Prompt Security Audit (openclaw)

Vercel Skills Installation

Install with the Vercel Skills CLI for this harness:

npx skills add prompt-security/clawsec --skill openclaw-audit-watchdog -a openclaw -y

Installation Options

You can get openclaw-audit-watchdog in two ways:

Option A: Bundled with ClawSec Suite (Recommended)

If you've installed clawsec-suite, you may already have this!

Openclaw-audit-watchdog is bundled alongside ClawSec Suite to provide crucial automated security audit capabilities. When you install the suite, if you don't already have the audit watchdog installed, it will be deployed from the bundled copy.

Advantages:

• Convenient - no separate download needed
• Standard location - installed to ~/.openclaw/skills/openclaw-audit-watchdog/
• Preserved - if you already have audit watchdog installed, it won't be overwritten
• Single verification - integrity checked as part of suite package

Option B: Standalone Installation (This Page)

Install openclaw-audit-watchdog independently without the full suite.

When to use standalone:

• You only need the audit watchdog (not other suite components)
• You want to install before installing the suite
• You prefer explicit control over audit watchdog installation

Advantages:

• Lighter weight installation
• Independent from suite
• Direct control over installation process

Standalone installation usually involves a network download from the published GitHub release. Verify the release source and archive integrity before installing it on production hosts.

Continue below for standalone installation instructions.

Release Artifact Verification

For standalone installs, verify the signed release manifest before trusting SKILL.md, skill.json, or the archive. The skill.json file is the package metadata/SBOM source, and the release pipeline signs checksums.json with the ClawSec release key.

set -euo pipefail

SKILL_NAME="openclaw-audit-watchdog"
VERSION="0.1.9"
REPO="prompt-security/clawsec"
TAG="${SKILL_NAME}-v${VERSION}"
BASE="https://github.com/${REPO}/releases/download/${TAG}"
ZIP_NAME="${SKILL_NAME}-v${VERSION}.zip"
TMP_DIR="$(mktemp -d)"
trap 'rm -rf "$TMP_DIR"' EXIT

RELEASE_PUBKEY_SHA256="711424e4535f84093fefb024cd1ca4ec87439e53907b305b79a631d5befba9c8"

curl -fsSL "$BASE/checksums.json" -o "$TMP_DIR/checksums.json"
curl -fsSL "$BASE/checksums.sig" -o "$TMP_DIR/checksums.sig"
curl -fsSL "$BASE/signing-public.pem" -o "$TMP_DIR/signing-public.pem"
curl -fsSL "$BASE/$ZIP_NAME" -o "$TMP_DIR/$ZIP_NAME"
curl -fsSL "$BASE/SKILL.md" -o "$TMP_DIR/SKILL.md"
curl -fsSL "$BASE/skill.json" -o "$TMP_DIR/skill.json"

ACTUAL_PUBKEY_SHA256="$(openssl pkey -pubin -in "$TMP_DIR/signing-public.pem" -outform DER | shasum -a 256 | awk '{print $1}')"
if [ "$ACTUAL_PUBKEY_SHA256" != "$RELEASE_PUBKEY_SHA256" ]; then
  echo "ERROR: signing-public.pem fingerprint mismatch" >&2
  exit 1
fi

openssl base64 -d -A -in "$TMP_DIR/checksums.sig" -out "$TMP_DIR/checksums.sig.bin"
openssl pkeyutl -verify -rawin -pubin \
  -inkey "$TMP_DIR/signing-public.pem" \
  -sigfile "$TMP_DIR/checksums.sig.bin" \
  -in "$TMP_DIR/checksums.json" >/dev/null

hash_file() {
  if command -v shasum >/dev/null 2>&1; then
    shasum -a 256 "$1" | awk '{print $1}'
  else
    sha256sum "$1" | awk '{print $1}'
  fi
}

verify_manifest_file() {
  asset="$1"
  path="$2"
  expected="$(jq -r --arg asset "$asset" '.files[$asset].sha256 // empty' "$TMP_DIR/checksums.json")"
  if [ -z "$expected" ]; then
    echo "ERROR: checksums.json missing $asset" >&2
    exit 1
  fi
  actual="$(hash_file "$path")"
  if [ "$actual" != "$expected" ]; then
    echo "ERROR: checksum mismatch for $asset" >&2
    exit 1
  fi
}

expected_archive="$(jq -r '.archive.sha256 // empty' "$TMP_DIR/checksums.json")"
if [ -z "$expected_archive" ]; then
  echo "ERROR: checksums.json missing archive.sha256" >&2
  exit 1
fi
actual_archive="$(hash_file "$TMP_DIR/$ZIP_NAME")"
if [ "$actual_archive" != "$expected_archive" ]; then
  echo "ERROR: archive checksum mismatch" >&2
  exit 1
fi

verify_manifest_file "SKILL.md" "$TMP_DIR/SKILL.md"
verify_manifest_file "skill.json" "$TMP_DIR/skill.json"

echo "Signed release manifest, archive, SKILL.md, and skill.json verified."

Only install or extract the archive after this verification succeeds.

