The Agent Skills Directory

[SAFE]: The skill consists exclusively of markdown instructions and templates; it does not contain executable code, scripts, or binary files that could pose a direct execution risk.- [SAFE]: Tool usage is limited to read-only archaeological research using standard developer utilities like git and gh, and specific Model Context Protocol (MCP) servers for services like Notion, Slack, and Datadog.- [SAFE]: The skill incorporates robust defenses against indirect prompt injection from ingested data sources (such as chat logs or pull request comments) by enforcing strict citation rules and separating evidence from inference. \n
Ingestion points: Tool outputs from git, gh, and various MCPs (Slack, Notion, Linear, Sentry, Datadog, Databricks).\n
Boundary markers: Explicit instructions to "Quote, don't paraphrase" and use a "confidence framework" to label evidence tiers.\n
Capability inventory: Focused on read-only search and retrieval; no capabilities for file modification, environment changes, or arbitrary network egress were found.\n
Sanitization: Relies on specific model instructions to treat external data as historical evidence rather than executable instructions.

why