The Agent Skills Directory

[METADATA_POISONING]: The example project configuration files specify highly inflated or non-existent version numbers for major dependencies.
examples/nextjs/package.json: Lists next@^16.3.6 (stable is 15.x) and typescript@^7.0.2 (stable is 5.x).
examples/express/package.json: Lists jest@^30.5.2 (stable is 29.x).
examples/fastapi/requirements.txt: Lists fastapi>=0.141.1 (stable is 0.115.x) and pytest>=9.1.1 (stable is 8.3.x).
[INDIRECT_PROMPT_INJECTION]: The skill provides logic for ingesting untrusted data from Docker Hub webhooks, which creates a potential surface for indirect injection attacks.
Ingestion points: Handlers located in examples/nextjs/app/webhooks/docker-hub/[token]/route.ts, examples/express/src/index.js, and examples/fastapi/main.py accept external JSON payloads.
Boundary markers: While the code performs structural validation (e.g., parsePush), it does not implement delimiters or warnings to prevent the agent from processing instructions embedded within the payload strings (like the tag field).
Capability inventory: The skill is capable of logging payload content and performing network lookups against Docker Hub APIs.
Sanitization: Includes basic type validation and field existence checks for the tag and repo_name fields.

docker-hub-webhooks