.html

@@ -8755,24 +8755,6 @@
                 { t: "mono", v: "pull_request" },
                 { t: "text", v: "Deploy & environments" },
               ],
-              [
-                { t: "mono", v: "auto-update-pr-branches.yml" },
-                {
-                  t: "text",
-                  v: "Upkeep — keeps open PRs rebased on their base",
-                },
-                { t: "mono", v: "push · pull_request" },
-                { t: "text", v: "—" },
-              ],
-              [
-                { t: "mono", v: "auto-update-pr-branches-dispatch.yml" },
-                {
-                  t: "text",
-                  v: "Upkeep — on-demand handler for the same rebase, via repository dispatch",
-                },
-                { t: "mono", v: "repository_dispatch" },
-                { t: "text", v: "—" },
-              ],
               [
                 { t: "mono", v: "zap-baseline.yml" },
                 {

package/typescript/create-only/.github/workflows/auto-update-pr-branches-dispatch.yml

@@ -1,27 +0,0 @@
-# Seeded by Lisa on first setup — this file is YOURS.
-# Lisa will not overwrite it. (copy-overwrite assets ARE replaced each run.)
-# You can customize this file — Lisa will not overwrite it.
-
-name: Auto-update PR branches (Dispatch Handler)
-
-on:
-  repository_dispatch:
-    types:
-      - auto-update-pr-branches
-      - auto-update-pr-branch
-
-permissions:
-  contents: write
-  pull-requests: write
-  id-token: write
-
-jobs:
-  dispatch:
-    uses: CodySwannGT/lisa/.github/workflows/reusable-auto-update-pr-branches-dispatch.yml@main
-    with:
-      action: ${{ github.event.action }}
-      ref_name: ${{ github.event.client_payload.ref_name || '' }}
-      repo_full_name: ${{ github.event.client_payload.repo_full_name }}
-      pr_number: ${{ github.event.client_payload.pr_number || 0 }}
-    secrets:
-      CLAUDE_CODE_OAUTH_TOKEN: ${{ secrets.CLAUDE_CODE_OAUTH_TOKEN }}

package/typescript/create-only/.github/workflows/auto-update-pr-branches.yml

@@ -1,31 +0,0 @@
-# Seeded by Lisa on first setup — this file is YOURS.
-# Lisa will not overwrite it. (copy-overwrite assets ARE replaced each run.)
-# You can customize this file — Lisa will not overwrite it.
-
-name: Auto-update PR branches
-
-on:
-  push:
-    branches:
-      - main
-      - staging
-      - dev
-  pull_request:
-    types: [opened, reopened, ready_for_review]
-
-permissions:
-  contents: write
-  pull-requests: write
-  id-token: write
-
-jobs:
-  autoupdate:
-    uses: CodySwannGT/lisa/.github/workflows/reusable-auto-update-pr-branches.yml@main
-    with:
-      event_name: ${{ github.event_name }}
-      ref_name: ${{ github.ref_name }}
-      pr_base_ref: ${{ github.event.pull_request.base.ref || '' }}
-      pr_number: ${{ github.event.pull_request.number || 0 }}
-      repo_full_name: ${{ github.repository }}
-    secrets:
-      CLAUDE_CODE_OAUTH_TOKEN: ${{ secrets.CLAUDE_CODE_OAUTH_TOKEN }}

