@cat-factory/executor-harness 1.126.0 → 1.129.0

package/dist/agent.d.ts

@@ -53,6 +53,22 @@ export declare function testSecretEnv(secrets: TestSecretSpec[] | undefined): Re
  * `process.env` — and a second copy is a second place to forget the redaction.
  */
 export declare function secretEnv(secrets: TestSecretSpec[] | undefined): Record<string, string>;
+/**
+ * Which refs a REUSED (warm-pool) explore checkout must end up holding: the branch to explore, and
+ * the repo's own base branch beside it.
+ *
+ * Its own function, and named, because the bug it removes is a swap between two branch names both
+ * in scope at the call site, and nothing downstream can tell them apart. A read-only reviewer's
+ * whole instruction is `git diff origin/<base>...HEAD`, so the base ref has to be as fresh as the
+ * branch: passing the explored branch as the base collapses the two refspecs into one, leaves
+ * `origin/<base>` at whatever tip the pool dir was first cloned at, and moves the merge base back
+ * to that tip. The diff then reports every commit merged into base since as part of the change
+ * under review, which is wrong in the direction nothing notices.
+ */
+export declare function exploreCheckoutRefs(job: Pick<AgentJob, 'branch' | 'repo'>): {
+    branch: string;
+    baseBranch: string;
+};
 /**
  * Whether a Ralph iteration ({@link AgentJob.validation} set) landed on a MULTI-REPO job (writable
  * peer repos or read-only reference repos). The post-commit validation command is only wired into

package/dist/agent.js

@@ -383,6 +383,21 @@ export function secretEnv(secrets) {
     registerKnownSecrets(secrets.map((s) => s.value));
     return Object.fromEntries(secrets.map(({ key, value }) => [key, value]));
 }
+/**
+ * Which refs a REUSED (warm-pool) explore checkout must end up holding: the branch to explore, and
+ * the repo's own base branch beside it.
+ *
+ * Its own function, and named, because the bug it removes is a swap between two branch names both
+ * in scope at the call site, and nothing downstream can tell them apart. A read-only reviewer's
+ * whole instruction is `git diff origin/<base>...HEAD`, so the base ref has to be as fresh as the
+ * branch: passing the explored branch as the base collapses the two refspecs into one, leaves
+ * `origin/<base>` at whatever tip the pool dir was first cloned at, and moves the merge base back
+ * to that tip. The diff then reports every commit merged into base since as part of the change
+ * under review, which is wrong in the direction nothing notices.
+ */
+export function exploreCheckoutRefs(job) {
+    return { branch: job.branch, baseBranch: job.repo.baseBranch };
+}
 /**
  * Read-only exploration: clone `branch`, run the agent making no edits, and return its
  * prose report — or, when `output.kind==='structured'`, the parsed JSON object as
@@ -408,12 +423,17 @@ async function runExploreMode(job, opts) {
         let workDir;
         if (job.persistentCheckout) {
             logger.info('agent(explore): preparing reused checkout');
+            // Both refs, resolved by {@link exploreCheckoutRefs}: which one is the base is the whole
+            // decision, so it is made there rather than inline here.
+            //
+            // `job.full` is deliberately not consulted: the fresh-clone leg inside
+            // `prepareExistingCheckout` always clones with full history, so a reused checkout already
+            // has the merge base a shallow explore clone would not.
             await prepareExistingCheckout({
                 dir,
                 repo: job.repo,
                 ghToken: job.ghToken,
-                branch: job.branch,
-                baseBranch: job.branch,
+                ...exploreCheckoutRefs(job),
                 existing: true,
                 signal: opts.signal,
             });

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@cat-factory/executor-harness",
-  "version": "1.126.0",
+  "version": "1.129.0",
   "description": "Container payload: a thin TypeScript wrapper that runs the Pi coding agent against a cloned repo and opens a PR. Runs in the Cloudflare Container (and, in local native mode, as a host process); carries no secrets.",
   "repository": {
     "type": "git",
@@ -25,14 +25,14 @@
     "access": "public"
   },
   "devDependencies": {
-    "@hono/node-server": "^2.1.0",
+    "@hono/node-server": "^2.1.1",
     "@types/node": "^26.2.0",
-    "hono": "^4.13.1",
+    "hono": "^4.13.3",
     "typescript": "7.0.2",
-    "vitest": "^4.1.10",
-    "@cat-factory/kernel": "0.309.0",
-    "@cat-factory/server": "0.294.0",
-    "@cat-factory/spend": "0.16.1"
+    "vitest": "^4.1.11",
+    "@cat-factory/kernel": "0.316.0",
+    "@cat-factory/server": "0.302.0",
+    "@cat-factory/spend": "0.16.9"
   },
   "scripts": {
     "build": "tsc -p tsconfig.json",

package/src/agent.ts

@@ -464,6 +464,25 @@ export function secretEnv(secrets: TestSecretSpec[] | undefined): Record<string,
   return Object.fromEntries(secrets.map(({ key, value }) => [key, value]))
 }
 
+/**
+ * Which refs a REUSED (warm-pool) explore checkout must end up holding: the branch to explore, and
+ * the repo's own base branch beside it.
+ *
+ * Its own function, and named, because the bug it removes is a swap between two branch names both
+ * in scope at the call site, and nothing downstream can tell them apart. A read-only reviewer's
+ * whole instruction is `git diff origin/<base>...HEAD`, so the base ref has to be as fresh as the
+ * branch: passing the explored branch as the base collapses the two refspecs into one, leaves
+ * `origin/<base>` at whatever tip the pool dir was first cloned at, and moves the merge base back
+ * to that tip. The diff then reports every commit merged into base since as part of the change
+ * under review, which is wrong in the direction nothing notices.
+ */
+export function exploreCheckoutRefs(job: Pick<AgentJob, 'branch' | 'repo'>): {
+  branch: string
+  baseBranch: string
+} {
+  return { branch: job.branch, baseBranch: job.repo.baseBranch }
+}
+
 /**
  * Read-only exploration: clone `branch`, run the agent making no edits, and return its
  * prose report — or, when `output.kind==='structured'`, the parsed JSON object as
@@ -490,12 +509,17 @@ async function runExploreMode(job: AgentJob, opts: RunOptions): Promise<AgentRes
       let workDir: string
       if (job.persistentCheckout) {
         logger.info('agent(explore): preparing reused checkout')
+        // Both refs, resolved by {@link exploreCheckoutRefs}: which one is the base is the whole
+        // decision, so it is made there rather than inline here.
+        //
+        // `job.full` is deliberately not consulted: the fresh-clone leg inside
+        // `prepareExistingCheckout` always clones with full history, so a reused checkout already
+        // has the merge base a shallow explore clone would not.
         await prepareExistingCheckout({
           dir,
           repo: job.repo,
           ghToken: job.ghToken,
-          branch: job.branch,
-          baseBranch: job.branch,
+          ...exploreCheckoutRefs(job),
           existing: true,
           signal: opts.signal,
         })