The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes the Bash tool to execute shell commands for validation, file statistics, and repository management.
Evidence: Use of wc, awk, and grep within tick validation logic and Done Criteria templates.
Evidence: Automated git operations including git add and git commit as part of the tick execution cycle.
Evidence: The execution.verifier field in the loop YAML allows for the definition and execution of arbitrary shell commands to verify tick success.
[DYNAMIC_EXECUTION]: The skill dynamically generates execution prompts and FlowMind YAML configurations at runtime based on the current state and project content.
Evidence: Implementation of a 'Meta-Prompting' system that builds sub-agent instructions using templates and variable interpolation from files like PRDs and loop state.
Evidence: Automatic generation of adversarial flowmind files (.flow.yaml) between ticks to perform diagnostic checks.
[PERSISTENCE]: The skill implements automated scheduling which allows agent tasks to persist and recur over time.
Evidence: Use of CronCreate and CronDelete via the /ll schedule and /ll stop commands to manage recurring execution intervals.
[INDIRECT_PROMPT_INJECTION]: The skill ingests data from external project files and interpolates it directly into prompts for sub-agents, creating an attack surface for data-driven instruction overrides.
Ingestion points: Reads from .lev/loops/ directory, project-specific PRD files (e.g., prd.json), and source documentation (e.g., dna/graph.yaml).
Boundary markers: Uses Markdown headers for structure but lacks explicit delimiters or instructions to ignore embedded commands within the interpolated variables (e.g., {Read the loop YAML. Read the PRD}).
Capability inventory: The skill possesses Read, Write, Edit, Bash, and Agent capabilities, allowing it to modify the filesystem and execute shell commands based on task outcomes.
Sanitization: No explicit sanitization or validation of the content within the ingested PRDs or source documents is performed before they are included in the prompt context.