The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill documents a platform architecture designed to ingest and process data and executable logic from external Git repositories, which constitutes an attack surface for indirect prompt injection and potentially untrusted code execution.
Ingestion points: System manifests (.infrahub.yml), GraphQL queries (.gql), object data (YAML), and Python-based automation logic (checks, transforms, generators) are loaded from external repositories as described in deployment-git-integration.md and infrahub-yml-reference.md.
Boundary markers: The skill explicitly mandates safety boundaries, instructing the agent to utilize branch-based development (workflow-branch-for-crud.md) and pre-merge dry-run validation (deployment-gql-dry-run.md) to isolate changes and prevent unintended modification of the production branch.
Capability inventory: The documentation identifies capabilities for executing arbitrary Python logic for data checks and transformations, as well as shell command execution via the infrahubctl CLI tool (rules/connectivity-python-environment.md).
Sanitization: The skill provides a dedicated 'Resources Testing Framework' (testing-resource-framework.md) and recommends YAML linting and live-schema GraphQL validation to ensure the integrity of ingested content.
[SAFE]: The skill incorporates strong security best practices for identity and access management, such as advice on using is_protected flags to lock sensitive attributes and explicit warnings against logging or echoing API tokens in shell environments (rules/connectivity-server-check.md).
[SAFE]: All external URLs and resources identified, including docs.infrahub.app, marketplace.infrahub.app, and github.com/opsmill/*, are official vendor-owned domains and repositories associated with the skill's author, OpsMill.

infrahub-common