The Agent Skills Directory

[DATA_EXFILTRATION]: The skill documentation explicitly states that the dashboard feature reads session data from ~/.claude/projects/. This directory contains sensitive information including past conversation history, system prompts, and project-specific context. Accessing these paths constitutes a high-risk data exposure. Additionally, the skill launches a web dashboard on localhost:7777, creating a potential network pathway for the accessed sensitive data.
[COMMAND_EXECUTION]: The skill defines several commands such as go ham and ham dashboard that perform file system operations and process management. The go ham command performs automated project platform detection and recursive directory analysis to generate configuration files throughout the project structure.
[EXTERNAL_DOWNLOADS]: The skill is sourced from an untrusted repository (github.com/kromahlusenii-ops/ham) which is not associated with any trusted organizations or well-known services. It requires an external Node.js environment to execute its dashboard functionality.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection because it ingests untrusted data from existing session history files.
Ingestion points: Reads session JSONL files from ~/.claude/projects/ (documented in SKILL.md).
Boundary markers: None identified in the provided description to separate historical session content from current instructions.
Capability inventory: Performs file system writes (creating CLAUDE.md files) and network operations (launching a local web server).
Sanitization: No evidence of sanitization or validation of the ingested session data before processing or display.

hierarchical-agent-memory