The Agent Skills Directory

[COMMAND_EXECUTION]: The skill performs shell command execution to run the graphify tool, using arguments derived from user input ($ARGUMENTS). The lack of explicit sanitization instructions for these inputs when interpolated into shell commands creates a potential surface for command injection.
[PERSISTENCE]: The skill instructs the agent to suggest or install a git post-commit hook (graphify hook install), which establishes a persistence mechanism that executes code automatically upon repository changes. Additionally, it maintains state by appending to a local file (~/.claude/skills/save-tokens/LEARNINGS.md) outside the project scope.
[INDIRECT_PROMPT_INJECTION]: The skill's primary function is to ingest and process a codebase using AST extraction and graph analysis. This processing of untrusted local data creates a vulnerability surface for indirect prompt injection, where malicious instructions embedded in code comments or strings could attempt to influence the agent's behavior.
[PROMPT_INJECTION]: The instructions contain strong directives for the agent to "Do NOT ask the user questions" and "Proceed autonomously." These commands instruct the agent to bypass standard interactive safety checks and operate with reduced human oversight.
[EXTERNAL_DOWNLOADS]: The skill provides automated recovery steps that include installing a third-party package (graphifyy) from a public registry using pipx, pip, or uv if the command-line tool is not detected on the system.

save-tokens