/magic:propose

Run a propose-then-act cycle for the current data processing task.

Steps:

1. Read workspace_state.md to understand the current objective and progress
2. Inspect the data — run discovery scripts as appropriate: detect_issues.py (cleaning skill, Tier A scriptable), detect_all_issues.py (profiling skill, Tier A scriptable), detect_patterns.py (exploration skill, Tier A scriptable). All three can be called directly via CLI.
3. Present findings as numbered options with:
   • What was found (issue type, affected columns, severity)
   • Proposed approach for each finding (clean, transform, synthesize, skip)
   • Estimated impact (rows affected, data loss risk)
4. Wait for user to choose which options to pursue
5. Record the decision in analysis_journal.md
6. Hand off to the appropriate skill for execution

Output format:

## Processing Proposal

### 1. [Issue title] — [severity: HIGH/MEDIUM/LOW]
**Affected:** [column(s)], [N rows] ([X%] of data)
**Finding:** [concrete description with sample values]
**Options:**
  A. [approach] — [estimated impact]
  B. [approach] — [estimated impact]
  C. Skip — flag for manual review
**Recommended:** [option] because [rationale]

### 2. ...

Key rules:

• Always present options before acting
• Use uncertainty language ("suggests", "appears to", "may indicate")
• Include confidence levels (high/medium/low) for each finding
• Never force a single path — always offer alternatives

See also: /magic:findings for raw findings, /magic:decide to record choices, /magic:spec to formalize as a data spec.