The Agent Skills Directory

[SAFE]: No malicious patterns or security risks were identified. The skill performs standard local file and command operations to provide cross-conversation continuity.\n- [COMMAND_EXECUTION]: The skill uses a provided Python script to manage its state via the command line.\n
Evidence: SKILL.md instructions include running python3 /home/workspace/Skills/handoff/scripts/handoff.py check at the start of every conversation.\n- [DATA_EXFILTRATION]: The skill reads and writes state information to a local file in the workspace data directory. This access is confined to the local filesystem and is necessary for the skill's intended function.\n
Evidence: scripts/handoff.py defines HANDOFF_FILE as /home/workspace/Data/handoff.json.\n- [PROMPT_INJECTION]: The skill creates a surface for indirect prompt injection because the agent is instructed to follow 'resume instructions' stored in a file from a previous session.\n
Ingestion points: Data is read from handoff.json during the boot check at the start of every conversation.\n
Boundary markers: Absent; the agent is instructed to 'Read the handoff context carefully' and follow the instructions in the resume field.\n
Capability inventory: The skill can execute local Python scripts and potentially trigger platform communication tools (SMS/Email).\n
Sanitization: Data is stored as structured JSON, but the content is treated as natural language instructions without explicit filtering or validation.

handoff