Sammā Suit

Add 8 security governance layers to your OpenClaw agent — budget controls, permissions, audit logging, kill switch, identity signing, skill vetting, process isolation, and gateway protection.

Security Scan

The skill claims to enforce heavy security controls but provides no code, no provenance, and asks only for an opaque SAMMA_API_KEY — the design is plausible but the instructions are vague and could permit exfiltration or privileged changes; proceed only after getting concrete implementation details and a trusted source.

Purpose & Capability

The name/description (8 governance layers) matches the single declared requirement (SAMMA_API_KEY) in the sense that a remote security service could implement these features. However, there is no homepage, repository, install mechanism, or code — so it's unclear how the claimed enforcement would be applied. That gap is disproportionate given the scope of the controls described (kill switch, signing, isolation, audit logging).

Instruction Scope

SKILL.md contains only high-level claims and metadata; it ends abruptly at "Run:" and provides no concrete runtime steps, endpoints, or data-flow descriptions. Because the skill's functionality implies intercepting hooks and sending/receiving security-relevant data, the lack of precise instructions means the agent (or the remote service) could be given broad discretion to read or transmit sensitive session, tool, or message data. The vagueness is a risk.

Install Mechanism

There is no install spec and no code files (instruction-only). That's lower surface area for local disk writes, but it also means the skill must rely on a remote service to provide enforcement. Absence of an install spec increases need for provenance (source repo, release artifacts) because the enforcement is opaque.

Credentials

Only one env var (SAMMA_API_KEY) is required and is declared as the primary credential. That is proportionate if the skill delegates to an external governance API. However, because there is no information about what the API consumes or logs (e.g., whether it receives full messages, tool inputs, or skill code), the single key could give broad access to sensitive data. Lack of scoping or documentation is a concern.

Persistence & Privilege

The skill is not marked always:true and is user-invocable; it does not request system-wide persistence or modification of other skills in the metadata. No direct privilege escalation is declared in the manifest.

Scan Findings in Context

[no_scan_findings] expected: The regex scanner found nothing because the skill is instruction-only with no code files. This is expected but provides no assurance about behavior — absence of findings is not evidence of safety.

What to consider before installing

This skill makes large claims (kill switch, auditing, signing, isolation) but provides no code, repo, or detailed instructions and asks only for an opaque SAMMA_API_KEY from an unknown source. Before installing or supplying an API key, ask the publisher for: (1) a public repository or homepage with source code and releases; (2) detailed docs describing exactly what data is sent to the SAMMA service, endpoints, and retention/acl policies; (3) how the skill modifies agent hooks or runtime configuration and what permissions it needs; (4) a minimal, auditable implementation or a sandboxed demo you can run locally; and (5) recommended scoping/rotation for the API key (least privilege, short-lived token, limited scope). If you must test now, do so in an isolated agent environment with sanitized inputs and an ephemeral API key, and do not provide production credentials or sensitive datasets until you verify provenance and the data-flow. If the publisher cannot produce source or clear data-flow guarantees, treat this skill as untrusted.

Sammā Suit — Security Governance for OpenClaw

You are helping the user install and configure Sammā Suit, an open-source security framework that adds 8 enforced governance layers to OpenClaw as lifecycle hooks.

What It Does

Sammā Suit intercepts OpenClaw's plugin hooks to enforce:

• NIRVANA — Kill switch. Blocks all activity if agent is terminated.
• DHARMA — Permissions. Checks tools against allowed permission set.
• SANGHA — Skill vetting. Blocks unapproved skills via allowlist + AST scanning.
• KARMA — Budget controls. Per-agent monthly spending caps with hard ceiling.
• BODHI — Isolation. Injects timeout, token, and resource limits per agent.
• METTA — Identity. Ed25519 cryptographic signing on outbound messages.
• SILA — Audit trail. Logs every tool call, message, and session event.

Installation

Run: