The Agent Skills Directory

[DATA_EXFILTRATION]: The skill is designed to read the local Claude Code chat history file (~/.claude/history.jsonl). This file contains highly sensitive information, including user messages, project details, and potentially hardcoded secrets or sensitive code snippets. The skill then transmits a summary of this data to an external Slack workspace using an MCP tool.
[COMMAND_EXECUTION]: The skill instructions direct the agent to utilize several tools from the Rube MCP set (RUBE_SEARCH_TOOLS, RUBE_MANAGE_CONNECTIONS, RUBE_MULTI_EXECUTE_TOOL) to perform network searches on HackerNews and send messages to Slack.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection. It processes untrusted data (chat history from history.jsonl) without utilizing boundary markers or sanitization. A malicious instruction previously stored in the chat history could be triggered during the analysis phase.
Ingestion points: The skill reads from ~/.claude/history.jsonl in Step 1 of the instructions.
Boundary markers: Absent. There are no instructions for the agent to use delimiters or warnings to ignore instructions embedded within the chat history.
Capability inventory: The skill has the capability to read local files (~/.claude/history.jsonl) and perform network operations (Slack messaging and HackerNews search via Rube MCP).
Sanitization: Absent. The skill extracts and analyzes raw content from the chat history without escaping or validating the input data.

developer-growth-analysis