The Agent Skills Directory

[PROMPT_INJECTION]: Indirect prompt injection surface via log analysis and external research.
Ingestion points: The skill ingests untrusted data from application logs (./logs.txt), CI/CD logs (gh run view), and external documentation via the researcher subagent.
Boundary markers: The instructions lack explicit boundary markers or warnings to the agent to treat these external inputs as data rather than instructions, which could allow a malicious log entry to influence agent behavior.
Capability inventory: The skill possesses high-privilege capabilities, including the ability to execute shell commands via the Bash subagent, modify the project codebase, and perform version control operations with git-manager.
Sanitization: No explicit sanitization or filtering of external data is mentioned in the workflow before processing.
[COMMAND_EXECUTION]: Extensive use of shell command execution for development tasks.
The skill workflow relies on the Bash subagent to execute commands for testing (npm test, bun test, pytest), building projects, and running diagnostic tools like tsc or eslint.
In workflow-ui.md, the skill invokes a local script located at .claude/skills/ui-ux-pro-max/scripts/search.py, creating a dependency on the integrity of local script files.
[DATA_EXFILTRATION]: Potential exposure of sensitive data through log file processing.
The skill is designed to read and analyze logs from various sources. Application and CI/CD logs often contain sensitive information such as API keys, authentication tokens, or personally identifiable information (PII) if the logging mechanism is not correctly configured. The skill's purpose is to find and fix errors, which often involves inspecting these potentially sensitive files.

fix