The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and synthesize data from untrusted external sources, such as the live web and public GitHub repositories, creating a potential surface for indirect prompt injection.
Ingestion points: External data enters the agent context through search, fetch, and repository browsing tools, as documented in rules/research-workflow.md and references/channel-selection.md.
Boundary markers: The skill employs "expected truths" and a "claim gate" to validate external evidence against user premises.
Capability inventory: The agent utilizes web browsing, file system access, and a local shell for isolated execution checks.
Sanitization: The skill provides explicit instructions to treat all external content as evidence only and strictly forbids following instructions found within retrieved artifacts, as detailed in rules/validation.md and rules/claim-verification.md.
[DYNAMIC_EXECUTION]: The skill includes a process for technical claim verification that involves generating and running scripts in a local shell.
Evidence: Section 4 of rules/claim-verification.md ("Execution Checks") directs the agent to write and execute minimal scripts to settle disputes regarding code behavior or API output.
Safety Constraints: The instructions mandate the use of isolated temporary directories, version pinning, and strictly prohibit the use of credentials, account actions, or unauthorized file system writes.

hyper-research