Send agent-to-agent pings with likes, comments, upvotes, adverts, and signed RustChain RTC payments across BoTTube, Moltbook, and LAN UDP.
Security Analysis
medium confidenceThe package largely matches its stated purpose, but there are multiple inconsistencies and risk signals (bundled server components, hard-coded/internal endpoints, undeclared credential needs, and developer docs referencing token files) that you should understand before installing or running it on a machine with sensitive data.
The skill's name/description (agent pings, RustChain payments, UDP/Webhook transports) align with the included code (many transport modules, RustChain transport, UDP/webhook servers, identity/keystore handling). However there are mismatches: the registry metadata claims "No install spec — instruction-only" while SKILL.md and README show pip install instructions and the package includes >80 source files (Python/JS). SKILL.md/README claim a public GitHub source but the registry listing shows Source: unknown / Homepage: none. These inconsistencies (packaged code vs 'instruction-only', and a missing homepage/source in metadata) are unexpected and worth verifying.
Runtime instructions and docs ask you to create ~/.beacon/config.json, run servers (webhook serve, dashboard), enable UDP broadcasts, and use mayday (which sends a reconstitution payload). The codebase contains a Flask backend (atlas/beacon_chat.py) that exposes relay endpoints and proxies chat to an LLM at a hard-coded internal IP (OLLAMA_URL = http://100.75.100.89:11434). The README claims mayday "never include private keys" but the mayday feature is described as sending journal digests, active goals and trust snapshots — these could leak sensitive internal data if misconfigured. The SKILL.md also says "No post-install telemetry" but the web UI and boot scripts fetch remote endpoints (e.g., swarmhub.onrender.com) at runtime. Overall the instructions/behavior permit network services and data sharing beyond what a minimal "ping" helper implies, so exercise caution.
Registry metadata lists no install spec (instruction-only) yet SKILL.md/README instruct pip/npm install and the package contains full source and CLI entry points. That mismatch is notable: this is not a pure 'instructions-only' skill — installing it will install runnable code. There is no external download from untrusted URLs in an install script inside the skill (no installer URLs flagged), but because a full codebase is bundled and pip install is recommended, installing executes more surface area than the registry metadata suggests.
The skill declares no required environment variables or primary credential, but many transports and runtime features normally require credentials (Discord tokens, RustChain wallet keys or RPC credentials, webhook endpoints, possible cloud API keys). The codebase and docs reference storage at ~/.beacon/identity (keystores), running servers on port 8071, and developer publisher instructions that mention reading tokens from local files (PUBLISH_CHECKLIST shows exporting a GitHub token from ~/git.txt). The package therefore expects or encourages secrets/config usage at runtime that are not declared in the metadata — this mismatch is a red flag for undeclared credential needs and possible accidental exposure.
always:false (good), but the code implements long-running components: a Flask backend (atlas/beacon_chat.py) intended to run on port 8071, a webhook server, UDP broadcast/listen, and CLI tools that create on-disk keystores (~/.beacon/). Those behaviors are consistent with the skill's scope but have elevated privilege/attack surface (network listeners, local keystores, inbox files). There's no indication the skill will force-enable itself platform-wide, but running it creates persistent services and files that should be run only with intent and on an appropriately isolated host.
Guidance
What to check before installing or running Beacon: - Verify source: the SKILL.md/README claim a GitHub repo (https://github.com/Scottcjn/beacon-skill) but the registry metadata shows Source: unknown and no homepage. Confirm the package you install matches the official repository and tag on GitHub before trusting it. - Run in isolation: because the package contains servers (Flask backend, webhook serve, UDP listeners) and writes keystore/config files under ~/.beacon/, test it inside a disposable VM, container, or sandbox — do not install on a machine with sensitive secrets or production workloads. - Inspect network endpoints: the code contains hard-coded endpoints (e.g., OLLAMA_URL at http://100.75.100.89:11434 and runtime fetches to swarmhub.onrender.com). Decide whether you trust those endpoints and search the codebase for any other hard-coded hosts before allowing outbound network access. - Credentials and keys: the skill advertises encrypted keystores but the package expects keys and may integrate with RustChain/Discord/etc. Expect to provide API tokens or wallet keys at runtime (not declared in registry metadata). Do not reuse high-value keys; create separate test wallets and tokens. Keep keystores password protected and verify file permissions (~/.beacon/ should be chmod 600 as docs suggest). - UDP/Webhook caution: UDP broadcast capability can leak metadata on local networks. Only enable broadcasts on trusted networks. If you run webhook servers, put them behind proper reverse proxies and TLS, and validate incoming payloads. - Mayday payloads: review what is included in a mayday before sending. Although docs say "never private keys," mayday can include journal digests and other state that may be sensitive; confirm exact fields and sanitize if necessary. - Financial flows: the README/advertise panel references Solana / RTC addresses and liquidity requirements. Treat any on-chain addresses and payment flows as real money — verify addresses and contract code before sending funds. - Developer docs: PUBLISH_CHECKLIST includes instructions that reference storing tokens in local files (~/git.txt) and exporting them — that is poor practice. Do not use secrets stored in plain files; use proper secret stores or environment variables with restricted permissions. - If unsure, audit: because full source is bundled, review the specific modules that will run in your environment (webhook, udp, rustchain transport, mayday, inbox) or ask someone with code-audit experience to review. If you have limited trust tolerance, avoid enabling network-facing features and limit the skill to read-only, offline usage.
Latest Release
v2.14.0
v2.14.0: Agent Scorecard dashboard, version sync across PyPI/npm/ClawHub
More by @Scottcjn
Published by @Scottcjn on ClawHub
