The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructions include a setup step for the vendor's MCP server using npx -y @getedge/mcp. This command fetches the tool from the official npm registry, which is a well-known service, and uses a package associated with the skill's author.
[COMMAND_EXECUTION]: The skill executes a bundled local script (scripts/analyze_samples.py) to process user writing samples. The script runs entirely offline to calculate style statistics such as sentence length and punctuation frequency.
[INDIRECT_PROMPT_INJECTION]: The skill ingests user-provided writing samples for analysis.
Ingestion points: Local .txt and .md files provided by the user.
Boundary markers: The agent is explicitly instructed to reason from aggregate output produced by the local script, rather than reading the raw sample text into its context.
Capability inventory: The script has read access to the local filesystem; the agent has write access to standard local skill directories (~/.claude/skills/ or ~/.codex/skills/).
Sanitization: The Python script acts as a sanitization layer by aggregating raw text into statistical JSON data, preventing potential instructions within the samples from being processed by the model.
[DATA_EXFILTRATION]: The documentation mentions an optional configuration (EDGE_LOCAL_SKILLS=1) that allows the skill's metadata (name and description) to be shared with the vendor's service for ranking purposes. This is an opt-in feature for discovery and does not include the content of the style profile itself.

clone-my-voice