The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes a shell command via bun run to process design data into an external gallery backend.
Evidence: bun run ~/.agents/skills/laniameda-gallery-ingest/scripts/ingest.ts '<payload-json>' in SKILL.md.
[REMOTE_CODE_EXECUTION]: The skill relies on and executes a TypeScript script sourced from an external repository.
Evidence: Usage of ingest.ts located in a separately installed skill directory.
[EXTERNAL_DOWNLOADS]: The skill recommends installing a dependency from a specific GitHub repository.
Evidence: bunx skills add https://github.com/laniamedaHQ/laniameda-gallery/tree/main/skills/laniameda-gallery-ingest in SKILL.md.
[DATA_EXFILTRATION]: Transfers design metadata, URLs, and image paths to the laniameda.gallery service.
Evidence: The ingest workflow configured to send data to the CONVEX_URL backend.
[CREDENTIALS_UNSAFE]: Contains a hardcoded user identifier used for identifying the gallery owner.
Evidence: KB_OWNER_USER_ID=278674008 in SKILL.md.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection by processing untrusted files in the references directory.
Ingestion points: Reads all image and markdown files in /mnt/skills/user/design-inspiration/references/ (SKILL.md).
Boundary markers: Absent. There are no instructions for the agent to distinguish between design constraints and malicious instructions within these files.
Capability inventory: Shell command execution (bun run), file listing (ls), and file reading (view).
Sanitization: Absent. The agent is directed to "internalize" the content and apply it directly to code generation tasks.

design-inspiration