Preflight Checks

Test-driven behavioral verification for AI agents. Catches silent degradation when agent loads memory but doesn't apply learned behaviors. Use when building agent with persistent memory, testing after updates, or ensuring behavioral consistency across sessions.

What to check before installing/running: 1) Inspect the two scripts (init.sh and add-check.sh) — they copy templates into whatever directory you run them from and will update AGENTS.md and scoring text. Run them in a safe/test workspace or set WORKSPACE_DIR to a controlled folder, and back up AGENTS.md and memory files first. 2) SKILL.md mentions run-checks.sh, sessions_list, and a message CLI, but run-checks.sh is not included and those CLIs are not declared; if you plan to automate checks you must verify those tools exist and what permissions they have. 3) The documentation and example answers include behavioral rules that encourage the agent to act autonomously (e.g., sending Telegram messages without asking). That is a policy/operational choice — only enable automated runs if you are comfortable with that autonomy. 4) The README suggests cloning a GitHub repo (external source) — if following that path, review the repository contents and history before executing any scripts. 5) Sed -i and grep/sed usage may be platform-dependent (macOS vs Linux); test on your platform. Overall the package appears to implement what it claims, but the above mismatches and autonomy guidance make it worth manual review and cautious deployment.

Type: OpenClaw Skill Name: preflight-checks Version: 1.0.0 The OpenClaw AgentSkills skill bundle 'preflight-checks' is designed for test-driven behavioral verification of AI agents. The `SKILL.md` and `README.md` provide clear instructions for the agent to perform self-checks, which are aligned with its stated purpose and do not contain any prompt injection attempts to subvert the agent. The shell scripts (`init.sh`, `add-check.sh`) perform benign local file operations (copying templates, modifying markdown files) within the workspace to set up and manage these checks. Crucially, the `examples/ANSWERS-prometheus.md` file explicitly instructs the agent *not* to save user data (emails, names) and to *ask permission* before posting to public channels, demonstrating a clear intent to promote secure and responsible agent behavior rather than malicious activity. No evidence of data exfiltration, malicious execution, persistence, obfuscation, or supply chain attacks was found.