Add failing tests for #1648: pdd sync subprocess runner may use installed pdd instead of checkout

[prompt-driven-github]

Summary

Adds failing tests that detect the bug reported in #1648, where _find_pdd_executable() in multiple modules prefers shutil.which("pdd") over [sys.executable, "-m", "pdd"], causing sync subprocesses to execute an installed binary (e.g., 0.0.276) rather than the current source-checkout version (e.g., 0.0.278.dev0).

Root Cause

_find_pdd_executable() in three separate files returns shutil.which("pdd") immediately when any pdd binary exists on PATH, so the else: [sys.executable, "-m", "pdd"] branch in callers is never reached. The same antipattern exists in the MCP runner tool. This contradicts the documented intent in pdd/__main__.py and the correct pattern already used in pdd/agentic_change_orchestrator.py.

Files with the bug:

• pdd/agentic_sync_runner.py:236 — primary _find_pdd_executable()
• pdd/server/routes/commands.py:434 — duplicate copy
• pdd/server/jobs.py:199 — duplicate copy
• utils/mcp/pdd_mcp_server/tools/runner.py:66 — MCP runner

Test Files

• Unit tests: tests/test_agentic_sync_runner.py (Tests 1 & 6: TestBuildCommandInterpreterParity)
• Unit tests: tests/test_jobs_sync_failure_detection.py (Test 2: TestBuildSubprocessCommandArgsInterpreterParity)
• Unit tests: tests/test_server_routes_commands.py (Test 3: TestBuildPddCommandArgsInterpreterParity) — new file
• Unit tests: tests/test_mcp_runner.py (Test 4: TestRunPddCommandInterpreterParity + prompt regression guards) — new file
• Unit tests: tests/test_run_pdd_sync_helper.py (Test 5) — new file
• E2E tests: tests/test_e2e_sync_interpreter_parity.py — new file (real PATH manipulation, no mocks)

Prompt Files Fixed

Prompt specifications encoded the wrong executable-resolution strategy and have been corrected to mandate [sys.executable, "-m", "pdd"]:

• utils/mcp/prompts/runner_python.prompt
• pdd/prompts/agentic_sync_runner_python.prompt
• pdd/prompts/server/routes/commands_python.prompt
• pdd/prompts/server/jobs_python.prompt

What This PR Contains

• 6 failing unit tests that reproduce the reported bug (fail on current code, will pass once fixed)
• 4 failing E2E tests that verify the bug using real OS PATH manipulation (no shutil.which mocking)
• 2 passing prompt regression guards (prompt was already corrected; guards against reintroduction)
• Prompt file fixes ensuring code regenerated from specs won't reintroduce the bug

Test Verification

All 6 behavioral unit tests and 4 E2E tests fail on current buggy code with clear error messages pointing to the exact binary path resolved incorrectly. Example:

AssertionError: Expected sys.executable ('/opt/venv/bin/python') as first element,
got '/opt/anaconda3/bin/pdd'. Bug: _find_pdd_executable() returned the PATH binary.

Next Steps

1. Fix _find_pdd_executable() in pdd/agentic_sync_runner.py to always return None
2. Fix _find_pdd_executable() in pdd/server/routes/commands.py to always return None
3. Fix _find_pdd_executable() in pdd/server/jobs.py to always return None
4. Fix run_pdd_command() in utils/mcp/pdd_mcp_server/tools/runner.py to use [sys.executable, "-m", "pdd"]
5. Fix _run_pdd_sync() in tests/test_one_session_eval.py
6. Verify all 10 tests pass
7. Run full test suite to check for regressions
8. Mark PR as ready for review

To auto-fix using PDD:

pdd fix https://github.com/promptdriven/pdd/issues/1648

Fixes #1648

---

Generated by PDD agentic bug workflow