soulkeeper
por cassh100k
Coding Agents & IDEsResumen
**Identity persistence for AI agents.**.
Descripción
Qué hace esta skill
SoulKeeper gives AI agents three Python scripts: one parses soul files into structured rules JSON, one scores session transcripts for behavioral drift, and one injects context-aware reminders before the agent responds. Agents lose their defined behavior between sessions without knowing it, and violations are traced to specific source lines.
Violations are cited with specific line numbers from the source soul files, making corrections actionable rather than vague.
Cuándo usarla
- Checking a session transcript for tone and behavior violations after a long agent run
- Injecting rule reminders before an agent responds to a sensitive or complex request
- Parsing SOUL.md files into structured JSON for use in automated validation
- Running CI checks on agent output that return exit code 1 on high drift scores
- Resetting an agent's behavioral context at session start with a heartbeat reminder
Flujo de trabajo de ejemplo
User asks: check this session transcript for drift
- 1
Run audit.py against the workspace to parse soul files into soul_rules.json
- 2
Pass the transcript and rules file to drift.py --report
- 3
drift.py matches each agent response against violation patterns by severity category
- 4
Produce a scored report with rule IDs, violation counts, and SOUL.md line references
Drift report showing score 42 (MODERATE DRIFT) with 2 critical and 3 high-severity violations traced to specific lines in SOUL.md
Instalación
clawhub install cassh100k/soulkeeperEjecuta el comando de la CLI o pega el prompt en Claude Code, Cursor, Codex o el agente de programación que uses.