The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill uses eval "$(~/.claude/skills/econstack/bin/econstack-slug)" to execute shell code generated by a local binary. This pattern allows for arbitrary code execution if the binary's output is manipulated or if the binary itself is compromised.
[COMMAND_EXECUTION]: Multiple local binaries located in ~/.claude/skills/econstack/bin/ are executed silently to handle updates (econstack-update-check), state management (econstack-slug), and data logging (econstack-learnings-read, econstack-learnings-log).
[EXTERNAL_DOWNLOADS]: Users are prompted to download and update a parameter database from an external GitHub repository (https://github.com/charlescoverdale/econstack-data.git) that is not owned by the skill author and is not on the list of trusted vendors.
[DATA_EXFILTRATION]: A "learnings" system captures project insights and logs them using a local binary (econstack-learnings-log). While the skill documentation claims "Nothing transmitted", the use of external binaries for data management cannot be verified through static analysis and represents a potential data handling risk.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection through its processing of untrusted user input.
Ingestion points: The AskUserQuestion tool is used in Step 1 to collect project descriptions and counterfactuals.
Boundary markers: No delimiters or safety warnings are present to separate user data from instructions.
Capability inventory: The skill possesses the Bash tool (for binary execution) and the Write tool (for file creation) across its scripts.
Sanitization: No sanitization or validation of user-provided text is performed before it is used to influence the brainstorming and reporting logic.

longlist