The Agent Skills Directory

[DATA_EXPOSURE]: The skill reads session transcripts from local directories such as ~/.claude/projects/ and ~/.local/share/opencode/. Access to these files is necessary for the skill's stated purpose of auditing past activities, and no data is exfiltrated externally.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from past session logs. The instructions mitigate the risk of the agent executing embedded instructions by requiring the output to be displayed in a code block, ensuring the content is treated as data rather than instructions.
[COMMAND_EXECUTION]: The skill uses shell scripts to process session data. It implements defensive measures, including validating session IDs against a strict alphanumeric whitelist and using a dedicated escaping function for SQL queries to prevent injection via manipulated directory names or session identifiers.

session-outline