The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface as it ingests untrusted external data during the research phase.
Ingestion points: The skill fetches data from research/web and research/deep endpoints (web intelligence and free-text research) in references/research-context.md.
Boundary markers: There are no explicit delimiters or instructions to ignore embedded commands within the research data processing instructions.
Capability inventory: The skill possesses the capability to write to the local filesystem (pack-manifest.md, output folders) and perform network operations via content creation APIs (POST /api/content/create).
Sanitization: No specific sanitization or escaping of external research content is defined before interpolation into the generation workflow.
[SAFE]: The skill implements several safety and quality control mechanisms:
Spend Discipline: Mandatory cost estimation (POST /content/estimate) and user confirmation before executing batch generations.
Quality Assurance: The 'Analyze Gate' (POST /api/content/analyze) ensures visual assets are inspected for artifacts and brand alignment before being presented as complete.
Credential Safety: Instructions in references/account-resolver.md correctly advise using environment variables (RECOUP_API_KEY) and warn against extracting OAuth credentials.
User Oversight: The skill is instructed to surface borderline analysis results to the user rather than silently passing them, and it explicitly stops at asset generation without automatic posting/scheduling.

recoup-content-asset-pack