The Agent Skills Directory

[COMMAND_EXECUTION]: Executes inline JavaScript via node -e to parse package.json and extract a list of scripts and dependencies for auditing purposes.
[DATA_EXFILTRATION]: Accesses metadata of sensitive configuration files (.env*) using ls -la. While file contents are not read, exposing the existence and presence of environment files is flagged as metadata exposure.
[PROMPT_INJECTION]: Vulnerable to indirect prompt injection because it ingests untrusted data from CLAUDE.md and uses that context to suggest or apply code changes via the Edit tool.
Ingestion points: Reads CLAUDE.md, package.json, and directory structures during the audit process (SKILL.md).
Boundary markers: Absent; no delimiters or explicit instructions to ignore embedded commands are used when reading external file content.
Capability inventory: Bash, Edit, Read, Glob, Grep, and AskUserQuestion tools are enabled.
Sanitization: Absent; the skill does not validate or sanitize file content before incorporating it into the audit logic or using it to suggest edits.

claudemd-auditor