The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes several bundled Python scripts (e.g., prepare_pet_run.py, compose_atlas.py, extract_strip_frames.py) to perform deterministic image processing, such as atlas assembly, frame slicing, and validation. It also executes common shell utilities like jq, cp, and rm to maintain job manifests and manage temporary files within the skill's specific environment at ${CODEX_HOME}.
[INDIRECT_PROMPT_INJECTION]: The skill includes a brand discovery phase that ingests data from the web using search tools. This data is distilled into visual cues and motifs which are then interpolated into prompts for the image generation engine. Although this introduces a surface for indirect injection, the skill mitigates this by instructing the agent to treat external cues only as broad inspiration and explicitly forbidding the verbatim copying of logos, UI text, or marks. (Ingestion: External website data processed by a discovery subagent; Boundary Markers: Explicit instructions in SKILL.md to avoid copying trademarks and to treat inputs as broad cues; Capability Inventory: Local file system access, shell command execution, and script invocation; Sanitization: Formatting of external data into a structured brief with basic whitespace compaction.)
[EXTERNAL_DOWNLOADS]: The skill uses web search capabilities to research brand identities and visual styles for mascot creation. This functionality is documented as a primary feature of the discovery worker to assist in generating context-aware pet sprites.

hatch-pet