The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes the local command python -m skore_skills review choices to retrieve project status in JSON format.
Evidence: Found in SKILL.md under the 'Sequence' section.
The command belongs to the project's specific tooling and is used for state retrieval.
[DYNAMIC_EXECUTION]: The skill dynamically identifies and loads other skills based on user input and the current project state.
Evidence: SKILL.md specifies loading skills such as sync-ml-reports, export-ml-project, and frame-ml-problem depending on which row the user selects to change.
[INDIRECT_PROMPT_INJECTION]: The skill processes structured data (JSON) from an external command and presents it to the user.
Ingestion points: Output from python -m skore_skills review choices.
Boundary markers: The skill instructions enforce strict handling of the JSON schema (e.g., changeable, framing, read_only keys).
Capability inventory: The skill itself cannot modify the environment directly; it only triggers other skills.
Sanitization: It uses defined schema roles to determine if a value can be changed or if it should be treated as read-only context.

review-ml-choices