The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/report.py identifies and executes commands found within the target repository (such as package.json scripts, Makefiles, or files in scripts/) using subprocess.run via /bin/zsh -lc. This execution is triggered when flags like --probe-verification or --probe-runtime are provided.
[CREDENTIALS_UNSAFE]: The tool includes logic to probe the host environment for the presence and validity of sensitive API tokens, including GITHUB_TOKEN, LINEAR_API_KEY, NOTION_TOKEN, SENTRY_AUTH_TOKEN, and VERCEL_TOKEN. Results of these probes are included in the generated readiness report.
[DATA_EXFILTRATION]: The script reads the user's local configuration file at ~/.codex/config.toml to identify installed plugins and active MCP servers, exposing local environment metadata to the agent's context.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted project data (manifests, documentation, and scripts) to derive an execution plan. A malicious repository could define dangerous commands in its package.json or Makefile that the skill would then suggest or execute during the readiness assessment.

repo-readiness-bootstrap