The Agent Skills Directory

[DATA_EXFILTRATION]: The skill accesses sensitive data files outside the current project directory, specifically the user's Claude Code interaction history located at ~/.claude/projects/ and the global settings file ~/.claude/settings.json. These files contain logs of previous conversations, including code snippets, system interactions, and configuration details that may contain proprietary information or developer secrets.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external sources which could influence agent behavior:
Ingestion points: Project conversation history (.jsonl files) is read by scripts/analyze.sh, and GitHub code search results are ingested via scripts/github-discovery.sh (using the gh CLI).
Boundary markers: There are no explicit boundary markers or instructions to ignore embedded commands within the processed history or search data.
Capability inventory: The skill possesses the capability to execute shell commands, perform web fetches, and modify global user configuration files.
Sanitization: While jq is used to filter specific fields from the history logs, the scripts/analyze.sh script uses manual string concatenation (printf) to build JSON output, which is vulnerable to injection if tool names in the history contain specifically crafted characters. The resulting summary is then interpreted by the agent, potentially allowing malicious content in past logs to trigger unintended actions.
[EXTERNAL_DOWNLOADS]: The skill fetches documentation and feature metadata from docs.claude.com and anthropic.com. These are official vendor domains and are considered safe references for configuration guidance.
[COMMAND_EXECUTION]: The skill relies on several local bash scripts (analyze.sh, analyze-claude-md.sh, github-discovery.sh, fetch-features.sh) to perform data extraction and project analysis. These scripts are executed via the bash tool as part of the core optimization workflow.

claude-code-analyzer