The Agent Skills Directory

[SAFE]: The skill instructions emphasize a 'read-only' posture, explicitly forbidding file mutation, package installation, and stateful git actions without user approval. This design significantly limits the potential for unauthorized changes or persistent malicious activity.
[COMMAND_EXECUTION]: The skill utilizes standard inspection commands (such as rg for searching) and read-only repository access. These capabilities are restricted to information gathering and are consistent with the skill's intended research purpose.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection due to its core function of ingesting and analyzing external web content and internal codebase files.
Ingestion points: External research sources (web pages, vendor docs, standards) and internal repository artifacts (source code, logs, configs, git history) as outlined in the research workflows.
Boundary markers: The skill does not define specific markdown delimiters for external content but mandates that the agent lead with citations, distinguish facts from interpretation, and maintain an evidence ledger to track sources.
Capability inventory: Access to repository search tools (rg), file reading, external web search via MCP, and the ability to dispatch parallel agents for specific research lanes.
Sanitization: No automated sanitization is implemented; however, the requirement to cross-check important claims across independent sources and document 'Unknowns' in the evidence ledger serves as a procedural mitigation against malicious data.

research