The Agent Skills Directory

[SAFE]: The skill implements strong security boundaries by explicitly forbidding the agent from publishing, scheduling, or sending content autonomously. Step 5 and the 'Boundaries' section in SKILL.md and CARD.md mandate human approval for all public-facing assets.
[SAFE]: Role isolation is enforced to prevent context bleed and maintain specialized focus. references/role-contracts.md provides distinct instructions ('contracts') for six different roles, ensuring that each agent handles only a specific part of the pipeline.
[INDIRECT_PROMPT_INJECTION]: The skill contains an attack surface for indirect prompt injection as it is designed to ingest external data from URLs and founder recordings.
Ingestion points: The Signal scout and Researcher roles in references/role-contracts.md are instructed to read from external URLs. The Founder intake process in references/founder-content-loop.md ingests interview transcripts and voice notes.
Boundary markers: The skill uses a structured 'Campaign Record' (references/campaign-record.md) to delimit data and requires the Researcher to strictly separate verified claims from inferences and unproven statements.
Capability inventory: The skill is limited to reading data and writing markdown records/drafts within the local .agents/newsroom/ directory. It does not utilize shell commands, network exfiltration tools, or dynamic execution environments.
Sanitization: Verification is built into the workflow; the Researcher must provide a URL for every consequential claim and the Editor role performs a 'Claim integrity' check to ensure no unsupported content is added. Human review is the final sanitization gate.

suede-newsroom