The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill accesses a local configuration file at ~/.claude/skills/epiphan-house-rules-skill/reference/identifiers.md which is explicitly stated to contain sensitive information including owner IDs, portal details, emails, and seats.- [COMMAND_EXECUTION]: The skill uses the execute_sql tool to perform database queries on the HubSpot CRM, providing the agent with broad read access to structured data.- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from CRM notes and call summaries to generate customer-facing email drafts, creating a potential attack surface for indirect prompt injection.
Ingestion points: Data is ingested via hubspot_get_deal (notes) and clari_get_call_summary (transcripts and sentiment).
Boundary markers: The skill does not define specific delimiters or instructions to treat external CRM data as literal content, which could allow instructions hidden in notes to influence agent behavior.
Capability inventory: The agent has the ability to query the database (execute_sql), update CRM records (hubspot_create_contact_note), and create email drafts (create_draft).
Sanitization: While the skill employs check_my_copy for brand voice review, there is no technical sanitization to filter out malicious prompt injection payloads from the ingested data.

dead-deal-recovery-skill