The Agent Skills Directory

[DATA_EXFILTRATION]: The skill accesses a sensitive configuration path at ~/.config/RevorSkill/.env to manage authentication credentials. Although the instructions include safety measures such as key masking and avoiding chat disclosure, the interaction with a dotfile in the user's home directory is a sensitive operation used in conjunction with network calls to the revor.ai domain.
[INDIRECT_PROMPT_INJECTION]: The skill ingests and processes untrusted recipient data and outreach content, establishing an attack surface for indirect prompt injection. 1. Ingestion points: User-provided or searched contact details (profile_url, address, phone) and message drafts in SKILL.md and references/dispatch/dispatch.md. 2. Boundary markers: The instructions lack specific structural delimiters or warnings to prevent the agent from executing instructions that may be embedded within the ingested contact data. 3. Capability inventory: The skill performs network operations via MCP tools (revor_send_outreach, revor_like_linkedin_post) and REST API endpoints described in SKILL.md and references/dispatch/dispatch.md. 4. Sanitization: There is no documentation of explicit sanitization or escaping of external content before it is interpolated into outgoing API payloads.

revor-outreach