The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze untrusted external data, such as public policy documents and social media trends, creating an attack surface for indirect prompt injection.
Ingestion points: Protocols in protocols/public-issue-protocol.md and references/source-and-evidence-rules.md explicitly instruct the agent to retrieve and analyze external official texts and third-party reports.
Boundary markers: The skill employs a Public Evidence Ladder (references/public-evidence-ladder.md) and Evidence Boundary Summary (templates/evidence-boundary-summary.md) to logically categorize data reliability.
Capability inventory: The skill leverages the Agent Skills tool to interact with the environment and generate analytical drafts and action plans.
Sanitization: Instructions mandate down-grading conclusions and using question lists when sources are unverified, acting as a procedural safeguard.
[SAFE]: No malicious code, exfiltration patterns, or obfuscation techniques were found. The skill follows best practices for evaluating public evidence and contains evaluation tests in evals/crossframe-public-smoke-tests.md designed to prevent common misuses, such as treating social media sentiment as verified fact.

crossframe-public