R Util Proj Cross

---
name: r-util-proj-cross
description: >
  Cross-project analysis for workspace project sets. Reads MEMORY.md, PLAN.md,
---

# /rdf:r-util-proj-cross — Cross-Project Analysis

Cross-project analysis for workspace project sets. Reads MEMORY.md, PLAN.md,
AUDIT.md, and CLAUDE.md from all projects to identify shared patterns,
duplicate effort, convention drift, and alignment opportunities.

## Projects

The workspace root is the parent directory of the current project (or
CWD itself when it is a non-git directory containing project repos).
Scan all directories under the workspace root that contain a
`CLAUDE.md` or `.git/` directory.

If the resolved workspace root is `$HOME` or `/`, stop and ask the user
for an explicit root — a home directory is not a workspace. Cap the scan
at 25 sibling directories and report when the cap is hit.

Also scan shared-library repos if present (directories whose name ends
in `_lib` or that a workspace profile's cross-project reference lists).

## Protocol

### 1. Open Work Overlap

Compare open PLAN.md items across projects. Flag:
- Same class of fix needed in multiple projects (e.g., "copyright
  year update", "bash 4.1 compat fix")
- Shared infrastructure improvements (test parallelism, CI matrix,
  Dockerfile patterns)
- Deferred items that could be batched across projects

### 2. Audit Finding Patterns

If AUDIT.md exists in multiple projects, compare:
- Same finding class appearing across projects
- Shared false positives (candidates for parent CLAUDE.md)
- Common severity distribution patterns

### 3. Convention Drift

Compare key patterns across project CLAUDE.md files:
- Commit message format consistency
- Verification checklist completeness
- Test infrastructure parity (parallelism, CI targets, OS matrix)
- Variable naming conventions
- Shell standards adherence

Flag differences that should be aligned. Distinguish intentional
per-project differences from unintentional drift.

### 4. Shared Library Drift

For each shared library the workspace profile's cross-project reference
lists (or any `*_lib` repo found in the scan), check all consuming
projects:

- Compare version variables across canonical and consumer copies
- Compute `sha256sum` of all copies — warn if checksums differ
  (even at same version, detects project-specific edits)
- Flag any project-specific references that leaked into the library
- Check that wrapper scripts (e.g., `files/tlog`) are consistent

### 5. MEMORY.md Lessons

Read all MEMORY.md "Lessons Learned" / "Mistakes Made" / feedback
memory sections. Flag:
- Lessons from one project applicable to others
- Anti-patterns documented in one project but not enforced in parent
  CLAUDE.md
- Patterns that should be promoted to shared conventions

### 6. Output

```
# Cross-Project Analysis ({date})

## Open Work Overlap ({N} batch opportunities)
| Item | Projects | Action |
|------|----------|--------|
| {description} | {project-a}, {project-b} | Batch across both |

## Convention Drift ({N} differences)
| Convention | {project-a} | {project-b} | {project-c} | Aligned? |
|-----------|-------------|-------------|-------------|----------|
| Commit format | {format-a} | {format-a} | {format-b} | NO |

## Shared Library Drift ({N} mismatches)
| Library | Canonical | {consumer-a} | {consumer-b} | Status |
|---------|-----------|--------------|--------------|--------|
| {lib} | 1.2.0 | 1.2.0 | 1.1.9 | DRIFT |

## Lessons to Propagate ({N})
- {lesson from project X} → add to parent CLAUDE.md § {section}

## Recommendations
1. {numbered action items}
```

## Rules
- Read-only — do NOT modify any files
- Report convention drift as differences, not violations — some
  differences are intentional (e.g., a project with a deliberately
  different commit-message format)
- For library drift, the canonical copy is authoritative
- Focus on actionable items — skip trivial differences
- If a project has no PLAN.md/AUDIT.md/MEMORY.md, note it and
  continue with available data