The Agent Skills Directory

Terminal Command Execution: The skill instructs the agent to interact directly with the user's shell environment. It specifically requires executing PowerShell scripts (Invoke-PptxPipeline.ps1) or Bash scripts (invoke-pptx-pipeline.sh) to build PowerPoint decks. While this is part of the core functionality, executing shell commands based on skill instructions carries inherent risks regarding command integrity and terminal environment control.
System Environment Modification: The canonical-deck.md reference defines a protocol for upgrading the PowerShell version on the host machine if it is found to be incompatible. Although the skill mandates seeking explicit user approval before proceeding, the instruction to modify system-level software represents a potential privilege escalation or configuration management concern.
Indirect Prompt Injection Surface: The coaching state protocol documented in coaching-state.md involves capturing the user's initial_request verbatim and storing it in a persistent state file. This data is then used to ground future coaching sessions. As this field ingests untrusted user input without explicit sanitization instructions, it creates a potential surface for indirect prompt injection where a user's initial request could influence subsequent agent behavior across session recoveries.
Ingestion points: references/coaching-state.md defines the initial_request field to store the "Original customer request verbatim."
Boundary markers: No specific delimiters or "ignore instructions" markers are defined for the storage or retrieval of this verbatim field.
Capability inventory: The skill utilizes send_to_terminal and read_file capabilities, which are used to execute logic derived from the coaching state.
Sanitization: The skill does not explicitly define sanitization or validation logic for the captured verbatim request before it is processed in future turns.

dt-coaching-foundation