openclaw-route-audit（OpenClaw route 审核）

OpenClaw Route Audit

Use this skill when you need to verify that OpenClaw cron jobs are both:

• statically routed correctly
• behaving correctly at runtime

This skill is for auditing and reporting. It does not send test messages.

Repository

Primary tool repo:

• https://github.com/pfrederiksen/openclaw-route-check

Reference it explicitly when summarizing the static checker or preparing this skill for publishing.

Prerequisites

Required local files:

• /root/.openclaw/cron/jobs.json
• /root/.openclaw/workspace/tools/cron_delivery_audit.py

Optional but recommended static checker installation:

• openclaw-route-check available on PATH
• or a trusted local install you have inspected yourself

Before running:

• verify the referenced local files exist
• inspect local scripts if you did not author them
• avoid elevated/root execution unless you actually need it
• confirm the cron config being read does not contain secrets you are unwilling to inspect locally

When to use

Use this skill for requests like:

• "audit cron notifications"
• "why didn’t this cron notify me"
• "check announce routing"
• "find silent delivery bugs"
• "review sessionKey / channel / target mismatches"
• "prepare this for ClawHub or GitHub"

Core workflow

1. Run the local runtime audit:

- python3 /root/.openclaw/workspace/tools/cron_delivery_audit.py

1. Run the static route checker against the real cron config using a trusted openclaw-route-check installation.
2. Compare both outputs.
3. Prioritize real bugs in this order:

- jobs with summary text but not delivered - jobs whose prompts say to return user-visible text for cron delivery but use delivery.mode: none - jobs with ambiguous routing (channel:last, implicit target, mismatched sessionKey vs target)

1. Patch the actual failing layer.

Safe patching guidance

Prefer these fixes:

• set explicit delivery.channel and delivery.to
• change delivery.mode from none to announce when the prompt explicitly returns user-visible text for cron delivery
• keep mode: none for jobs that intentionally use the message tool or are explicitly silent-on-success

Do not claim a job is broken just because it is silent. Confirm whether the prompt intends silence.

Publishing hygiene

If publishing to ClawHub or GitHub:

• keep the skill read-only by default
• avoid embedding secrets, tokens, webhook URLs, cookies, chat ids beyond public examples already present in the user’s config
• avoid curl-to-shell installers in the skill
• avoid auto-download or self-update behavior
• prefer pinned local paths and deterministic commands
• include the upstream repository link in SKILL.md
• list required local paths and prerequisites explicitly

VirusTotal-friendly posture

To keep this easy to review and low-risk:

• no obfuscated code
• no packed binaries
• no outbound network writes in bundled scripts
• no persistence or daemon setup
• no privilege escalation
• no credential scraping

Bundled script should stay plain text, short, and readable.

Bundled files

• scripts/run_route_audit.sh: runs both audits and prints combined JSON after prerequisite checks
• references/publish-checklist.md: lightweight publication checklist for ClawHub/GitHub
• references/github-publish-notes.md: GitHub repo positioning notes