The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill researches target company websites and incorporates the results into generated marketing content, creating a surface for indirect instructions to influence agent behavior.\n
Ingestion points: Website content is fetched using WebFetch in Step 2 (Sub-step A) for research purposes.\n
Boundary markers: There are no explicit delimiters or 'ignore embedded instructions' warnings provided when website content is passed into variables like websiteContent or email body paragraphs.\n
Capability inventory: The skill has the capability to write to the vendor's CRM and notification systems using environment-stored tokens and performs network operations via curl.\n
Sanitization: No sanitization or validation of the fetched website content is performed before it is interpolated into prompts or API payloads.\n- [COMMAND_EXECUTION]: The skill utilizes shell commands to interact with vendor services and manage local credentials.\n
Evidence: Employs curl for GraphQL mutations to crm.psquared.dev and REST calls to notifications.psquared.dev. It also uses grep and cut to retrieve API keys from a local .env file.\n- [EXTERNAL_DOWNLOADS]: The skill communicates with several external endpoints to fetch company data and push generated demos.\n
Evidence: Accesses crm.psquared.dev, app.psquared.dev, and notifications.psquared.dev. These domains belong to the author's infrastructure (psquared-development).\n- [REMOTE_CODE_EXECUTION]: The skill instructions include piping remote data to a local Python interpreter for processing.\n
Evidence: Step 5 describes piping JSON output from notifications.psquared.dev to python3 -c for parsing. While this pattern is often a risk, the specific Python script used (import json,sys; print(json.load(sys.stdin)['data']['total'])) is a static, benign utility for data extraction from the vendor's own API.

inboxmate-demo-with-outreach