picoclaw-security-guardian

Picoclaw security posture skill with advisory awareness, configuration drift detection, and supply-chain verification guidance.

Security Scan

The skill's code, instructions, and required inputs align with its stated purpose (Picoclaw advisory checks, deterministic profile generation, drift detection, and supply-chain verification); nothing requested is disproportionate or unrelated.

Purpose & Capability

Name/description map to the actual files and runtime behavior: scripts and libs implement advisory filtering, deterministic profile generation, drift diffing, and checksum+signature verification. Declared binaries (node) and optional env vars are Picoclaw-specific and proportionate.

Instruction Scope

SKILL.md and the scripts only read Picoclaw config files / watched files, produce confined outputs under PICOCLAW_HOME, and require a local advisory feed or checksum manifests. The test/regression instructions do start Docker, run a local HTTP registry, and exercise Picoclaw-specific install flows — these are test harness actions, documented and isolated to pre-release regression.

Install Mechanism

No install spec is declared (instruction-only skill). All code is included in the repo and nothing is downloaded from third-party URLs. Tests pull docker images and use system tools, but there is no external archive download at install-time in the skill manifest itself.

Credentials

No required credentials or unrelated env vars are requested. Optional env vars are Picoclaw-specific (PICOCLAW_HOME, PICOCLAW_CONFIG, etc.). The code inspects config files within Picoclaw paths only, which is consistent with the stated purpose.

Persistence & Privilege

Skill does not request always:true and is described as read-only/on-demand in v0.0.1. It does not modify other skills or system-wide agent settings; outputs are confined to PICOCLAW_HOME and the code enforces path confinement and symlink checks.

Assessment

This package appears internally consistent and implements the features it claims. Before installing or running tests: 1) Run the scripts in an isolated/test environment (the regression harness runs Docker and builds/runs Picoclaw code). 2) Only provide verified advisory feeds, checksum manifests, and public keys from trusted sources — the skill enforces signed manifests by default and using '--allow-unsigned*' flags weakens guarantees. 3) Check your PICOCLAW_HOME and any --watch / --artifact paths you pass to avoid scanning or exposing unrelated files. 4) Review the regression/test harness before running it on a machine with sensitive data (it spawns containers, generates keys, and exercises install paths). Overall the code reads as a focused Picoclaw security helper — there are no unexplained network endpoints or unrelated credential requests.

Detailed architecture/operator docs: wiki/modules/picoclaw-security-guardian.md.

Goal

Provide Picoclaw with the same support-matrix security capabilities ClawSec tracks for mature platform modules:

Skill name	supported platform	security feed	config drift	agent posture-review lane	chain of supply verification
picoclaw-security-guardian	Picoclaw	Yes	Yes	Separate package	Yes

Threat model

Picoclaw is a lightweight AI gateway that can expose chat channels, a Web UI, tool execution, MCP servers, credentials, schedulers, and embedded/router deployments. This skill focuses on the trust boundaries where those features become security-sensitive.

Default safety posture

• Read-only by default.
• No scheduler creation in v0.0.1.
• No outbound network by default.
• Writes only explicit report/profile outputs under $PICOCLAW_HOME/security/clawsec/ unless the operator supplies test-local temporary paths.
• Advisory checks fail closed when verification state is not verified unless the operator passes --allow-unsigned for a documented emergency/offline window.

Security advisory awareness

Use scripts/check_advisories.mjs with a local feed/cache and verification state:

node scripts/check_advisories.mjs   --feed ~/.picoclaw/security/clawsec/feed.json   --state ~/.picoclaw/security/clawsec/feed-verification-state.json

The script filters advisories for picoclaw, ai-gateway, empty/all-platform advisories, or affected package entries containing picoclaw.

Drift protection

Generate a deterministic profile:

node scripts/generate_profile.mjs   --output ~/.picoclaw/security/clawsec/current-profile.json

Compare against an approved baseline:

node scripts/check_drift.mjs   --baseline ~/.picoclaw/security/clawsec/baseline-profile.json   --current ~/.picoclaw/security/clawsec/current-profile.json   --fail-on critical

Critical drift includes public Web UI enablement, Web UI auth disablement, workspace restriction disablement, unsigned/insecure verification mode, verified-feed regression, and watched-file/release-artifact fingerprint changes.

Chain-of-supply verification

Verify a Picoclaw release artifact against a checksum manifest plus detached signature. Signed manifest verification is required for a passing supply-chain verdict:

node scripts/verify_supply_chain.mjs \
  --artifact ./picoclaw \
  --checksums ./checksums.json \
  --signature ./checksums.json.sig \
  --public-key ./feed-signing-public.pem

Checksum-only mode is integrity-only, not provenance. Use --allow-unsigned-checksums only for short, documented offline triage windows; it should not satisfy production install verification.

Operator review notes

• Treat public UI binding (0.0.0.0, -public) as a critical review item until auth and network allowlists are proven.
• Treat MCP servers as separate trust boundaries; review each server's filesystem, network, and credential access.
• Treat third-party OpenWrt/LuCI wrappers as separate supply-chain artifacts. Verify provenance before installing them on routers.
• Never leave unsigned advisory mode enabled in recurring or production checks.

Validation

python utils/validate_skill.py skills/picoclaw-security-guardian
node skills/picoclaw-security-guardian/test/profile.test.mjs
node skills/picoclaw-security-guardian/test/drift.test.mjs
node skills/picoclaw-security-guardian/test/supply_chain.test.mjs
bash -n skills/picoclaw-security-guardian/test/picoclaw_security_guardian_sandbox_regression.sh

Pre-release install regression

Before publishing v0.0.1 release artifacts, run the isolated install lane from the repo root:

skills/picoclaw-security-guardian/test/picoclaw_security_guardian_sandbox_regression.sh

The regression installs the skill through Picoclaw's own find_skills / install_skill path from a local ClawHub-compatible registry into an isolated Docker-hosted Picoclaw workspace with isolated HOME, PICOCLAW_HOME, and PICOCLAW_WORKSPACE. It verifies signed release-artifact preflight inputs, confirms Picoclaw's skill loader can list/load the installed skill, then runs the installed copy's profile, drift, advisory fail-closed, advisory filtering, and supply-chain verification paths against Picoclaw-style config.json and launcher-config.json files.