The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it parses and executes "Agent Prompts" extracted from GitHub PR comments.
Ingestion points: PR comments (general and inline) fetched via gh api (SKILL.md, references/api-reference.md).
Boundary markers: No explicit delimiters or instructions to ignore nested prompts are present.
Capability inventory: File-write operations for code fixes, command execution (git, gh, npm, pytest), and network access via GitHub API.
Sanitization: Employs sed to remove HTML tags but does not validate the semantic content of the extracted instructions (references/api-reference.md).
[COMMAND_EXECUTION]: The skill invokes several CLI tools including gh for API interaction, git for version control, and various test runners (npm, pytest, make, ruff, black) to verify code changes.
[EXTERNAL_DOWNLOADS]: Fetches PR metadata and discussion threads from the GitHub API, which is a trusted and well-known service. It may also execute dependency installation commands like npm install or pip install if required by the project environment.

qodo-pr-resolver