The Agent Skills Directory

[DATA_EXFILTRATION]: The skill accesses sensitive configuration files that manage agent permissions and execution settings.\n
Files targeted include ~/.claude/settings.json, ~/.claude/settings.local.json, and ~/.claude/permissions.allow.\n
The skill is designed to read, duplicate for backup, and modify these files to remove redundant permission entries using shell utilities.\n- [COMMAND_EXECUTION]: The skill provided instructions include multiple bash command blocks intended for execution by the agent to perform system audits and file operations.\n
The shell commands utilize tools such as grep for searching configuration files, jq for manipulating JSON permission structures, and find/du for identifying large or stale files.\n
File system modifications are performed via cp for backups and mv for soft-deletion (moving items to a trash directory).\n- [INDIRECT_PROMPT_INJECTION]: The skill ingests data from external files which could be manipulated by an attacker to include hidden instructions that influence the agent's behavior during the cleanup workflow.\n
Ingestion points: The agent is instructed to read content from SKILL.md files within ~/.claude/skills/ and markdown files within ~/.claude/**/memory/ to identify redundancy and staleness.\n
Boundary markers: Absent. There are no instructions for the agent to use specific delimiters or to disregard natural language instructions found within the data being scanned.\n
Capability inventory: The skill provides the agent with capabilities to modify permissions, move or rename files, and execute shell commands based on findings from the ingested data.\n
Sanitization: Absent. The skill does not implement validation or filtering for the content retrieved from the external skill or memory files.

config-gc