Agent Builder scans generated `.analysis` artifacts and required root `.memlog.md` as agent-source violations

[hughdw]

Clear description of the problem

bmad-agent-builder requires a builder process log at <agent>/.memlog.md for resume detection, and its Analyze workflow writes reports to <agent>/.analysis/<timestamp>/.

However:

1. scan-path-standards.py flags the required root .memlog.md as a high-severity “Prompt file at skill root” violation.
2. scan-path-standards.py scans generated .analysis/ Markdown/JSON files, including findings.json, where the Analyze schema requires absolute paths in standards. Those paths are then flagged as non-portable absolute paths.
3. prepass.py counts generated .analysis/ artifacts as part of the agent tree, so prior HTML reports can inflate token/file metrics for later Analyze runs.

This makes a clean agent appear to fail lint and lets prior analysis output influence subsequent analysis.

Steps to reproduce

1. Create an agent with Agent Builder, ensuring it has the required root-level .memlog.md.

2. Run Analyze on the agent, which creates:

   <agent>/.analysis/<timestamp>/findings.json
   <agent>/.analysis/<timestamp>/agent-analysis-report.html
   <agent>/.analysis/<timestamp>/agent-analysis-report.md
   

3. Run:

   uv run scripts/scan-path-standards.py <agent-path>
   uv run scripts/prepass.py <agent-path>

Expected vs actual behavior

Expected

• Root .memlog.md is treated as Agent Builder operational metadata, not a capability/prompt file.
• .analysis/ is treated as generated analysis output and excluded from path lint and pre-pass token/file counting.
• Lint findings reflect only authored Agent files.

Actual

• .memlog.md produces a high-severity structure finding.
• .analysis/.../findings.json produces high-severity absolute-path findings.
• .analysis/ HTML/Markdown/JSON files are included in pre-pass metrics.

Proposed fix

• In scan-path-standards.py:
  • Exempt root .memlog.md from the “only SKILL.md at root” rule.
  • Exclude .analysis/ from scanned Markdown/JSON files.
• In prepass.py:
  • Add .analysis to skipped directories.

Your environment

• Model: GPT-5 / Codex
• IDE: Codex Desktop
• BMad version: 6.11.0
• OS: macOS
• Python runner: uv

[bmadcode]

Closed by #114. All three symptoms reported here are fixed and verified (8 findings → 5 on a sample agent; the .memlog.md structure finding and both .analysis/findings.json absolute-path findings are gone).

Three adjacent defects found during review were kept out of scope and are tracked in #115 — most importantly, bmad-workflow-builder carries a near-verbatim copy of the same scanner with both original defects still present, so this bug class is only half-closed.

---

🤖 Posted by BMad Bot on behalf of the maintainers.