The Agent Skills Directory

[COMMAND_EXECUTION]: The scripts/gather-metadata.py script utilizes subprocess.run() to execute git commands for retrieving current branch, commit, and remote URL information. The implementation is safe as it uses argument lists and execution timeouts, mitigating shell injection risks.
[INDIRECT_PROMPT_INJECTION]: The skill ingests content from external sources such as codebase files and the open web, which could contain malicious instructions designed to influence the agent's behavior during synthesis.
Ingestion points: Data enters the agent's context through file reading in code-analyzer and code-pattern-finder, as well as web content retrieved by the web-searcher agent.
Boundary markers: The system uses Markdown and YAML frontmatter as delimiters; however, it lacks specific instructions to the agent to disregard potential instructions embedded within the ingested research data.
Capability inventory: The skill has the capability to perform file system operations (read/write/move), network fetches, and system metadata retrieval, which could be abused if an injection is successful.
Sanitization: No automated sanitization or filtering of external content is performed before it is processed by the AI models.

researching-codebases