Operational requirements

Required runtime:

• openclaw
• node
• bash

Optional runtime:

• sendmail for local MTA delivery
• SMTP relay via PROMPTSECSMTPHOST / PROMPTSECSMTPPORT
• git only if PROMPTSECGITPULL=1

This skill is not always-on by default, but when invoked it creates or updates an unattended openclaw cron job. Review the configured DM/email recipients and the host's openclaw/SMTP environment before enabling it.

Goal

Create (or update) a daily cron job that:

1. Runs:

• openclaw security audit --json
• openclaw security audit --deep --json

1. Summarizes findings (critical/warn/info + top findings)

1. Sends the report to:

• a user-selected DM target (channel + recipient id/handle)
• an optional email recipient only when PROMPTSECEMAILTO is configured

Default schedule: daily at 23:00 (11pm) in the chosen timezone.

Delivery:

• DM to the configured target
• Optional email only when an explicit recipient is configured
• Persistence via openclaw cron (unattended recurring job)

Usage Examples

Example 1: Quick Start (Environment Variables)

For automated/MDM deployments, set environment variables before invoking:

export PROMPTSEC_DM_CHANNEL="telegram"
export PROMPTSEC_DM_TO="@yourhandle"
export PROMPTSEC_EMAIL_TO="[email protected]" # optional
export PROMPTSEC_TZ="America/New_York"
export PROMPTSEC_HOST_LABEL="prod-server-01"

# Then invoke the skill
/openclaw-audit-watchdog

The skill will automatically configure and create the cron job without prompts. If PROMPTSECEMAILTO is omitted, the job remains DM-only.

Example 2: Interactive Setup

If environment variables aren't set, the skill will prompt minimally:

User: /openclaw-audit-watchdog

Agent: Setting up daily security audit watchdog...
       What channel should I use for delivery? (e.g., telegram, slack)
User: telegram

Agent: What's the recipient ID or handle?
User: @myhandle

Agent: Optional email recipient? (leave blank to disable email)
User: [email protected]

Agent: Which timezone for the 23:00 daily run? (default: UTC)
User: America/Los_Angeles

Agent: ✓ Created cron job "Daily security audit (Prompt Security)"
       Schedule: Daily at 23:00 America/Los_Angeles
       Delivery: telegram → @myhandle, email → [email protected]

Example 3: Updating Existing Job

If a job already exists, the skill updates it instead of creating duplicates:

User: /openclaw-audit-watchdog

Agent: Found existing "Daily security audit (Prompt Security)" job.
       Current: Daily at 23:00 UTC → telegram:@oldhandle

       Update delivery target? (current: telegram:@oldhandle)
User: slack:#security-alerts

Agent: ✓ Updated cron job
       Schedule: Daily at 23:00 UTC
       Delivery: slack:#security-alerts

Example 4: What Gets Delivered

Each day at the scheduled time, you'll receive a report like:

🔭 Daily Security Audit Report
Host: prod-server-01
Time: 2026-02-16 23:00:00 America/New_York

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
SUMMARY
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
✓ Standard Audit: 12 checks passed, 2 warnings
✓ Deep Audit: 8 probes passed, 1 critical

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
CRITICAL FINDINGS
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
[CRIT-001] Unencrypted API Keys Detected
→ Remediation: Move credentials to encrypted vault or use environment variables

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
WARNINGS
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
[WARN-003] Outdated Dependencies Found
→ Remediation: Run `openclaw security audit --fix` to update

[WARN-007] Weak Permission on Config File
→ Remediation: chmod 600 ~/.openclaw/config.json

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
Run `openclaw security audit --deep` for full details.

Example 5: Custom Schedule

Want a different schedule? Set it before invoking:

# Run every 6 hours instead of daily
export PROMPTSEC_SCHEDULE="0 */6 * * *"
/openclaw-audit-watchdog

Example 6: Multiple Environments

For managing multiple servers, use different host labels:

# On dev server
export PROMPTSEC_HOST_LABEL="dev-01"
export PROMPTSEC_DM_TO="@dev-team"
/openclaw-audit-watchdog

# On prod server
export PROMPTSEC_HOST_LABEL="prod-01"
export PROMPTSEC_DM_TO="@oncall"
/openclaw-audit-watchdog

Each will send reports with clear host identification.

Example 7: Suppressing Known Findings

To suppress audit findings that have been reviewed and accepted, pass the --enable-suppressions flag and ensure the config file includes the "enabledFor": ["audit"] sentinel:

# Create or edit the suppression config
cat > ~/.openclaw/security-audit.json <<'JSON'
{
  "enabledFor": ["audit"],
  "suppressions": [
    {
      "checkId": "skills.code_safety",
      "skill": "clawsec-suite",
      "reason": "First-party security tooling — reviewed by security team",
      "suppressedAt": "2026-02-15"
    }
  ]
}
JSON

# Run with suppressions enabled
/openclaw-audit-watchdog --enable-suppressions

Suppressed findings still appear in the report under an informational section but are excluded from critical/warning totals.

Suppression / Allowlist

The audit pipeline supports an opt-in suppression mechanism for managing reviewed findings. Suppression uses defense-in-depth activation: two independent gates must both be satisfied.

