The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The script slash-command-dep-map.sh and the subagent guide in subagent-dependency-mapper.md use npx to run the claude-flow@alpha package. Runtime execution of external packages from public registries introduces supply-chain risks and potential for arbitrary code execution.
[COMMAND_EXECUTION]: The skill executes various system tools and package manager commands including npm audit, md5sum, and mkdir. These are triggered during the analysis phase and are used for dependency auditing and local cache management.
[DATA_EXFILTRATION]: As part of its security audit phase, the skill is configured to send package data to external services such as Snyk and GitHub. This includes reading environment variables for authentication (e.g., SNYK_TOKEN) and transmitting manifest content over the network.
[PROMPT_INJECTION]: The file skill-report.json contains a pre-written security audit of the skill, making authoritative safety claims like 'Verdict: SAFE' and 'not blocked.' This is a self-referential pattern intended to bypass or influence independent security analysis.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted project manifest data and incorporates it into an HTML visualization template. In subagent-dependency-mapper.md (lines 482-489), project metadata is assigned to innerHTML without sanitization, posing a Cross-Site Scripting (XSS) risk if an attacker-controlled project is analyzed.

when-mapping-dependencies-use-dependency-mapper