The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes the ax command-line utility for data ingestion and proposal management. It also runs a bundled Python script scripts/turns.py to parse local session logs. These operations are consistent with the skill's stated purpose of session analysis.
[INDIRECT_PROMPT_INJECTION]: The skill processes historical session logs located in ~/.claude/projects/. Since these logs contain the output of all tools used in previous sessions (including content from external websites or third-party files), they constitute an untrusted data source. A malicious payload encountered in a previous session could theoretically influence the skill to propose harmful 'hook' or 'automation' changes to the agent's environment.
[DATA_EXPOSURE]: The skill reads sensitive session logs from ~/.claude/projects/ which contain a full history of agent interactions, code, and tool outputs. This data is processed locally by the bundled script and the ax toolset.
[PRIVILEGE_ESCALATION]: The skill suggests creating or modifying files in ~/.claude/hooks/, which are shell scripts executed before tool calls. While the skill instructs the agent to 'apply only when asked' and requires user triage, the ability to generate executable hooks is a high-privilege capability that requires careful user oversight.

tune