Activation Requirements

1. CLI flag: The --enable-suppressions flag must be passed at invocation.
2. Config sentinel: The configuration file must include "enabledFor" with "audit" in the array.

If either gate is absent, all findings are reported normally and the suppression list is ignored.

Config File Resolution (4-tier)

1. Explicit --config <path> argument
2. OPENCLAWAUDITCONFIG environment variable
3. ~/.openclaw/security-audit.json
4. .clawsec/allowlist.json

Config Format

{
  "enabledFor": ["audit"],
  "suppressions": [
    {
      "checkId": "skills.code_safety",
      "skill": "clawsec-suite",
      "reason": "First-party security tooling — reviewed by security team",
      "suppressedAt": "2026-02-15"
    }
  ]
}

Sentinel Semantics

• "enabledFor": ["audit"] -- audit suppression active (requires --enable-suppressions flag too)
• "enabledFor": ["advisory"] -- only advisory pipeline suppression (no effect on audit)
• "enabledFor": ["audit", "advisory"] -- both pipelines honor suppressions
• Missing or empty enabledFor -- no suppression active (safe default)

Matching Rules

• checkId: exact match against the audit finding's check identifier (e.g., skills.code_safety)
• skill: case-insensitive match against the skill name from the finding
• Both fields must match for a finding to be suppressed

Installation flow (interactive)

Provisioning (MDM-friendly): prefer environment variables (no prompts).

Required env:

• PROMPTSECDMCHANNEL (e.g. telegram)
• PROMPTSECDMTO (recipient id)

Optional env:

• PROMPTSECEMAILTO (email recipient; if unset, email delivery stays disabled)
• PROMPTSEC_TZ (IANA timezone; default UTC)
• PROMPTSECHOSTLABEL (label included in report; default uses hostname)
• PROMPTSECINSTALLDIR (stable path used by cron payload to cd before running runner; default: ~/.config/security-checkup)
• PROMPTSECGITPULL=1 (runner will git pull --ff-only if installed from git)
• OPENCLAWAUDITCONFIG (suppression config path to persist into the cron payload)
• PROMPTSECSENDMAILBIN (explicit sendmail path)
• PROMPTSECSMTPHOST, PROMPTSECSMTPPORT, PROMPTSECSMTPHELO, PROMPTSECSMTPFROM (SMTP relay settings)

Path expansion rules (important):

• In bash/zsh, use PROMPTSECINSTALLDIR="$HOME/.config/security-checkup" (or absolute path).
• Do not pass a single-quoted literal like '$HOME/.config/security-checkup'.
• On PowerShell, prefer: $env:PROMPTSECINSTALLDIR = Join-Path $HOME ".config/security-checkup".
• If path resolution fails, setup now exits with a clear error instead of creating a literal $HOME directory segment.

Interactive install is last resort if env vars or defaults are not set. Keep prompts minimal: DM target is required, email is optional, and the user should see a concise preflight review before persistence is enabled.

Create the cron job

Use the cron tool to create a job with:

• schedule.kind="cron"
• schedule.expr="0 23 *"
• schedule.tz=<installer tz>
• sessionTarget="isolated"
• wakeMode="now"
• payload.kind="agentTurn"
• payload.deliver=true

Before creating or updating the job, print a preflight review that explicitly states:

• this action creates or updates an unattended recurring job,
• the required runtime (openclaw, node, bash),
• the configured DM target,
• whether email is enabled and to which recipient,
• the install directory and timezone used for execution.

Payload message template (agentTurn)

Create the job with a payload message that instructs the isolated run to:

1. Run the audits

• Prefer JSON output for robust parsing:

- openclaw security audit --json - openclaw security audit --deep --json

1. Render a concise text report:

Include:

• Timestamp + host identifier if available
• Summary counts
• For each CRITICAL/WARN: checkId + title + 1-line remediation
• If deep probe fails: include the probe error line

1. Deliver the report:

• DM to the chosen user target using message tool

Email delivery requirement

Email delivery is optional. Only promise or attempt it when PROMPTSECEMAILTO is configured.

If PROMPTSECEMAILTO is set, attempt delivery in this priority order:

A) If a local sendmail-compatible binary is available, use it first.

B) Otherwise, fallback to the configured SMTP relay:

• PROMPTSECSMTPHOST
• PROMPTSECSMTPPORT
• optional PROMPTSECSMTPHELO
• optional PROMPTSECSMTPFROM

If neither path is possible, still DM the user and include a line:

• "NOTE: could not deliver email to <PROMPTSECEMAILTO> via configured sendmail/SMTP path"

If PROMPTSECEMAILTO is not set, the cron payload must explicitly describe email as disabled rather than implying a default recipient.

Idempotency / updates

Before adding a new job:

• cron.list(includeDisabled=true)
• If a job with name matching "Daily security audit" exists, update it instead of adding a duplicate:

- adjust schedule tz/expr - adjust DM target

Suggested naming

• Job name: "Daily security audit (Prompt Security)"

Minimal recommended defaults (do not auto-change config)

The cron’s report should suggest fixes but must not apply them.

Do not run openclaw security audit --fix unless explicitly asked.