The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill possesses an indirect prompt injection attack surface as it is designed to ingest and process untrusted data from the public web via the WebSearch and WebFetch tools.
Ingestion points: The agent fetches external content from URLs discovered during the research process, as documented in SKILL.md and references/search-playbook.md.
Boundary markers: The skill implements a robust verification framework defined in CONTEXT.md and references/verification-policy.md. It requires the agent to resolve a subject's identity using a fingerprint of professional tokens before substantive research begins.
Capability inventory: The skill has the capability to search the web, fetch external pages, and write research reports to a project-local directory (research/people/).
Sanitization: The instructions emphasize source-opening discipline, distinguishing between self-reported and independent sources, and maintaining citation integrity across all output sections to mitigate the impact of adversarial content.
[COMMAND_EXECUTION]: The package includes a Python utility script, scripts/validate.py, which is used to verify the skill's structural integrity and file hashes.
The script performs local file reads, regex checks, and SHA-256 hash verifications to ensure the skill matches its recorded observable behavior.
It is a static analysis tool that does not perform network operations or execute remote code.
[EXTERNAL_DOWNLOADS]: The skill performs external network operations to retrieve public professional records and primary sources from the internet.
These downloads are the primary mechanism for the skill's intended purpose and are restricted to the public professional record, with explicit prohibitions against accessing private or sensitive personal data.

verified-person-research