The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The inmemoria sub-skill frequently executes npx in-memoria, which downloads and runs code from an unverified third-party NPM package (in-memoria by user pi22by7) at runtime.
[EXTERNAL_DOWNLOADS]: The jira sub-skill relies on the mcporter tool and atlassian MCP server, which are downloaded and executed from external sources without being part of an established trusted vendor list.
[PROMPT_INJECTION]: Instructions in miniproject/SKILL.md employ psychological manipulation, stating the agent 'will be rewarded' for high-quality codebase mapping, a technique used to influence AI behavior outside of standard parameters.
[COMMAND_EXECUTION]: Multiple shell scripts (miniproject.sh, wiki, get_ticket_summary.sh) execute complex command chains involving git, awk, sed, and find using inputs derived from the local environment and project files, which may be susceptible to argument injection.
[COMMAND_EXECUTION]: The skill hardcodes an absolute path to the author's local home directory (/home/zenobius/) in miniproject/SKILL.md and miniproject.sh, which can lead to command failures or execution of incorrect scripts on other systems.
[DATA_EXFILTRATION]: The in-memoria component is designed to 'learn' and index the entire codebase (npx in-memoria learn ./src), creating a risk of exposing architectural patterns and sensitive logic to a third-party tool with unverified telemetry behavior.

projectmanagement