The Agent Skills Directory

[Data Exposure & Exfiltration] (SAFE): The skill demonstrates strong security awareness. It explicitly instructs users and the AI agent not to include sensitive data like API keys, passwords, or tokens in handoff files. The validate_handoff.py script and the references/handoff-template.md emphasize secret detection and removal as a mandatory step before finalizing documents.
[Command Execution] (LOW): The skill utilizes several local Python scripts (create_handoff.py, validate_handoff.py, list_handoffs.py, check_staleness.py) to automate tasks. These are standard utility scripts used for local project management. The commands observed in SKILL.md use static script paths and well-defined arguments, presenting minimal risk of command injection.
[Prompt Injection] (SAFE): No evidence of malicious prompt injection, behavioral overrides, or safety bypasses. The 'important context' and 'ignore' instructions are used legitimately within the context of managing session handoffs.
[Unverifiable Dependencies] (SAFE): The provided scripts appear to use standard Python libraries (like os, re, sys, pathlib). No external package installations or remote script executions (e.g., curl|bash) were found in the analyzed files.
[Persistence Mechanisms] (SAFE): The skill stores state information in a project-local directory (.claude/handoffs/). It does not attempt to modify shell profiles, system services, or registry keys for persistence.
[Indirect Prompt Injection] (LOW): While the skill processes handoff files which could theoretically contain instructions, it includes a robust validation step and a resume checklist that encourages human-in-the-loop verification of the context before proceeding.

session-handoff