The Agent Skills Directory

[COMMAND_EXECUTION]: The skill's instructions make extensive use of the Bash tool with unquoted placeholders such as {domain_name}, {PR_NUMBER}, and {domain_slug} within shell commands. This pattern presents a command injection risk if these variables are populated with malicious shell metacharacters.
[COMMAND_EXECUTION]: The skill relies on the GH_TOKEN environment variable to authenticate gh (GitHub CLI) commands for querying pull request files, listing issues, and creating new issues in repositories.
[EXTERNAL_DOWNLOADS]: The skill fetches metadata and file SHAs from the vendor's own GitHub repositories, such as fellowship-dev/spec-kit, to identify architectural drift. These references are consistent with the skill's stated purpose for the fellowship-dev environment.
[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface as it reads content from local repository files and incorporates that data into its logic and reporting output.
Ingestion points: File content from docs/code-structure.md, ARCHITECTURE.md, QUALITY_SCORE.md, and various coverage report files (e.g., coverage/index.html).
Boundary markers: Absent; the skill does not use delimiters or instructions to ignore embedded commands when processing ingested file content.
Capability inventory: Bash tool for command execution, Write tool for updating local documentation, and GitHub CLI for remote issue management.
Sanitization: No explicit sanitization or validation of the external file data is implemented before it is processed or interpolated into subsequent commands.

entropy