The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The execution flow in SKILL.md includes a recursive search for sensitive keywords like api_key, password, secret, and token across the target directory using grep. Because the results are printed directly to standard output, any discovered credentials would be exposed to the AI agent's context and session logs without redaction.
[REMOTE_CODE_EXECUTION]: The skill utilizes npx to execute the claude-flow package for multiple phases of the audit (audit-pipeline, security-scan, performance-report, bottleneck detect). These commands do not pin the package to a specific version or a trusted source, creating a supply-chain vulnerability where a compromised package version could lead to arbitrary code execution on the user's machine.
[EXTERNAL_DOWNLOADS]: The use of npx triggers runtime downloads of packages from the npm registry. Without version constraints or integrity verification, the skill is susceptible to package hijacking or malicious updates from unverified third-party sources.
[COMMAND_EXECUTION]: The skill executes a comprehensive bash script that performs a wide range of system operations, including directory creation (mkdir), file writing (cat >), and complex command pipelines using jq, grep, and npm. While these operations are consistent with the tool's purpose, they represent a significant capability tier that could be abused if the input parameters are manipulated.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to process external project data which presents an injection attack surface.
Ingestion points: The skill reads all source files in the user-provided target_path (e.g., via grep and npx tools).
Boundary markers: Absent; there are no instructions to the agent to treat file content as data rather than instructions.
Capability inventory: The skill has the ability to execute shell commands, run remote packages via npx, write files to the disk, and perform dependency audits.
Sanitization: Absent; the content of files found in the target directory is processed directly by the audit tools without filtering.

production-readiness