The Agent Skills Directory

[COMMAND_EXECUTION]: The skill relies on executing various shell commands to inspect the local environment (e.g., lsof, ps), manage Docker containers (docker build, docker compose up, docker inspect), and perform stack-specific build operations across multiple languages.
[EXTERNAL_DOWNLOADS]: The skill is configured to fetch official Docker base images from well-known registries including Docker Hub (Node.js, Python), Microsoft Container Registry (.NET), and Google Container Registry (distroless), as well as download application dependencies through standard package managers like npm and pip.
[PROMPT_INJECTION]: An indirect prompt injection surface is present because the skill's core functionality involves reading and analyzing arbitrary files from project repositories (source code, manifests, and configuration files). These files could potentially contain malicious instructions intended to manipulate the agent's logic during the investigation phase.
Ingestion points: Project repository files, including manifests (package.json, pyproject.toml, pom.xml, etc.) and application source code.
Capability inventory: Extensive execution of shell commands (docker, curl, lsof, ps) and build tools (npm, pip, dotnet, go, mvn, gradle) throughout the skill and its references.
Boundary markers: The skill encourages structuring extracted metadata into a 'contract' summary, providing a layer of separation between data ingestion and artifact generation.
Sanitization: Analysis focuses on structural extraction of metadata (e.g., ports, environment variable keys, and dependency names) rather than executing logic directly from the ingested content.

containerizar-aplicacao