The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill triages repositories by reading configuration files and project headers (e.g., package.json, composer.json, plugin headers). A maliciously crafted repository could provide poisoned data intended to manipulate the agent's understanding of the project or its subsequent workflow recommendations.\n
Ingestion points: The script scripts/detect_wp_project.mjs reads various project files using fs.readFileSync and traverses directories using fs.readdirSync.\n
Capability inventory: The resulting triage report informs the agent's choices for commands and guardrails.\n
Boundary markers: Output is structured JSON, but contents are derived from untrusted external repository files.\n
Sanitization: The script extracts strings for display and recommendation without explicit sanitization for downstream prompt or shell safety.\n- [COMMAND_EXECUTION]: The script dynamically generates recommended shell commands (e.g., npm run lint, composer install) based on scripts and configurations found within the repository's package.json and composer.json files. An attacker with control over the repository could inject malicious commands into these files, which the agent might then be prompted to execute.\n
Evidence: The buildRecommendations function in scripts/detect_wp_project.mjs constructs command strings based on the presence of specific script keys in project metadata files.

wp-project-triage