The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes untrusted user-supplied text, which presents a surface for indirect prompt injection.
Ingestion points: Untrusted content is ingested via conversational input or by reading file paths as specified in the Invocation Modes section of SKILL.md.
Boundary markers: Absent. The instructions do not employ delimiters or specific directives to ignore embedded instructions within the user text.
Capability inventory: The skill prompt in SKILL.md does not have access to tools for network exfiltration, file system modifications, or arbitrary shell execution.
Sanitization: Absent. The skill does not describe methods for escaping or filtering the input text.
[COMMAND_EXECUTION]: The repository includes scripts/validate-package.py for developer maintenance.
The script reads local files (SKILL.md, README.md, and .claude-plugin/plugin.json) to verify metadata consistency.
The operations are limited to string matching and JSON parsing using standard Python libraries and do not execute arbitrary shell commands.
[EXTERNAL_DOWNLOADS]: Installation instructions in README.md suggest using the skills CLI tool via npx to fetch the skill from skills.sh.
The README.md also references the Claude Code plugin marketplace for installation. These references facilitate skill distribution and setup.

humanizer