The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The 'Credentials Management' section directs the agent to locate and parse a centralized 'access keys file' (e.g., ~/Documents/Access.txt) to extract API keys for OpenAI, Anthropic, Render, and Replicate. This encourages storing multiple high-value secrets in an unencrypted plain text file.
[DATA_EXFILTRATION]: The skill instructions require the agent to access a sensitive file path (~/Documents/Access.txt). Accessing such files to retrieve credentials for project configuration constitutes a significant data exposure risk.
[COMMAND_EXECUTION]: The mandatory TDD workflow involves the agent executing various shell commands to run tests and linters (e.g., npm test, pytest, ruff, mypy).
[DYNAMIC_EXECUTION]: The skill implements a workflow where the agent writes and immediately executes its own code and tests, creating an attack surface for dynamically generated content.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from project specifications and todo lists in _project_specs/ (e.g., active.md, current-state.md). It lacks specific boundary markers or sanitization, and when combined with the agent's capability to execute shell commands and write files, this creates a vulnerability to indirect prompt injection.

base