The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill's implementation playbook contains code that communicates with external security and package metadata services, including registry.npmjs.org, pypi.org, and bundlephobia.com. These connections are used to fetch vulnerability data and package statistics necessary for the auditing process.
[COMMAND_EXECUTION]: The skill provided in the resources uses shell commands to interact with package managers such as npm and pip. These commands are used to discover project dependencies and facilitate automated security remediation.
[DATA_EXFILTRATION]: Network requests identified in the implementation playbook transmit package names and versions to external registries. While this involves sending data out of the environment, it is limited to public package metadata required for auditing and does not appear to involve sensitive local configuration files or secrets.
[PROMPT_INJECTION]: The skill presents an attack surface for indirect prompt injection due to its processing of untrusted dependency manifest files.
Ingestion points: The DependencyDiscovery class in resources/implementation-playbook.md reads and parses multiple dependency file types (e.g., package.json, requirements.txt).
Boundary markers: The instructions do not define strict boundaries or delimiters to isolate data from these files from the agent's instruction context.
Capability inventory: The skill has the capability to execute shell commands, write to the filesystem (via update scripts), and perform network operations.
Sanitization: Manifest content is processed without explicit validation or sanitization, which could theoretically allow a maliciously crafted manifest file to influence the agent's behavior during an audit.

codebase-cleanup-deps-audit