The Agent Skills Directory

[PROMPT_INJECTION]: The skill is designed to ingest and 'inspect' third-party repositories containing instruction files (SKILL.md). This creates a surface for Indirect Prompt Injection, where instructions within a source repository could attempt to override agent behavior or exfiltrate data during the porting process.
Ingestion points: The scripts/list_claude_skills.py script identifies external skill files, which the agent is then instructed to read and inspect.
Boundary markers: No specific delimiters or safety instructions are provided to the agent to treat external SKILL.md content as untrusted data.
Capability inventory: The skill possesses file read/write capabilities via the provided Python scripts.
Sanitization: There is no evidence of sanitization or filtering of the content read from external repositories.
[COMMAND_EXECUTION]: The skill relies on the execution of two local Python scripts (scripts/list_claude_skills.py and scripts/import_skill.py). The import_skill.py script performs high-risk file system operations, including shutil.rmtree (recursive directory deletion) when the --force flag is used, and shutil.copytree. While functional for porting skills, these capabilities could be misused if the agent is directed toward sensitive system paths.
[DATA_EXFILTRATION]: Multiple files, including SKILL.md and scripts/import_skill.py, contain hardcoded absolute paths (e.g., C:\Users\nd851\.codex\skills). This constitutes information disclosure by revealing a specific local username ('nd851') associated with the environment for which the skill was developed.

import-claude-skills