The Agent Skills Directory

[COMMAND_EXECUTION]: The skill invokes numerous bash and Node.js helper scripts (e.g., generate-sprint-contract.js, enrich-sprint-contract.sh, codex-companion.sh) located in the plugin's root directory to handle state validation, contract generation, and review processes.
[DYNAMIC_EXECUTION]: It dynamically determines its installation path (HARNESS_PLUGIN_ROOT) using environment variables and directory backtracking, subsequently using this computed path to execute internal scripts.
[PROMPT_INJECTION]: The workflow reads task descriptions and 'Definition of Done' criteria from Plans.md and passes them to a sub-agent via the Agent tool. This creates a surface for indirect prompt injection where untrusted data in the plan could influence sub-agent behavior.
Ingestion points: Plans.md is parsed for task IDs and descriptions in Step 1.
Boundary markers: No specific boundary markers or instruction-ignoring delimiters are specified for the interpolated content.
Capability inventory: The skill utilizes Bash, Edit, Read, and ScheduleWakeup across its operational flow.
Sanitization: No sanitization or validation of the content extracted from Plans.md is performed before processing.
[DATA_EXPOSURE]: The skill manages state, locks, and task contracts within the .claude/state/ directory, which involves reading and writing potentially sensitive project workflow data.

harness-loop