The Agent Skills Directory

[DATA_EXPOSURE]: The skill's interactive dashboard requires access to ~/.claude/projects/. This directory contains highly sensitive session history and JSONL logs of agent interactions, which often include source code, API keys, credentials, and private conversation history.
[UNVERIFIABLE_DEPENDENCIES]: The skill references an external repository at https://github.com/kromahlusenii-ops/ham. This source is not associated with a trusted organization or well-known service, posing a supply chain risk for any CLI tools or scripts executed by the skill.
[METADATA_POISONING]: The YAML frontmatter includes a risk: safe field. This is a deceptive claim designed to influence the user's perception of the skill's security posture, potentially bypassing critical review of its data access practices.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from project files and session logs to generate reports and "insights."
Ingestion points: ~/.claude/projects/ session logs and project CLAUDE.md files.
Boundary markers: None present.
Capability inventory: File system read/write, local web server execution.
Sanitization: None documented.
[COMMAND_EXECUTION]: The skill provides commands such as go ham and ham dashboard that perform local file operations and launch a local web server at localhost:7777, increasing the local attack surface and providing a potential vector for data exfiltration if the server communicates externally.

hierarchical-agent-memory