The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes several Python scripts that execute local commands, primarily git, via subprocess.run(). These operations are used to manage project history, detect file renames for provenance, and track changes in documentation. The commands use argument lists to prevent shell injection and are focused on local repository management.- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest external data including raw sources (articles, papers) and human-provided feedback in the audit/ directory. This creates an attack surface where untrusted data could contain instructions that attempt to override agent behavior during wiki compilation. However, this is a core requirement for a knowledge-base skill and is managed by the agent's internal safety filters.- [DATA_EXPOSURE]: The session_discovery.py script identifies and reads agent conversation history stored in local directories like ~/.claude/projects and ~/.codex/sessions. While this involves sensitive conversational data, the access is a documented primary feature of the skill to 'build a dated timeline from Git and session history,' and the data remains within the local wiki structure.

llm-wiki