The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill generates new Agent Skills (Markdown files with logic) and shell scripts based on user input and session history. It then registers these new capabilities into the agent's runtime environment using the skill_manage tool. Although this is the primary intended functionality, the assembly and registration of executable instructions at runtime is a known security vector.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted session history and user answers to automatically generate instructions for new skills, creating a potential path for malicious instructions to influence the generated output.
Ingestion points: skills/observer/SKILL.md (scanning session transcripts via session_search) and skills/interviewer/SKILL.md (interactive user responses).
Boundary markers: Absent; there are no explicit instructions or delimiters provided to the agent to prevent it from obeying commands that might be embedded within the session data being analyzed.
Capability inventory: The skill utilizes write_file for local filesystem access and skill_manage for registering new skills within the agent's ecosystem.
Sanitization: Partial; while shell_quote is mentioned for the generation of kanban setup scripts, there is no evidence of sanitization or escaping for the natural language instructions interpolated into the generated sub-skill templates.

workflow-architect