The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes multiple bash scripts from the local path ~/.claude/skills/econstack/bin/ during initialization and operation, such as econstack-update-check, econstack-slug, and econstack-learnings-read.- [REMOTE_CODE_EXECUTION]: The initialization process includes eval "$(~/.claude/skills/econstack/bin/econstack-slug)", which executes the output of a script directly as shell commands. This pattern can be exploited if the script output is influenced by external data.- [EXTERNAL_DOWNLOADS]: The skill instructions direct users to download and update data from a third-party GitHub repository: https://github.com/charlescoverdale/econstack-data.git.- [DATA_EXFILTRATION]: The skill processes potentially sensitive documents and possesses both Bash and Skill tools. While it claims local-only storage for its learning system, the technical capability to transmit data to external repositories exists.- [PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface when processing user-supplied documents.
Ingestion points: Reads markdown files provided via the [file or directory] argument (SKILL.md, Step 1).
Boundary markers: No specific delimiters or safety instructions are used when ingesting the document content.
Capability inventory: Access to Bash and Write tools allows for significant system impact.
Sanitization: No evidence of sanitization or content validation before processing the untrusted document data.

econ-audit