The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes data from the conversation context (such as file paths, PR numbers, and comment bodies) which could potentially contain malicious instructions.
Ingestion points: User-supplied file paths, PR/Issue numbers, and comment text retrieved from conversation history as specified in SKILL.md.
Boundary markers: No explicit delimiters for user text are used, though the skill performs extension-based validation for files and regex validation for numeric identifiers.
Capability inventory: The skill reads local files, executes shell commands (gh, curl, jq, git), and performs authenticated network uploads to GitHub.
Sanitization: scripts/upload.sh uses bash arrays for all external command executions to prevent argument and shell injection and validates file types against a supported whitelist.
[EXTERNAL_DOWNLOADS]: The skill performs network operations to upload visual assets to GitHub.
Evidence: scripts/upload.sh uses curl to upload assets to uploads.github.com and gh api to fetch repository metadata.
Note: These operations target a well-known service (GitHub) and are necessary for the skill's primary function.
[COMMAND_EXECUTION]: The skill executes local shell commands and CLI tools to perform its automation tasks.
Evidence: scripts/upload.sh invokes gh, curl, jq, and git via subprocess calls.
The implementation uses safe practices for command assembly, specifically employing bash arrays to avoid risks associated with shell evaluation of user-influenced strings.

attach-github-assets