The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The Docker Compose configuration provided for self-hosting the Firecrawl backend includes hardcoded default credentials for its internal database and services.
Evidence: In references/self-host-firecrawl/docker-compose.yml, POSTGRES_PASSWORD is set to postgres and BULL_AUTH_KEY defaults to CHANGEME. While the documentation warns against public exposure, these are insecure defaults.
[PROMPT_INJECTION]: The skill ingests data from arbitrary third-party websites during the price research and coupon hunting phases, creating an indirect prompt injection surface.
Ingestion points: Web content from various storefronts and coupon aggregators is fetched via Firecrawl or built-in tools as defined in the Step 2 and Step 3 instructions of SKILL.md.
Boundary markers: The instructions tell the agent to "extract just these fields" and "never dump whole pages," providing a logical boundary for processing untrusted data.
Capability inventory: The agent can execute shell commands through scripts/firecrawl.sh and perform network requests to the scraping backend.
Sanitization: The skill requires the agent to transform raw page content into structured fields or summaries, which mitigates direct pass-through of malicious instructions.
[COMMAND_EXECUTION]: The skill relies on a local shell script to interact with its scraping backend.
Evidence: scripts/firecrawl.sh is used to execute curl commands for performing searches and scraping operations.
[EXTERNAL_DOWNLOADS]: The documentation provides instructions to download and run the Firecrawl infrastructure using Docker.
Evidence: references/self-host-firecrawl/docker-compose.yml pulls official images for Firecrawl, Redis, and RabbitMQ from well-known registries. These images are correctly pinned to specific cryptographic digests to ensure integrity.

shop-scout