The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill explicitly instructs the agent in references/codebase-understanding.md to "Analyze dotenv files and configuration". Environment files typically contain sensitive information such as API keys, database credentials, and other private secrets, leading to high-risk data exposure.
[EXTERNAL_DOWNLOADS]: In references/research-phase.md, the skill uses the repomix --remote <github-repo-url> command to download and consolidate content from external GitHub repositories. Fetching content from unverified remote sources introduces the risk of processing malicious data.
[COMMAND_EXECUTION]: The skill utilizes external command-line tools such as gh (GitHub CLI) and repomix within its research workflow. These tools are used to interact with remote services and process external data, which can be exploited if inputs or the target sources are malicious.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to analyze existing codebases (local and remote) to generate implementation plans, creating a significant surface for indirect prompt injection where malicious instructions embedded in the analyzed code could influence the agent's behavior.
Ingestion points: Data enters the agent's context through codebase scouts defined in references/codebase-understanding.md and remote repository analysis via repomix in references/research-phase.md.
Boundary markers: There are no specified delimiters or instructions to ignore potential commands within the analyzed codebase or generated research reports.
Capability inventory: The skill can execute shell commands (gh, repomix) and has file system write access to create the plans/ directory structure.
Sanitization: The skill does not implement any sanitization, validation, or filtering logic for the content ingested from external or local repositories.

planning