The Agent Skills Directory

[COMMAND_EXECUTION]: The skill makes extensive use of high-privilege command-line interfaces to manage cloud environments and source code repositories.
Evidence: Frequent use of aws, kubectl, gcloud, az, gh, and the vendor's own planton CLI throughout the instruction set.
Evidence: Capability to execute arbitrary user-defined commands via planton service env run as described in references/service.local-env-vars.md.
Context: These operations are fundamental to the skill's stated purpose as a DevOps tool. The instructions implement a strict protocol requiring explicit user consent for any state-changing mutations.
[INDIRECT_PROMPT_INJECTION]: The skill regularly processes data from external systems that could contain malicious instructions designed to influence the agent.
Ingestion points: Build logs retrieved via get_service_pipeline_logs, GitHub Actions logs from get_external_ci_run_logs, and git metadata (commit messages, PR descriptions) read during activity briefings in references/craft.organization-activity.md.
Boundary markers: The SKILL.md and references/craft.organization-activity.md files explicitly instruct the agent to treat logs and commit content as data and state that instructions found there cannot authorize commands or change the agent's scope.
Capability inventory: The agent has broad capabilities to modify cloud infrastructure, manage secrets, and open pull requests using platform tools.
Sanitization: The skill mandates sanitization of data when drafting GitHub issues in references/craft.filing-platform-gaps.md.
[EXTERNAL_DOWNLOADS]: The skill involves cloning external repositories and referencing remote configuration files.
Evidence: Instructions to clone the plantonhq/planton repository to inspect infrastructure-as-code modules in references/infra.deployment-model.md.
Context: These references point to the official resources of the skill author (plantonhq) on well-known platforms like GitHub, falling within the trusted vendor context.

planton