fix: an empty backlog is indistinguishable from broken discovery

[Jammy2211]

Overview

Split out of #211 / #212, which fixed the broken discovery roots but deliberately left this behind as separate work.

When a conductor's selection mode finds nothing it prints a flat message — feature agent: no feature prompts found in PyAutoMind. and siblings. That is consistent with two very different situations: the backlog genuinely holds no prompts of that work-type, or discovery is pointed somewhere wrong (a bad PYAUTO_MIND, a non-Mind checkout, a layout the discoverer has not learned).

Case two is what #211 was. Three conductors reported an empty backlog for four weeks while sitting on 87 prompts, and nobody noticed — because the message a broken root produces is the message an empty backlog produces. That ambiguity, not the one-line wrong path, is why the bug survived. Diagnosing the empty case is the signal it lacked.

Plan

• Teach the sizing faculty to explain an empty discovery result against the Mind it was pointed at.
• Draw the three distinctions that matter: not a Mind checkout / no such work-type / genuinely empty backlog.
• Have the three conductors print that reason instead of the flat sentence, keeping their exit codes.
• Lock the three causes as mutually distinguishable in tests.

Detailed implementation plan

Affected Repositories

• PyAutoBrain (primary, and only)

Branch Survey

Repository	Current Branch	Dirty?
./PyAutoBrain	claude/automind-task-planning-tj0sdn (restarted from origin/main f4e9341)	clean

Suggested branch: claude/automind-task-planning-tj0sdn (session-designated)

Implementation Steps

1. agents/faculties/sizing/_sizing.py — add empty_discovery_reason(mind, work_type) beside discover_prompts. Classifies an empty result:

   • path is not a directory → bad path
   • directory has neither draft/ nor active.md → not a Mind checkout, name PYAUTO_MIND
   • no draft/<work-type>/ and no legacy flat root → no such work-type, and list the work-types that do exist
   • roots exist but hold no prompts → genuinely empty backlog

   Diagnosis only — callers keep their own exit codes, and discover_prompts keeps its signature and result.

2. The three conductors print the reason on their empty paths: _feature.py, _bug.py, _refactor.py.

3. tests/test_conductor_discovery.py — six new tests, including one asserting the three causes produce three distinct strings, and one covering a freshly-spawned Mind (has active.md, no draft/ yet) which must read as an empty backlog rather than a bad path.

Key Files

• agents/faculties/sizing/_sizing.py — gains empty_discovery_reason
• agents/conductors/{feature,bug,refactor}/ — call sites
• tests/test_conductor_discovery.py — extended

Scope

Message-and-diagnosis only. No change to discovery behaviour, ranking, or exit codes; the end-to-end non-empty assertions from #212 must keep passing unchanged.

Testing

pytest tests/ plus manual exercise of all three empty causes against fixture Minds, and confirmation the populated case is byte-identical.