The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes session transcripts which are inherently untrusted external data. While this presents an attack surface where malicious instructions could be embedded in a transcript, the skill mitigates this risk through an extensive sanitization pipeline.
Ingestion points: The skill reads JSONL, JSON, and Markdown transcript files from the local filesystem (e.g., in scripts/share_session.py).
Boundary markers: The skill does not rely on LLM-based parsing of the transcript; instead, it uses a Python script to perform structural normalization and cleaning, which prevents the transcript content from influencing the agent's logic during the upload process.
Capability inventory: The skill can perform network POST requests to the vendor's domain, read local transcript files, and update specific fields in the user's local profile (profile.json).
Sanitization: scripts/share_session.py implements multiple layers of protection: SECRET_REDACTIONS (removes GitHub tokens, API keys, and private keys), METADATA_PATTERNS (strips host-injected XML/Markdown envelopes), and HOST_BLOCK_DROP_RE (drops blocks containing system notifications or environment context).
[DATA_EXFILTRATION]: The skill transmits conversation transcripts to https://lovstudio.ai. This is the primary intended function of the skill, and the destination is a vendor-owned domain associated with the author. The skill includes logic to prevent accidental leakage of sensitive environment data and credentials during this process, fulfilling a best-practice approach to session sharing.
[COMMAND_EXECUTION]: The skill executes Python scripts (scripts/share_session.py and scripts/profile_store.py) to perform its tasks. These scripts use standard libraries and do not invoke arbitrary shell commands with untrusted input. The profile_store.py script includes path validation to ensure only permitted sections of the user profile are modified.

lov-share-session