Roll back the PR-self-check-in guard and draft-merge workaround (#152)

Two Claude-side settings that drove PR churn — draft-by-default PRs and
"Autofix pull requests" (auto-subscribe + hourly self-check-in) — have been
turned off at the source. The guards the corpus grew to smooth those over now
guard a phantom, so remove them:

- checks/pretooluse-guard.mjs: drop the deferred-self-check-in branch
  (send_later / ScheduleWakeup / create_trigger). The remote-branch-delete
  guard stays.
- .claude/settings.json + bootstrap.md: PreToolUse matcher back to `Bash`.
- checks/test/guard.test.mjs: drop the self-check-in fixtures.
- skills/merge-to-main: drop the "un-draft before merging" step — PRs are no
  longer opened as draft.

Kept from the lessons pass: the get_status-misses-Actions-CI gotcha in
git-github-advanced — reading CI via check runs is setting-independent.

Closes #154. Refs #151.


Claude-Session: https://claude.ai/code/session_01HzjetxNGYuN5wwF1vRdkA8

Co-authored-by: Claude <noreply@anthropic.com>

Author: missingbulb

.claude/settings.json

@@ -16,7 +16,7 @@
     ],
     "PreToolUse": [
       {
-        "matcher": "Bash|send_later|create_trigger|ScheduleWakeup",
+        "matcher": "Bash",
         "hooks": [
           {
             "type": "command",

bootstrap.md

@@ -148,10 +148,10 @@ The corpus's enforceable rules run as deterministic checks — usage, configurat
 
 No ordering constraint: Stop fires at end of turn, long after the SessionStart sync (Method B) or submodule update (Method A) has populated `.claudinite/`.
 
-**2.** Register the PreToolUse guard alongside it (same file; skip if present). It deterministically blocks actions the corpus forbids outright — currently remote-branch-delete pushes (which fail in this environment) and deferred PR self-check-in scheduling (a `send_later`/`ScheduleWakeup`/`create_trigger` call to "confirm CI goes green" or re-arm a PR watch — query the check-run status directly instead):
+**2.** Register the PreToolUse guard alongside it (same file; skip if present). It deterministically blocks commands the corpus forbids outright — currently remote-branch-delete pushes, which fail in this environment:
 
 ```json
-{ "hooks": { "PreToolUse": [ { "matcher": "Bash|send_later|create_trigger|ScheduleWakeup", "hooks": [
+{ "hooks": { "PreToolUse": [ { "matcher": "Bash", "hooks": [
   { "type": "command", "command": "node $CLAUDE_PROJECT_DIR/.claudinite/checks/pretooluse-guard.mjs" }
 ] } ] } }
 ```

checks/pretooluse-guard.mjs

@@ -1,51 +1,23 @@
 #!/usr/bin/env node
-// Claude Code PreToolUse guard: blocks actions the corpus forbids outright, before
-// they run. Exit 2 blocks the tool call and feeds stderr back to the agent.
-// Registered per-repo — see bootstrap.md. Two forbidden actions today:
-//   1. Bash pushes that delete a remote branch (the delete-push fails in this
-//      environment, so it can never succeed).
-//   2. Scheduling a deferred self-check-in to confirm CI / babysit a PR
-//      (send_later / ScheduleWakeup / create_trigger): a webhook that won't push
-//      CI-success is a reason to *query* the status directly, not to poll on a
-//      re-arming cadence.
+// Claude Code PreToolUse guard: blocks Bash commands that delete a remote branch
+// (the delete-push fails in this environment, so it can never succeed).
+// Exit 2 blocks the tool call and feeds stderr back to the agent. Registered
+// per-repo — see bootstrap.md.
 let input = '';
 process.stdin.on('data', (d) => { input += d; });
 process.stdin.on('end', () => {
   let payload = {};
   try { payload = JSON.parse(input); } catch { /* no payload → allow */ }
-  const name = payload.tool_name ?? '';
-
-  if (name === 'Bash') {
-    const cmd = payload.tool_input?.command ?? '';
-    const deletesRemoteBranch =
-      /\bgit\s+push\b[^\n;&]*\s(--delete|-d)\s/.test(cmd) ||
-      /\bgit\s+push\b[^\n;&]*\s\S+\s+:\S/.test(cmd);
-    if (deletesRemoteBranch) {
-      process.stderr.write(
-        'Blocked: never delete a remote branch — a current environment bug makes the delete-push fail, so it cannot succeed. Leave the branch; it can be deleted from the GitHub UI if needed.'
-      );
-      process.exit(2);
-    }
-    process.exit(0);
+  if (payload.tool_name !== 'Bash') process.exit(0);
+  const cmd = payload.tool_input?.command ?? '';
+  const deletesRemoteBranch =
+    /\bgit\s+push\b[^\n;&]*\s(--delete|-d)\s/.test(cmd) ||
+    /\bgit\s+push\b[^\n;&]*\s\S+\s+:\S/.test(cmd);
+  if (deletesRemoteBranch) {
+    process.stderr.write(
+      'Blocked: never delete a remote branch — a current environment bug makes the delete-push fail, so it cannot succeed. Leave the branch; it can be deleted from the GitHub UI if needed.'
+    );
+    process.exit(2);
   }
-
-  // send_later / create_trigger arrive MCP-namespaced (mcp__<server>__send_later);
-  // ScheduleWakeup is a built-in. Match by suffix so server casing doesn't matter.
-  const schedulesWakeup =
-    name === 'ScheduleWakeup' || name.endsWith('send_later') || name.endsWith('create_trigger');
-  if (schedulesWakeup) {
-    const text = payload.tool_input?.message ?? payload.tool_input?.prompt ?? '';
-    const isSelfCheckIn =
-      /\bself[-\s]?check[-\s]?in\b/i.test(text) ||
-      /\bre-?arm\b/i.test(text) ||
-      /\bconfirm\b[^.]*\bgreen\b/i.test(text);
-    if (isSelfCheckIn) {
-      process.stderr.write(
-        "Blocked: don't schedule a deferred self-check-in to confirm CI or babysit a PR. A webhook not delivering CI-success is a reason to *query*, not to poll on a re-arming cadence. Ask the platform directly for the check-run / workflow-run status (e.g. pull_request_read get_check_runs) on a short backoff bounded to the run's own duration, until it settles: green → report and stop, red → act. Don't re-arm — a stale-but-open PR awaiting human review is a safe state to end your turn on."
-      );
-      process.exit(2);
-    }
-  }
-
   process.exit(0);
 });

