The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from multiple external sources to influence its content generation process.
Ingestion points: Reads data from wiki/business/voice.md, the business 'vault' (learnings and research dossiers), and raw transcript files in the raw/ directory.
Boundary markers: The skill contains explicit instructions to avoid 'generic ghostwriter' output and enforces strict verification rules such as 'Ask before you assert,' 'Biographical claims are receipt-only,' and 'Admission before prose' to ensure grounding.
Capability inventory: The skill has the capability to write new files to wiki/business/content/, update index.md and hub/business.md logs, and execute local audit scripts.
Sanitization: It employs a mandatory scripts/graph-audit check as a filing gate to validate that all durable file changes meet structural and integrity requirements.
[COMMAND_EXECUTION]: The skill instructs the agent to execute a local validation utility.
Evidence: It requires running scripts/graph-audit on the complete content artifact and every touched durable file before the workflow is considered complete. This command is used for internal consistency and integrity verification.

content-engine