The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill instructions direct the agent to collect high-privilege credentials from the user, including Google Service Account JSON files, HubSpot Private App Tokens, and GoHighLevel API keys.
[DATA_EXFILTRATION]: By default, the skill creates dashboards with a "public, no-login URL," which exposes client data to the internet. The skill documentation explicitly identifies Lead and Communications pages as high-risk surfaces for PII exposure under this default posture.
[COMMAND_EXECUTION]: The skill executes local Python scripts, such as scripts/quality_gates/audit_provenance_lint.py, to perform audits on data folders.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external sources that could contain malicious instructions meant to influence the agent's summary or recommendations.
Ingestion points: Website scraping of client URLs (Phase 0), customer review aggregation (Phase 2), and competitor benchmarking (Phase 2).
Boundary markers: The skill references external checklists (privacy-sanitizer-checklist.md) but does not implement explicit delimiters for the data it processes.
Capability inventory: The agent has network access to third-party APIs, file-writing capabilities to generate dashboard code, and the ability to trigger automated messaging.
Sanitization: Lacks specific input validation or escaping mechanisms for the scraped website content or aggregated reviews before processing.
[PERSISTENCE]: The "Weekly Engagement Automation" feature establishes recurring touchpoints (Email, SMS, Voicemail) that persist beyond the initial dashboard build to maintain client engagement.
[DYNAMIC_EXECUTION]: The skill uses scripts/build_dashboard.py to dynamically generate the client-facing reporting surface based on extracted brand assets and connected data sources.

kai-client-dashboard