The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's primary function is to audit instructions authored by other users, which creates a surface for indirect prompt injection attacks where malicious skills could attempt to influence the auditor's behavior.
Ingestion points: Uses skill-get and skill-file-get to read the bodies, descriptions, and bundled files of other agent skills.
Boundary markers: Includes a dedicated "Untrusted content" section that explicitly instructs the agent to treat all ingested fields as data to be analyzed rather than instructions to follow, and to ignore any imperative framing found in the audited content.
Capability inventory: Has capabilities to emit_report, edit_report, execute-sql, and scout-scratchpad-remember.
Sanitization: Instructs the agent to only quote sanitized snippets in reports and explicitly prohibits reproducing secret values, requiring redaction instead.
[EXTERNAL_DOWNLOADS]: The skill periodically fetches updated best-practice guidelines and specifications from external URLs.
Fetches configuration from Anthropic's official GitHub repository (anthropics/skills).
Retrieves documentation from the agentskills.io specification site.
Downloads guidelines from the official Anthropic Claude documentation.
[COMMAND_EXECUTION]: The skill uses conditional logic to execute SQL queries via execute-sql for usage discovery. This is used to identify "high-leverage" skills by analyzing telemetry data such as tool-call events and version churn.

signals-scout-skills-store