The Agent Skills Directory

[DATA_EXFILTRATION]: The skill contains a 'Session Transcript Integration' feature that reads the agent's internal session logs located in ~/.claude/projects/ and exports them to a GitHub Gist using an external tool. While this is presented as a functional feature and the instructions require the agent to ask the user for permission, it involves transmitting the full conversation history (which may contain sensitive internal data or tool outputs) to an external service.
[INDIRECT_PROMPT_INJECTION]: The skill ingests and processes external data from chatwithgpt.substack.com via a sitemap and individual article fetches in scripts/update-cache.py. This data is then used to suggest references and inform article content. This creates a surface where malicious instructions embedded in the external Substack content could influence the agent's behavior during article generation, although the target domain appears to belong to the skill author.
[COMMAND_EXECUTION]: The instructions rely heavily on the execution of local shell commands, including ls, git log, and python3 to run internal management scripts. It also uses uvx to execute external Python packages.
[EXTERNAL_DOWNLOADS]: The skill utilizes the uvx utility to run the claude-code-transcripts package from a remote registry at runtime to facilitate the transcript export feature.

write-article