clawsec-suite

ClawSec Suite

Vercel Skills Installation

Install with the Vercel Skills CLI for this harness:

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

Operational Notes

• Required runtime: node, npx, openclaw, curl, jq, shasum, openssl, unzip
• Side effects: setup scripts install an advisory hook under ~/.openclaw/hooks, optionally create an unattended openclaw cron job, and use npx clawhub@latest install for guarded installs
• Network behavior: fetches signed advisory feed artifacts and remote catalog metadata unless you pin local paths
• Trust model: the suite can recommend removal or block risky installs, but removal/install overrides stay approval-gated

This means clawsec-suite can:

• monitor the ClawSec advisory feed,
• track which advisories are new since last check,
• cross-reference advisories against locally installed skills,
• recommend removal for malicious-skill advisories and require explicit user approval first,
• and still act as the setup/management entrypoint for other ClawSec protections.

Included vs Optional Protections

Built into clawsec-suite

• Embedded signed advisory trust set: advisories/feed.json, feed.json.sig, checksums.json, checksums.json.sig, and feed-signing-public.pem
• Portable heartbeat workflow in HEARTBEAT.md
• Advisory polling + state tracking + affected-skill checks
• OpenClaw advisory guardian hook package: hooks/clawsec-advisory-guardian/
• Setup scripts for hook and optional cron scheduling: scripts/
• Guarded installer: scripts/guarded_skill_install.mjs
• Dynamic catalog discovery for installable skills: scripts/discover_skill_catalog.mjs

Installed separately (dynamic catalog)

clawsec-suite does not hard-code add-on skill names in this document.

Discover the current catalog from the authoritative index (https://clawsec.prompt.security/skills/index.json) at runtime:

SUITE_DIR="${INSTALL_ROOT:-

Fallback behavior:

• If the remote catalog index is reachable and valid, the suite uses it.
• If the remote index is unavailable or malformed, the script falls back to suite-local catalog metadata in skill.json.

Installation

Cross-shell path note

• In bash/zsh, keep path variables expandable (for example, INSTALL_ROOT="$HOME/.openclaw/skills").
• Do not single-quote home-variable paths (avoid '$HOME/.openclaw/skills').
• In PowerShell, set an explicit path:
  • $env:INSTALL_ROOT = Join-Path $HOME ".openclaw\\skills"
• If a path is passed with unresolved tokens (like \$HOME/...), suite scripts now fail fast with a clear error.

Option A: Via clawhub (recommended)

Option B: Manual download with signature + checksum verification

OpenClaw Automation (Hook + Optional Cron)

After installing the suite, enable the advisory guardian hook:

The setup script prints a preflight review before it installs and enables the persistent hook.

Optional: create/update a periodic cron nudge (default every 6h) that triggers a main-session advisory scan:

The cron setup script prints a preflight review before it creates or updates the unattended job.

What this adds:

• scan on agent:bootstrap and /new (command:new),
• compare advisory affected entries against installed skills,
• consider advisories with application: "openclaw" (and legacy entries without application for backward compatibility),
• notify when new matches appear,
• and ask for explicit user approval before any removal flow.

Restart the OpenClaw gateway after enabling the hook. Then run /new once to force an immediate scan in the next session context.

Guarded Skill Install Flow (Double Confirmation)

When the user asks to install a skill, treat that as the first request and run a guarded install check:

Behavior:

• If no advisory match is found, install proceeds.
• If --version is omitted, matching is conservative: any advisory that references the skill name is treated as a match.
• If advisory match is found, the script prints advisory context and exits with code 42.
• Then require an explicit second confirmation from the user and rerun with --confirm-advisory:

This enforces:

1. First confirmation: user asked to install.
2. Second confirmation: user explicitly approves install after seeing advisory details.

