The Agent Skills Directory

[COMMAND_EXECUTION]: The helper script scripts/collect-vision-diff.mjs executes git commands using Node.js spawnSync. While parameters are primarily derived from the repository state, user-supplied arguments like --base and --target are passed to the script, which in turn are passed to git.
[PROMPT_INJECTION]: The skill processes potentially untrusted data from git diffs and repository files to influence modifications to the critical VISION.md file, creating an indirect prompt injection surface.
Ingestion points: The script scripts/collect-vision-diff.mjs reads file contents and git diff output to identify candidate lines for the project vision.
Boundary markers: The script structures output into TSV and Markdown files, but lacks explicit instructions or markers to prevent the agent from obeying malicious instructions embedded within the processed text.
Capability inventory: The agent is explicitly instructed to patch the VISION.md file and route instructions to other project components (skills, research docs, etc.) based on its analysis of this data.
Sanitization: Although candidate lines are truncated to 500 characters, no formal sanitization or escaping of the extracted text is performed before it is presented to the agent.

sync-vision