package/typescript/create-only/.github/workflows/required-checks-drift.yml

@@ -1,91 +0,0 @@
-# Seeded by Lisa on first setup — this file is YOURS.
-# Lisa will not overwrite it. (copy-overwrite assets ARE replaced each run.)
-# -----------------------------------------------------------------------------
-# Required-checks snapshot drift — the REMOTE arm (#2426)
-# -----------------------------------------------------------------------------
-# `.github/required-checks.json` holds two reviewed snapshots that police each
-# other: the contexts your ruleset requires, and what each `skip_jobs` token
-# silences. Two snapshots inside one repository can only catch each other
-# drifting from the CODE. Neither can see the RULESET change, because rulesets
-# are edited in an admin console — which is exactly how one repo's list went
-# from ten required contexts to eleven with every test still green for a day.
-#
-# So this runs the `--remote` arm on a schedule, and the pull-request path stays
-# offline. That split is the whole design: an enforced gate that needed network
-# and `gh` auth on every run would flake, and a flaky gate gets skipped — which
-# reintroduces the false-green class the guard exists to refuse. This workflow
-# never runs on `pull_request`, so it cannot wedge a merge.
-#
-# `--remote` is also the arm that STRUCTURALLY CANNOT LIE (#2476). It evaluates
-# the skip rules against the contexts it just read from the live ruleset, not
-# against the committed cache, so it answers correctly even in a repository
-# whose `required_contexts` has never been transcribed — which is exactly the
-# state the offline arm refuses to answer in. Arming this workflow is therefore
-# the fastest way to get a real answer.
-#
-# Reading a repository ruleset needs a token with administration:read. The
-# default GITHUB_TOKEN generally does not have it, so provide a PAT as the
-# repository secret RULESET_READ_TOKEN. Without it this job goes red rather
-# than reporting a clean bill of health it never checked.
-#
-# Until `ruleset.repo` / `ruleset.ids` are filled in, this reports UNCONFIGURED
-# and passes. Find the ids with:
-#   gh api repos/OWNER/NAME/rulesets --jq '.[] | "\(.id) \(.name)"'
-
-name: 🔒 Required-Checks Drift
-
-on:
-  schedule:
-    # Weekly. A ruleset can be edited in the admin console on a day nobody
-    # touched the repository, so change-triggered checking alone would miss it.
-    - cron: '0 8 * * 1'
-  workflow_dispatch:
-
-permissions:
-  contents: read
-
-jobs:
-  drift:
-    name: 🔒 Ruleset snapshot vs live ruleset
-    runs-on: ubuntu-latest
-    timeout-minutes: 10
-
-    steps:
-      - name: 📥 Checkout repository
-        uses: actions/checkout@v6
-        with:
-          persist-credentials: false
-
-      - name: 🔧 Setup Node.js
-        uses: actions/setup-node@v6
-        with:
-          package-manager-cache: false
-          node-version: '22.21.1'
-
-      - name: 🔒 Compare the committed snapshot against the live ruleset
-        env:
-          GH_TOKEN: ${{ secrets.RULESET_READ_TOKEN || github.token }}
-        run: |
-          set -euo pipefail
-          # An absent input is a FAILURE, never a skip — the same flip #2933
-          # made to the `🔒 Skipped Required Checks` job in quality.yml, for the
-          # same reason: this workflow is installed by the same `lisa apply`
-          # that installs the prover and the seed, so neither can be absent
-          # without the install being broken, and a weekly run that reports
-          # success having compared nothing is the drift it exists to catch.
-          if [ ! -f scripts/check-skipped-required-checks.mjs ]; then
-            echo "::error title=Skipped-required-check prover missing::scripts/check-skipped-required-checks.mjs does not exist, so the committed snapshot was compared against nothing. Run \`lisa apply\` to install the Lisa-managed script."
-            exit 1
-          fi
-          if [ ! -f .github/required-checks.json ]; then
-            echo "::error title=Required-checks declaration missing::.github/required-checks.json does not exist, so there is no committed snapshot to compare against the live ruleset. It is create-only and Lisa ships a seed — run \`lisa apply\` to get it."
-            exit 1
-          fi
-          configured="$(node -e "const d=require('./.github/required-checks.json');process.stdout.write(String(Boolean(d.ruleset&&d.ruleset.repo&&Array.isArray(d.ruleset.ids)&&d.ruleset.ids.length)))")"
-          if [ "$configured" != "true" ]; then
-            echo "UNCONFIGURED: .github/required-checks.json has no \`ruleset.repo\` / \`ruleset.ids\`, so drift cannot be checked."
-            echo "Fill them in to arm this: gh api repos/OWNER/NAME/rulesets --jq '.[] | \"\\(.id) \\(.name)\"'"
-            echo "::warning title=Required-checks drift UNCHECKED::Fill in ruleset.repo and ruleset.ids in .github/required-checks.json to arm the remote drift check."
-            exit 0
-          fi
-          node scripts/check-skipped-required-checks.mjs --remote