The Agent Skills Directory

[COMMAND_EXECUTION]: The skill modifies the repository's .git/hooks/ directory by installing commit-msg and pre-push hooks. These are persistence mechanisms that execute code automatically on developer machines during git operations.
[REMOTE_CODE_EXECUTION]: The skill generates and writes GitHub Action workflows to .github/workflows/. These scripts execute code in remote CI/CD environments (GitHub runners) and can be configured to trigger on various repository events.
[COMMAND_EXECUTION]: The skill performs extensive file system operations, including creating provider-specific configuration directories (e.g., .claude/skills/, .github/agents/, .cursor/rules/, .windsurf/rules/) and writing multiple reformatted skill files and an audit report (ENVOY_REPORT.md).
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection as it ingests untrusted data from local files and propagates it across different agent runtimes.
Ingestion points: Reads all SKILL.md files within the members/ directory as source material for translation.
Boundary markers: None identified; the skill performs direct content extraction and reformatting (e.g., removing YAML frontmatter) without adding delimiters or warnings to the output files.
Capability inventory: The skill possesses significant file-write capabilities, including access to restricted directories like .git/hooks and .github/workflows.
Sanitization: None identified; the skill assumes all content in the members/ directory is safe and adheres to expected formats.

the-envoy