The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill mentions the installation of the 'allium' plugin using the command /plugin install allium@juxt in SKILL.md. This refers to a plugin provided by JUXT, a well-known and reputable technology consultancy. The download is part of the intended functionality for behavioral specification tracking.
[PROMPT_INJECTION]: The skill implements a workflow that processes 'Epics' authored by users, which are then decomposed into tasks for agents. This creates a surface for indirect prompt injection.
Ingestion points: User-authored epic objectives and constraints enter the agent context via the Team Leader (references/team-leader.md).
Boundary markers: No explicit delimiter-based isolation is described for the user-provided epic content.
Capability inventory: The agent roles have the capability to execute shell commands for git operations, CI/CD tasks (mise), and project-specific tooling (bees).
Sanitization: The workflow relies on a 'Validator' role (references/validator.md) to run technical checks and CI suites, providing a safety layer against malicious code generation, though it does not explicitly sanitize the natural language inputs.
[COMMAND_EXECUTION]: The skill makes extensive use of CLI tools including git, gh, mise, and language-specific test runners. It includes specific defensive instructions to prevent security regressions, such as running gitleaks on all commits and utilizing the 1Password CLI (op) in a way that avoids revealing sensitive values in output logs.

agent-loop