The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes a local Python engine (backlink_gap.py) located within the plugin root directory to perform the mathematical gap analysis and scoring. This is a standard execution pattern for skills providing complex data processing capabilities.
[DATA_EXPOSURE]: The skill reads and writes to specific local directories, including the user's home directory (~/.claude-marketing/brands/) for brand profiles and a plugin data directory (${CLAUDE_PLUGIN_DATA}) for results. This access is scoped to the skill's own operational metadata and project outputs.
[INDIRECT_PROMPT_INJECTION]: The skill ingests external CSV data exported from third-party SEO tools (Ahrefs, Semrush, etc.). While this presents a theoretical surface for indirect prompt injection if the CSV content contains malicious instructions, the data is processed by a specialized Python script rather than being directly interpreted as instructions by the LLM.
Ingestion points: Reads backlink CSV exports and brand profile JSON files (SKILL.md Step 2, 3, 4).
Boundary markers: The process uses structured file paths and CLI flags to separate input data from the execution environment.
Capability inventory: Local script execution (python), file system read/write, and MCP tool invocation for web fetching.
Sanitization: The skill relies on the logic within backlink_gap.py for data processing and scorecard validation (Step 6).

backlink-gap