The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests user-supplied project briefs and plans to generate SOUL.md files, which serve as the personality and instruction set (system prompts) for specialized sub-agents. Maliciously crafted content in the project brief could potentially influence the behavior of these sub-agents.
Ingestion points: plan.json (processed by bootstrap_pipeline.py) and the interactive discovery process that populates brief.md.
Boundary markers: Templates use structured headers, but do not explicitly include instructions for sub-agents to ignore potentially malicious embedded commands within the ingested text.
Capability inventory: Sub-agents are configured with terminal, file, and kanban toolsets, allowing them to execute shell commands and modify the filesystem within the project workspace.
Sanitization: The skill performs regex validation on profile names and slugs to ensure they conform to platform requirements. It utilizes json.dumps for serializing tool and skill lists to prevent injection into configuration files.
[DYNAMIC_EXECUTION]: The skill employs a bootstrap script (scripts/bootstrap_pipeline.py) that dynamically generates an executable shell script (setup.sh) by interpolating project metadata into templates located in the assets/ directory.
Logic: The script uses string replacement to fill templates with values such as project titles, tenants, and roles derived from the input plan.
Context: The generated script is intended to be run to automate the creation of project workspaces and the configuration of agent profiles.
[COMMAND_EXECUTION]: The monitoring utility (scripts/monitor.py) executes platform-native CLI tools to track the state of the production pipeline.
Evidence: Uses subprocess.run to invoke the hermes CLI for listing and showing task details. The calls are implemented using argument lists rather than shell strings, following safe execution patterns.

kanban-video-orchestrator