sentinel-export

Sentinel Export — Compliance Lint + Format Rendering

Overview

Produces submission-ready documents from project docs. Runs compliance lint (AI writing pattern detection) and renders output using user-provided templates.

Policy lives in .sentinel/export/compliance.py. This skill orchestrates only.

Language Behavior

Default to English. If the user writes in Chinese, respond in Chinese. Match the user's language throughout.

Workflow

Step 0: Read Export Directory

List docs/export/ contents:

• Templates (.tex, .pptx, .docx)
• Rubrics (*rubric*)
• README.md

If docs/export/ doesn't exist, create it with output/ subdirectory.

Step 1: Identify Content Sources

Use AskUserQuestion: "Which documents do you want to export?"

Options:

• Specific files (e.g., ARCHITECTURE.md, docs/sessions/2026-03-06.md)
• "All project docs"
• Custom selection

Step 2: Determine Output Format

Use AskUserQuestion: "What output format?"

• Markdown → proceed directly
• PDF / PPTX / DOCX → check docs/export/ for matching template
  • Template found → confirm with user
  • No template → STOP: "No [format] template found in docs/export/. Please add one and re-run."

Step 3: Run Compliance Lint

PYTHONPATH=.sentinel python -m export.compliance --input <file>

For multiple files, run on each and aggregate results.

Step 4: Present Findings

Categorize by type, show counts:

## Compliance Report

| Type | Count | Auto-fixable |
|------|-------|--------------|
| Chain-of-thought (T1) | 3 | 0 |
| Structural tells (T2) | 7 | 4 |
| Phantom references (T3) | 1 | 0 |
| Statistical (T4) | - | - |

Total: 11 findings (4 auto-fixable, 7 need review)

Use AskUserQuestion with options:

1. "Apply auto-fixes (4 items) and review the rest"
2. "Show me all findings — I'll fix manually"
3. "Skip compliance"

Step 5: Apply Fixes

If user chose auto-fix:

1. Apply auto-fixes via apply_auto_fixes()
2. Re-run compliance to get fresh offsets
3. Show delta: "Fixed 4 items. 7 findings remain for manual review."
4. List remaining findings with line numbers and suggestions

Step 6: Rubric Check (optional)

If a rubric file exists in docs/export/ (filename contains "rubric"):

• Parse rubric criteria
• Evaluate content against criteria
• Present score and gaps

If no rubric: skip silently.

Step 7: Render Output

• Markdown: Write cleaned content to docs/export/output/
• Non-markdown: Invoke the appropriate document skill with the user's template
  • PDF: Use LaTeX template → compile
  • PPTX: Use /pptx skill with template
  • DOCX: Use /docx skill with template
• All output goes to docs/export/output/

Step 8: Final Summary

## Export Complete

- Source: ARCHITECTURE.md
- Format: Markdown
- Compliance: 11 findings → 4 auto-fixed, 7 reviewed
- Output: docs/export/output/ARCHITECTURE-export.md

Anti-Patterns

Do NOT	Do Instead
Generate random template for non-md	Ask user to provide template in docs/export/
Auto-fix Type 4 findings	Flag for human rewrite
Auto-fix borderline vocabulary (T2-D2)	Show suggestion, let user decide
Skip compliance without user consent	Always run; let user opt out at Step 4
Overwrite source documents	Write to docs/export/output/ only
Describe compliance patterns here	Reference compliance.py — policy in code
Scan inside code blocks or front matter	compliance.py handles segmentation
Run Type 4 without asking	Type 4 is opt-in; ask user if they want statistical analysis