The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted input from component files which are then used to drive file modification and shell command execution.
Ingestion points: The skill reads .uxm and .md files located in ./fluxwing/components/ and ./fluxwing/library/ (referenced in Step 3 of SKILL.md).
Boundary markers: Absent. The instructions do not define delimiters or provide warnings for the agent to ignore potential instructions embedded within the component data.
Capability inventory: The skill utilizes Write, Edit, and Bash tools. It performs file updates (Step 5) and executes a shell command for validation (Step 5a).
Sanitization: Absent. Data extracted from the component files is interpolated into the agent's workflow without explicit validation or escaping.
[COMMAND_EXECUTION]: The skill executes a script located outside its own directory structure using the shell.
Evidence: The validation step (Step 5a) executes uv run {SKILL_ROOT}/../uxscii-component-creator/scripts/quick_validate.py. This relies on the existence of a sibling directory containing executable code, which may be outside the user's intended scope for this specific skill.
[METADATA_POISONING]: There is a discrepancy between the author identified in the skill metadata and the official author context.
Evidence: The skill files (SKILL.md and SKILL.json) identify the author as "Trabian", while the provided system context identifies the official author as "jackspace". This inconsistency in metadata can be misleading regarding the origin of the skill.

Fluxwing Component Expander