The Agent Skills Directory

[PROMPT_INJECTION]: The static detector identified strings typically associated with prompt injection ('ignore previous instructions') within SKILL.md. However, contextual analysis reveals these are part of a 'Retrieval Safety' section designed to instruct the agent to ignore malicious directives found in external documentation. This is a security feature, not a vulnerability.
[INDIRECT_PROMPT_INJECTION]: The skill utilizes a data ingestion flow where framework documentation is fetched from external sources. To mitigate potential injection attacks from these sources, the skill mandates a 'Retrieval Safety' protocol: it treats fetched content as untrusted data, instructs the agent to extract only specific technical information (API definitions, usage examples), and explicitly commands the agent to ignore any embedded directives targeting the model's behavior.
[CREDENTIALS_UNSAFE]: The provided reference files, particularly security-checklist.md and observability-checklist.md, include robust guidelines to prevent the accidental exposure of secrets, PII, and authentication tokens in source code, version control, and application logs.
[DEPENDENCY_SECURITY]: The security-checklist.md contains a detailed framework for managing supply chain risks. It provides specific instructions for different package managers (npm, pnpm, yarn) to verify lockfiles, audit dependencies, and strictly control the execution of lifecycle scripts, which are common vectors for malware.
[REMOTE_CODE_EXECUTION]: The skill does not perform any unauthorized remote code execution. All mentioned tools (e.g., axe-core, lighthouse, pa11y) are well-known, industry-standard development and testing utilities invoked by the user for auditing purposes.

source-driven-development