The Agent Skills Directory

[PROMPT_INJECTION]: The skill facilitates the processing of user-provided project data (specifications, plan files, and source code) to perform audits and generate visualizations. This creates an indirect prompt injection surface where malicious instructions embedded in the project files could theoretically influence the agent's behavior.
Ingestion points: The system ingests content from the plans/ and specs/ directories, as well as implementation source code and Git logs during the auditing phase described in references/spec-drift-auditor.md.
Boundary markers: The provided Node.js scripts use specific regex patterns to isolate YAML frontmatter from document body; however, the spec-drift-auditor agent does not explicitly define delimiters to separate instructions from the data it inventories.
Capability inventory: The spec-drift-auditor agent has access to Bash, Read, WebFetch, and WebSearch tools to perform repository analysis and external lookups.
Sanitization: There is no evidence of explicit sanitization or instruction filtering on the content of the files processed by the scripts or the auditor agent.

specops