The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: Fetches a package catalog and individual skill files from the author's GitHub repository at github.com/Mathews-Tom/armory.
[COMMAND_EXECUTION]: Orchestrates various shell commands including curl, git, rsync, rm, awk, and grep to manage local files and perform network operations.
[COMMAND_EXECUTION]: Several procedures in the cookbook/ directory (notably remove.md, use.md, and sync.md) interpolate user-provided package names directly into shell commands. For example, rm -rf ~/.claude/skills/<name>/ and mkdir -p ~/.claude/skills/<name> are used without explicit sanitization of the <name> variable, which could lead to command injection or unauthorized file deletion if the input is maliciously crafted.
[INDIRECT_PROMPT_INJECTION]: The skill processes an external manifest.yaml file to determine package metadata and installation paths. Malicious content in this remote manifest could influence the agent's display or lead it to perform unintended actions.
Ingestion points: Remote manifest.yaml fetched from GitHub (SKILL.md).
Boundary markers: None used to delimit or ignore instructions within the manifest content.
Capability inventory: File system read/write, directory removal (rm -rf), network downloads, and file permission modification.
Sanitization: Uses standard text processing tools like awk and grep for metadata extraction, but lacks robust validation of remote data fields.
[PRIVILEGE_ESCALATION]: The skill documentation specifies that utility type packages are installed using chmod +x, allowing for the execution of downloaded scripts or binaries in the user's environment.

skill-library