The Agent Skills Directory

[PROMPT_INJECTION]: The instructions in SKILL.md focus on structural retrieval and model-neutral pattern abstraction. There are no attempts to override safety filters, disregard prior instructions, or enable unauthorized administrative modes.
[DATA_EXPOSURE_AND_EXFILTRATION]: The skill contains no hardcoded credentials or sensitive system paths. UUIDs and SHA-256 digests used in the manifest and case files are for audit-only purposes and are accompanied by explicit instructions never to copy them into downstream agent output.
[OBFUSCATION]: All content is in plain text or standard JSON format. Analysis of the manifest and case files confirmed that no Base64, zero-width characters, or homoglyph-based evasion techniques are present.
[UNVERIFIABLE_DEPENDENCIES_AND_REMOTE_CODE_EXECUTION]: The skill does not perform any external downloads or package installations. It operates entirely as a static reference library within the agent's provided context.
[INDIRECT_PROMPT_INJECTION]: The skill possesses a minimal attack surface as it is a retrieval-only layer with no execution capabilities. It implements structured guidance through the guidance-package-schema.md to maintain data integrity.
Ingestion points: User-provided abstract scene requests or repair requests processed by SKILL.md.
Boundary markers: The guidance-package-schema.md enforces a strict YAML structure, and SKILL.md instructs the agent to omit any case suggestions that conflict with user-locked facts.
Capability inventory: None; the skill has no access to file-writing tools, shell execution, or network communication.
Sanitization: The skill mandates the removal of all source identifiers, model-specific syntax, and unfiltered reference blocks before returning results.
[DYNAMIC_EXECUTION]: No dynamic code generation, runtime compilation, or unsafe deserialization patterns were detected. The skill consists exclusively of static documentation and reference data.

cinematic-scene-case-library