The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the agent to use npx -y godpowers@7 in steps 3 and 4. This command automatically downloads the specified package from the npm registry if it is not already present locally.
Evidence: npx -y godpowers@7 verify "<verify command>" and npx -y godpowers@7 record harden ... in SKILL.md.
[REMOTE_CODE_EXECUTION]: By using npx, the skill executes code from a remote, unverifiable package ('godpowers'). This package is not associated with any trusted organization or well-known service, and its logic is executed with the permissions of the agent's environment.
Evidence: The package is invoked as a CLI tool to 'verify' and 'record' security audit results.
[COMMAND_EXECUTION]: The skill uses shell commands (npx) and potentially passes user-provided or agent-generated strings (<verify command>) to the execution environment, which could lead to command injection if the strings are not properly sanitized.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze external codebases for security vulnerabilities. This creates an attack surface where malicious instructions could be embedded in the code being audited to influence the auditor agent's behavior.
Ingestion points: The skill analyzes the "current code" and probes entry points where "untrusted input enters".
Boundary markers: No explicit delimiters or instructions to ignore embedded prompts in the analyzed code are provided.
Capability inventory: The skill has the ability to execute shell commands via npx and modify local files like STATE.md.
Sanitization: No sanitization or validation of the ingested code content is mentioned before it is processed by the agent.

god-harden