checks/test/guard.test.mjs

@@ -34,29 +34,3 @@ test('allows ordinary pushes and non-Bash tools', () => {
     assert.equal(runGuard(payload).status, 0, JSON.stringify(payload));
   }
 });
-
-const CHECKIN =
-  'Self check-in for PR #43. Re-check the PR state, CI status, mergeability, and any new review comments. If nothing changed, re-arm the next check-in silently ~1h out. Stop once the PR is merged or closed.';
-
-test('blocks a deferred PR self-check-in across scheduling tools', () => {
-  for (const payload of [
-    { tool_name: 'mcp__claude-code-remote__send_later', tool_input: { delay_minutes: 60, message: CHECKIN } },
-    { tool_name: 'ScheduleWakeup', tool_input: { prompt: CHECKIN, delaySeconds: 3600 } },
-    { tool_name: 'mcp__claude-code-remote__create_trigger', tool_input: { prompt: 'Schedule a self check-in to re-check PR #7 hourly.' } },
-    { tool_name: 'ScheduleWakeup', tool_input: { prompt: 'Wait for CI, then confirm it goes green and report back.', delaySeconds: 600 } },
-  ]) {
-    const r = runGuard(payload);
-    assert.equal(r.status, 2, JSON.stringify(payload));
-    assert.match(r.stderr, /query|check-run|re-arm|directly/i);
-  }
-});
-
-test('allows legit scheduled reminders and unrelated tools', () => {
-  for (const payload of [
-    { tool_name: 'mcp__claude-code-remote__send_later', tool_input: { delay_minutes: 30, message: 'Remind me to review the design doc with Bob.' } },
-    { tool_name: 'ScheduleWakeup', tool_input: { prompt: 'Resume the migration once the data export finishes.', delaySeconds: 900 } },
-    { tool_name: 'Edit', tool_input: {} },
-  ]) {
-    assert.equal(runGuard(payload).status, 0, JSON.stringify(payload));
-  }
-});

skills/git-github-advanced/SKILL.md

@@ -87,6 +87,10 @@ When the escalation is itself a separate reusable workflow invoked via `workflow
 
 On `pull_request`, `actions/checkout` checks out the synthetic `refs/pull/N/merge` commit by default — an ephemeral merge of the PR head into the base — so `HEAD` is itself a merge commit, not the branch tip. Any job that reasons about commit *shape* or history (merge commits, commit count, `--first-parent`, commit messages) reads that synthetic merge and misfires: a "no merge commits" check false-positives even on a linear branch, since `HEAD` is a merge. Check out the head SHA (`ref: ${{ github.event.pull_request.head.sha || github.sha }}`) so CI evaluates the same branch tip a local run does.
 
+## To read a PR's CI result, look at its check runs — `get_status` misses Actions
+
+GitHub **Actions** reports results as **check runs**, not the legacy **commit statuses**, so `pull_request_read` `method: get_status` returns `state: pending`, `total_count: 0` for a PR whose Actions CI has already **passed** — reading falsely as "no CI / not started," which can skip a real gate or trigger an endless wait. Gate on CI by reading the **check runs for the PR head SHA** (`get_check_run`, or the workflow run for that SHA) instead. Target the head SHA directly rather than `actions_list`-ing the repo's runs — that returns every run, its output is huge, and a specific check-run/SHA query is both correct and cheap.
+
 ## Mark large committed fixtures `linguist-vendored` to fix language stats
 
 Large committed fixture files (full-page HTML, generated data dumps) can dwarf actual source by byte count and cause GitHub to mislabel the repo's primary language. Add a `.gitattributes` entry for each such path (e.g. `test/fixtures/*.html linguist-vendored`) to tell Linguist to ignore it; apply the same annotation whenever you add another large generated or fixture file.