The Agent Skills Directory

[PROMPT_INJECTION]: No evidence of instructions attempting to bypass safety filters or override agent behavior. The language is focused on professional video editing techniques.
[DATA_EXPOSURE_AND_EXFILTRATION]: No sensitive file access or network operations detected. File operations are restricted to project-specific paths and standard interchange formats like FCPXML and EDL.
[OBFUSCATION]: The content is fully transparent. No Base64, zero-width characters, or homoglyph attacks were found.
[UNVERIFIABLE_DEPENDENCIES_AND_RCE]: The skill references a local CLI tool ('cinefoundry'), which is consistent with the stated purpose and the author's context. No remote script downloads or third-party package installations are present.
[INDIRECT_PROMPT_INJECTION]: The skill manages client-provided notes and script revisions through the revise command.
Ingestion points: Untrusted data enters via --text and --note arguments in the cinefoundry revise command.
Boundary markers: Not explicitly defined in the prompt text, but data is handled within a structured CLI argument format.
Capability inventory: The skill triggers file exports, revision logging, and automated re-processing of clips.
Sanitization: Mandatory compliance checks ('ad-compliance-review') and independent reviewer paths are integrated into the workflow to validate revised text before it is used.
[DYNAMIC_CONTEXT_INJECTION]: No use of shell-execution placeholders or silent background commands in the skill documentation.

editing-and-versioning