The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill is designed to read and audit the agent's internal configuration files, specifically ~/.claude/settings.json and ~/.claude/settings.local.json. These files typically contain sensitive environment information, tool configurations, and potentially authentication secrets or tokens for connected services.
[COMMAND_EXECUTION]: The instructions provide several shell commands for the agent to execute, including complex find, du, grep, and jq pipelines. These commands are used to identify files for deletion and to programmatically modify the agent's JSON configuration files.
[PRIVILEGE_ESCALATION]: The skill provides instructions and specific jq commands to modify the agent's permission whitelist (permissions.allow in settings.json). While the intended use is to remove redundant entries, this mechanism provides a direct interface for altering the agent's security policies and access controls.
[INDIRECT_PROMPT_INJECTION]: The skill introduces a surface for indirect prompt injection by scanning and reading the contents of potentially untrusted local files.
Ingestion points: Files located in ~/.claude/skills/*/SKILL.md, ~/.claude/**/memory/*.md, and ~/.claude/hooks/ are read and processed to determine redundancy or staleness.
Boundary markers: The workflow relies on user confirmation ([y/n/skip]) for each item, which acts as a manual boundary, but no technical delimiters are specified for processing the file content.
Capability inventory: The skill uses mv for soft-deletion, jq for configuration modification, and standard shell utilities like find and grep.
Sanitization: There is no evidence of sanitization or filtering of the content read from the memory or skill files before it is processed or displayed to the user during the audit.

config-gc