The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses local shell commands to discover the latest versions of its internal constraints and the next steps in its workflow.
Evidence: ls -d ~/.claude/plugins/cache/edwinhu-plugins/workflows/*/lib/references/ds-common-constraints.md 2>/dev/null | sort -V | tail -1 in SKILL.md.
Evidence: ls -d ~/.claude/plugins/cache/edwinhu-plugins/workflows/*/lib/skills/ds-plan/SKILL.md 2>/dev/null | sort -V | tail -1 in SKILL.md.
These operations are limited to the vendor's specific directory and are used for internal workflow management.
[PROMPT_INJECTION]: The skill presents a surface for indirect prompt injection as it processes untrusted user input to generate a specification file that influences later automation phases.
Ingestion points: User answers provided via AskUserQuestion commands.
Boundary markers: The workflow requires a mandatory user confirmation and a separate review by a ds-spec-reviewer subagent before proceeding to data exploration.
Capability inventory: The skill can execute directory discovery commands (ls) and write/read files within the .claude/ directory.
Sanitization: No explicit content filtering or escaping is performed on user inputs before they are written to the specification file.

ds