The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructions and helper scripts (found in SKILL.md, scripts/README.md, and agents/skill.yaml) suggest installing the Sruja CLI using 'curl -fsSL https://sruja.ai/install.sh | bash'. This pattern involves executing remote code directly in the shell. While the domain belongs to the skill author, it is a significant execution vector.
[EXTERNAL_DOWNLOADS]: The skill references and downloads configuration and scripts from the author's official domains and repositories, specifically sruja.ai and github.com/sruja-ai/sruja.
[COMMAND_EXECUTION]: The skill contains shell scripts (scripts/collect-evidence.sh and scripts/validate-refine.sh) that execute the sruja CLI for tasks like 'discover', 'lint', and 'drift', and manage local file creation for results.
[PROMPT_INJECTION]: The skill exhibits risk for indirect prompt injection (Category 8). 1. Ingestion points: The skill reads local repository metadata and code structures through 'sruja discover' and file reads of '.sruja/context.json'. 2. Boundary markers: There are no explicit instructions or delimiters used to prevent the LLM from obeying instructions embedded in the analyzed code. 3. Capability inventory: The skill can execute subprocesses via the 'sruja' CLI and perform file system writes (e.g., repo.sruja). 4. Sanitization: No evidence of sanitization or escaping is present for the data gathered from the analyzed repository.

sruja-architecture