pm test --run reports a passing name-filtered Bun command as failed empty_run (Bun's stderr summary is not accepted as an execution receipt)

[unbraind]

Problem

pm test --run marks a passing, name-filtered Bun test command as failed with failure_category: "empty_run" and rewrites its exit code to 1, although Bun ran tests, all passed, and it exited 0. The same file without the filter is classified passed.

Reproduced on @unbrained/pm-cli 2026.10.6, Bun 1.3.5 (also reported on 2026.10.4 / Bun 1.3.14 in a package CI):

mkdir repro && cd repro && git init -q && pm init --yes
mkdir test && cat > test/x.test.ts <<'TS'
import { test, expect } from "bun:test";
test("alpha one", () => { expect(1).toBe(1); });
test("alpha two", () => { expect(2).toBe(2); });
test("beta", () => { expect(3).toBe(3); });
TS
id=$(pm create --title "Repro" --type Task --json | jq -r .item.id)
pm test "$id" --add command="bun test --test-name-pattern alpha test/x.test.ts"
pm test "$id" --add command="bun test test/x.test.ts"
pm test "$id" --run --json | jq -c '.run_results[]|{command,status,exit_code,failure_category}'

Result:

{"command":"bun test --test-name-pattern alpha test/x.test.ts","status":"failed","exit_code":1,"failure_category":"empty_run"}
{"command":"bun test test/x.test.ts","status":"passed","exit_code":0,"failure_category":null}

Bun's own output for the filtered command (stderr):

 2 pass
 1 filtered out
 0 fail
 2 expect() calls
Ran 2 tests across 1 file. [51.00ms]

The runner recognizes Bun's name filter (that is why it looks for a positive execution receipt at all), but its receipt patterns apparently don't accept Bun's summary ( N pass / Ran N tests across M files). When a filter is present, the missing receipt is read as "the filter matched nothing".

Why it matters

Agents link narrow, fast regression commands to items (pm test <id> --add command="bun test --test-name-pattern …"), which is exactly the TDD flow pm recommends. With this bug every filtered Bun receipt is a false failure. A --validate-close gate then blocks closing a correctly tested item. Agents learn to drop the filter and run whole suites, which costs time and tokens, or to stop linking Bun tests.

Expected

• Treat Bun's summary as a positive execution receipt: ^\s*(\d+) pass with N > 0, or Ran (\d+) tests? across with N > 0, on stdout or stderr (Bun prints its summary to stderr).
• empty_run only when the receipt shows 0 tests ran (e.g. Ran 0 tests, or Bun's "did not match any test names" message).
• Never rewrite a zero exit code to 1 without a stated, matched reason. Include the matched (or missing) receipt pattern in the run result so the classification is explainable.

[unbraind]

Canonical recovery/evidence context: pm-f05lsg. The completed empty-selection guard precursor is pm-c1bn; its shipped false-pass protection must remain intact.

This new report remains open for the next reviewed delivery. The all-status exact empty_run search identifies the closed guard lineage, rather than an existing open Bun-specific duplicate. Required acceptance should use real Bun stdout/stderr with positive filtered execution, a genuine zero-match negative control, and unchanged pm test/test-all exit semantics. The reported false-negative classification has not been independently reproduced here and is not claimed fixed by PR1416. No completed precursor is reopened or duplicate recovery feature created.

[unbraind]

Canonical implementation owner: pm-gh1418.

The implementation, real Bun controls and immutable closure evidence are included in PR #1421.