Embedded Advisory Feed Behavior

The embedded feed logic uses these defaults:

• Remote consolidated feed URL: https://clawsec.prompt.security/advisories/feed.json
• Feed contents: NVD CVEs, approved community advisories, and provisional GHSA-without-CVE advisories.
• Remote feed signature URL: ${CLAWSEC_FEED_URL}.sig (override with CLAWSEC_FEED_SIG_URL)
• Remote checksums manifest URL: sibling checksums.json (override with CLAWSEC_FEED_CHECKSUMS_URL)
• Local seed fallback: ~/.openclaw/skills/clawsec-suite/advisories/feed.json
• Local feed signature: ${CLAWSEC_LOCAL_FEED}.sig (override with CLAWSEC_LOCAL_FEED_SIG)
• Local checksums manifest: ~/.openclaw/skills/clawsec-suite/advisories/checksums.json
• Pinned feed signing key: ~/.openclaw/skills/clawsec-suite/advisories/feed-signing-public.pem (override with CLAWSEC_FEED_PUBLIC_KEY)
• State file: ~/.openclaw/clawsec-suite-feed-state.json
• Hook rate-limit env (OpenClaw hook): CLAWSEC_HOOK_INTERVAL_SECONDS (default 300)

Fail-closed verification: Feed signatures are required by default. Checksum manifests are verified when companion checksum artifacts are available. Set CLAWSEC_ALLOW_UNSIGNED_FEED=1 only as a temporary migration bypass when adopting this version before signed feed artifacts are available upstream.

Quick feed check

Exploitability Context

Advisories in the feed can include exploitability_score and exploitability_rationale fields to help agents prioritize real-world threats:

• Exploitability scores: high, medium, low, or unknown
• Context-aware assessment: Considers attack vector, authentication requirements, and AI agent deployment patterns
• Exploit availability: Detects public exploits and weaponization status

When processing advisories, prioritize by exploitability in addition to severity. A HIGH severity + HIGH exploitability CVE is more urgent than a CRITICAL severity + LOW exploitability CVE.

For detailed methodology, see the exploitability scoring documentation.

Heartbeat Integration

Use the suite heartbeat script as the single periodic security check entrypoint:

• skills/clawsec-suite/HEARTBEAT.md

It handles:

• suite update checks,
• feed polling,
• new-advisory detection,
• affected-skill cross-referencing,
• approval-gated response guidance for malicious/removal advisories,
• and persistent state updates.

Approval-Gated Response Contract

If an advisory indicates a malicious or removal-recommended skill and that skill is installed:

1. Notify the user immediately with advisory details and severity.
2. Recommend removing or disabling the affected skill.
3. Treat the original install request as first intent only.
4. Ask for explicit second confirmation before deletion/disable action (or before proceeding with risky install).
5. Only proceed after that second confirmation.

The suite hook and heartbeat guidance are intentionally non-destructive by default.

Advisory Suppression / Allowlist

The advisory guardian pipeline supports opt-in suppression for advisories that have been reviewed and accepted by your security team. This is useful for first-party tooling or advisories that do not apply to your deployment.

Activation

Advisory suppression requires a single gate: the configuration file must contain "enabledFor" with "advisory" in the array. No CLI flag is needed -- the sentinel in the config file IS the opt-in gate.

If the enabledFor array is missing, empty, or does not include "advisory", all advisories are reported normally.

Config File Resolution (4-tier)

The advisory guardian resolves the suppression config using the same priority order as the audit pipeline:

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

Config Format

Sentinel Semantics

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

Matching Rules

• checkId: exact match against the advisory ID (e.g., CVE-2026-25593 or CLAW-2026-0001)
• skill: case-insensitive match against the affected skill name from the advisory
• Both fields must match for an advisory to be suppressed

Required Fields per Suppression Entry

Shared Config with Audit Pipeline

The advisory and audit pipelines share the same config file. Use the enabledFor array to control which pipelines honor the suppression list:

Audit entries (with check identifiers like skills.code_safety) are only matched by the audit pipeline. Advisory entries (with advisory IDs like CVE-2026-25593 or CLAW-2026-0001) are only matched by the advisory pipeline. Each pipeline filters for its own relevant entries.

Optional Skill Installation

Discover currently available installable skills dynamically, then install the ones you want:

Machine-readable output is also available for automation:

Security Notes

• Always verify checksums.json signature before trusting its file URLs/hashes, then verify each file checksum.
• Verify advisory feed detached signatures; do not enable CLAWSEC_ALLOW_UNSIGNED_FEED outside temporary migration windows.
• Keep advisory polling rate-limited (at least 5 minutes between checks).
• Treat critical and high advisories affecting installed skills as immediate action items.
• If you migrate off standalone clawsec-feed, keep one canonical state file to avoid duplicate notifications.
• Pin and verify public key fingerprints out-of-band before first use.