The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes numerous shell commands, including ls, grep, wc, find, and git, to inspect the filesystem, directory structures, and repository history.
[DYNAMIC_EXECUTION]: The skill executes a local Python script from a path that is dynamically determined during runtime by searching through skill directories: python3 <workflow skill folder>/scripts/check-run.py --all.
[PERSISTENCE]: The skill offers to modify the user's environment by creating or repairing symlinks within the ~/.codex/skills/ directory, which impacts how the Codex tool loads and manages skills across sessions.
[INDIRECT_PROMPT_INJECTION]: The skill ingests and processes text from various local markdown files (such as SKILL.md, MEMORY.md, and README.md) and an external repository without utilizing explicit boundary markers or sanitization, creating a large attack surface for indirect prompt injection.
Ingestion points: Local repository markdown files, plugin instruction files, and the j-skills-evals repository.
Boundary markers: None identified.
Capability inventory: Shell command execution, file system modification (symlinks, .gitignore), and local script execution.
Sanitization: None identified.
[EXTERNAL_DOWNLOADS]: Fetches the j-skills-evals repository from the author's GitHub account (johnviklund/j-skills-evals) to the ~/.agents/ directory.
[CREDENTIALS_UNSAFE]: Accesses and scans for sensitive file paths, including .env files and configuration directories located in the user's home folder (~/.copilot/, ~/.codex/).

checkup