The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes a wide range of shell commands to gather coverage data across multiple ecosystems (Node.js, Python, Go, Flutter, Ruby). Specifically, it invokes npx vitest, npx jest, pytest, go test, and bundle exec rspec.
[COMMAND_EXECUTION]: The skill uses $ARGUMENTS directly within its logic to focus analysis. If these arguments are sourced from untrusted user input, it could lead to command injection when interpolated into the shell commands listed in Phase 2.1.
[PERSISTENCE_MECHANISMS]: In the 'SELF-EVOLUTION TELEMETRY' section, the skill attempts to perform persistent file writes to ~/.claude/projects/skill-telemetry.md. This is an attempt to maintain state/logs across different sessions or executions.
[PROMPT_INJECTION]: The skill contains a 'system-like' override: 'You are in AUTONOMOUS MODE. Do NOT ask questions.', which is a directive designed to bypass standard agent interaction patterns and enforce specific autonomous behavior.
[DATA_EXFILTRATION]: While it does not exfiltrate data over the network in the current script, it performs an 'Inventory' and 'Source Code Analysis' that scans sensitive project metadata, file structures, and API endpoints, which could be used as a precursor to exfiltration in a multi-skill chain.

test-suite