The Agent Skills Directory

[DYNAMIC_CONTEXT_INJECTION]: The skill utilizes the dynamic context pattern (!command) to execute shell operations at load time. These commands (echo, cat, ls, head, tr) are used to retrieve project metadata such as the current stage and sprint status from the production/ directory. The execution is limited to local read-only operations and does not include user-controllable arguments or network activity.
[INDIRECT_PROMPT_INJECTION]: The skill processes potentially untrusted content from the workspace to generate its recommendations.
Ingestion points: The skill reads several workspace files including .claude/docs/workflow-catalog.yaml, the name field from .claude/skills/*/SKILL.md, production/stage.txt, production/session-state/active.md, and production/sprint-status.yaml.
Boundary markers: There are no explicit boundary markers or instructions to the model to ignore embedded commands within the files being processed.
Capability inventory: The skill is configured with Read, Glob, and Grep capabilities. It does not possess file-writing or network-access capabilities.
Sanitization: No sanitization or validation is applied to the data retrieved from the workspace before it is interpolated into the reasoning prompt.

help