The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external Remotion source code (TSX/JS files) provided by the user to perform translations. This creates an attack surface where malicious natural language instructions could be embedded in the source code to influence the agent's behavior during the translation or documentation phases.
Ingestion points: The agent reads user-supplied source files in the Remotion project directory as described in the 'Workflow' section of SKILL.md.
Boundary markers: While the skill does not use explicit boundary markers in its prompts, it mandates a linting step using scripts/lint_source.py which filters out non-deterministic code patterns.
Capability inventory: The skill possesses the capability to execute shell commands (ffmpeg, npx, python3) and write to the local filesystem (rendering outputs and summaries).
Sanitization: The linter identifies functional blockers but does not explicitly sanitize for natural language prompt injection content.
[EXTERNAL_DOWNLOADS]: The skill and its generated templates reference external assets and libraries from trusted organizations and well-known services.
Evidence: References to Google Fonts (fonts.googleapis.com) and Cloudflare/UNPKG CDNs (cdnjs.cloudflare.com, unpkg.com) for loading Lottie and typography assets.
Evidence: Links to official vendor resources on GitHub (github.com/heygen-com/hyperframes) for technical interop documentation.
[COMMAND_EXECUTION]: The skill uses local utility scripts and system tools to perform its core functions of video rendering, diffing, and code analysis.
Evidence: Execution of ffmpeg and ffprobe for video processing and frame-accurate SSIM comparisons in scripts/render_diff.sh and scripts/frame_strip.sh.
Evidence: Use of python3 for the linting engine in scripts/lint_source.py.
Evidence: Use of npm and npx for installing dependencies and running the Remotion and HyperFrames CLI tools as documented in the test corpus instructions.
[DYNAMIC_EXECUTION]: The evaluation scripts utilize hardcoded Python blocks within shell scripts to process command output and generate JSON summaries.
Evidence: The use of `python3
<<'PY'heredocs inscripts/render_diff.shandscripts/frame_strip.sh` is a standard practice for data processing and does not involve the execution of untrusted or remote code.

remotion-to-hyperframes