Workspace Erstellen

2. Block deletion:

• Never auto-delete files.
• Replace delete actions with move-to-archive behavior.
• Use archive/ as destination.

3. Enforce data zones:

• data/raw is read-only after import.
• data/clean allows transformations.
• data/output stores generated results only.

4. Enforce one-way flow:

• Allow only RAW -> CLEAN -> OUTPUT.
• Never write from output or clean back into raw.

5. Enforce loop protection:

• Track processing state in workspace_state.json.
• Process each input file at most once.
• If already logged as processed, skip.

6. Enforce operation logging:

• Write action logs to logs/YYYY-MM-DD.log.
• Record created folders, moved files, validation outcomes, outputs, and errors.

Execution Logic (Deterministic)

Execute these steps in order and stop after completion:

1. Resolve scope:

• Set WORKSPACE_ROOT to current directory.
• Initialize/ensure workspace_log.json.
• Initialize/ensure workspace_state.json.

2. Create standard folders if missing:

• data/raw
• data/clean
• data/output
• automation
• reports
• dashboard
• logs
• archive
• archive/errors

3. Discover import candidates in WORKSPACE_ROOT:

• Allowed input extensions: .csv, .xlsx, .json.
• Ignore files already inside data/, archive/, and logs/.

4. Import phase:

• Move discovered .csv and .xlsx (and .json if present) to data/raw.
• Log every move.

5. Validate phase for raw inputs:

• Run format, encoding, and structure checks.
• CSV requirements: UTF-8 and delimiter ;.
• If CSV is non-compliant, convert to compliant format into data/clean and log conversion.

6. Error handling:

• On validation errors (for example missing columns, invalid dates), log error details.
• Move failing file to archive/errors.
• Do not silently alter source data in place.

7. Processing phase:

• Write transformed outputs only to data/clean and generated artifacts to data/output or reports.

8. Finalize:

• Update workspace_state.json with processed file fingerprints/status and status transitions.
• Append completion entry to daily log.
• Stop execution; do not re-trigger automatically.

Platform Architecture

1) Skill Registry

Maintain a registry of available skills, for example:

• workspace_erstellen
• csv_analysieren
• daten_bereinigen
• report_generieren
• dashboard_erstellen
• archivieren

Each registry entry must define:

• name
• input
• output
• scope
• safety_level

2) Skill Executor

Enforce:

• run exactly one skill at a time
• block direct skill-to-skill chaining outside executor control

Execution path:

• Command -> Skill Lookup -> Safety Check -> Execution -> Logging -> Finish

3) State Manager

Persist state in workspace_state.json, minimum schema:

{
  "processed_files": [],
  "active_skill": "",
  "last_execution": "",
  "status": "idle"
}

If an input file is already in processed_files, skip it.

4) Safety Guard

Block at least:

• delete operations
• writes outside workspace scope
• overwrite attempts on raw data
• unsafe mass changes
• self-trigger loops

AI Execution Contract

Allow only:

1. Analyze command.
2. Select one matching skill.
3. Execute that skill.
4. Log result.

Do not start new autonomous processes. Rule: one command -> one skill -> one result.

Idempotency Rules

• Re-running initialization must not duplicate folders.
• Re-running import must skip already processed files based on workspace_state.json.
• Re-running processing must not overwrite raw files.

Minimal Prompt

Use this prompt when invoking the skill:

Skill: Workspace erstellen
Goal: Create a secure workspace structure in the current directory.
Rules:
1) Work only inside the current directory.
2) Never delete files automatically.
3) Create folders: data/raw, data/clean, data/output, automation, reports, dashboard, logs, archive.
4) Move CSV and XLSX files into data/raw.
5) RAW files are read-only.
6) Processing output must go into data/output.
7) Write all operations into logs.
8) Prevent repeated processing using a workspace_log file.
8) Prevent repeated processing using a workspace_state file.
9) Stop after initialization is complete.
Final state: workspace ready for data processing.

Reference

Load references/foundation-rules.md for policy details and rationale. Load references/framework-architecture.md for strategic, platform, and roadmap details.