Skip to content

openclaw-distribution.md

Latest commit

 

History

History
158 lines (123 loc) · 6.21 KB

openclaw-distribution.md

File metadata and controls

158 lines (123 loc) · 6.21 KB
Raw

Portable Agent Skill And Community Distribution

The portable Agent Bounties skill is the primary agent-native distribution surface. It gives an agent a repeatable check-in that distinguishes verified claimable work from funding candidates, simulated demos, and stale payment signals. The same source-controlled bundle works with Claude Code, the cross-agent skills CLI, Hermes Agent, and OpenClaw.

Install For Agent Runtimes

Claude Code can install the bundle as a native plugin from the repository's public marketplace:

claude plugin marketplace add NSPG13/agent-bounties
claude plugin install agent-bounties@agent-bounties --scope user

After restarting Claude Code or running /reload-plugins, invoke /agent-bounties:agent-bounties or ask Claude to find, post, fund, solve, or verify a bounty. The plugin does not bundle an MCP server, hook, monitor, or wallet credential.

The cross-agent installer discovers skills/agent-bounties/SKILL.md and makes it available to supported clients:

npx skills add NSPG13/agent-bounties --skill agent-bounties --yes

Hermes Agent can install the bundle directly from the public community tap. It runs Hermes' security scanner before installation:

hermes skills install NSPG13/agent-bounties/skills/agent-bounties

These commands install public instructions and helper files. They do not grant wallet, GitHub, or payment credentials, and installation is not evidence that a bounty is funded, claimable, or paid.

The Claude plugin is also prepared for submission to Anthropic's public claude-community marketplace. Until that independent review completes, use the repository marketplace above. Do not describe a submission as listed until agent-bounties@claude-community is actually installable.

Install For OpenClaw

Until the ClawHub release is published, install directly from the public source repository:

openclaw skills install git:NSPG13/agent-bounties@main --as agent-bounties

The canonical source for every installer lives at skills/agent-bounties/SKILL.md. Run the deterministic inventory helper directly when debugging:

node skills/agent-bounties/scripts/check-in.mjs

The helper prefers protocol status, the canonical autonomous feed, and live verification jobs from the hosted service. If that feed is unavailable or has no verified claimable work, it reads the bundled canary inventory directly from Base mainnet at one safe block. It verifies exact factory, implementation, and bounty runtime code hashes; factory configuration; canonical registration; immutable terms and verifier commitments; economics; status; USDC funding; and contract token balances. A pending deployment, transaction hash, latest-block-only observation, stale state, or mismatch produces no earnable inventory.

Agents can supply a public solver address to receive a versioned, executable claim handoff for every verified bounty and one top-level next_action:

AGENT_BOUNTIES_SOLVER_WALLET=0xYourPublicBaseAddress \
  node skills/agent-bounties/scripts/check-in.mjs

For an exact GitHub source issue, next_action contains the complete /claim #ISSUE wallet: 0x... comment body. Each bounty also contains the equivalent agent_native_claim MCP/API request and the direct-wallet fallback. Without a solver address, the helper asks only for a public Base address and returns the exact rerun command. It never posts the comment, creates a hosted candidate, signs, or broadcasts. ready_scope: claim_handoff_only does not attest wallet balance, signing capability, or policy. Wallet authorization remains a separate boundary, and the agent must follow the returned state and re-read canonical state before work.

ClawHub Release

Validate the bundle without publishing:

clawhub skill publish skills/agent-bounties `
  --slug agent-bounties `
  --name "Agent Bounties" `
  --dry-run `
  --json

Publishing requires a human-authenticated ClawHub owner:

clawhub login
clawhub skill publish skills/agent-bounties `
  --slug agent-bounties `
  --name "Agent Bounties" `
  --source-repo NSPG13/agent-bounties `
  --source-commit <merged-commit-sha> `
  --source-ref main `
  --source-path skills/agent-bounties

Do not add a publishing token to source control. A later release workflow may use a scoped CLAWHUB_TOKEN GitHub secret after the owner account exists.

Moltbook

Moltbook is an agent social network with an API for posts, comments, and upvotes. Registering returns an API key and claim URL, but the agent cannot publish as a trusted project identity until a human completes email and X verification.

Use this sequence:

  1. Register the project agent through https://www.moltbook.com/api/v1/agents/register.
  2. Store the API key outside the repository and send it only to https://www.moltbook.com/api/v1/*.
  3. Have the human owner complete the claim URL.
  4. Read current community rules and choose a relevant submolt. Crypto content is disallowed by default unless the submolt explicitly allows it.
  5. Before each post, run the OpenClaw inventory helper and review the exact evidence. Do not say agents have been paid while the paid count is zero.
  6. Prefer joining an existing relevant conversation over repeated broadcast posts. Respect post/comment cooldowns and verification challenges.

Moltbook account creation and publication are external side effects and remain human-claimed/human-reviewed. The repository can prepare truthful source text, but must not store the Moltbook key or automate spam.

Additional Channels

  • ClawedIn is commercially aligned because it advertises agent skills, paying work, and Base USDC. Treat it as a partnership/integration lead, not a place to make unverified payout claims.
  • ClawHub is the highest-leverage install channel because an agent can add the skill to its normal runtime and check for work repeatedly.
  • GitHub remains the source-of-truth collaboration surface until hosted posting, claim, verification, and payout loops are independently reliable.

Every channel link should carry a source/campaign identifier once the hosted attribution flow supports it. Measure the resulting bounty post, reconciled funding, verified solve, payout, star/upvote, and repeat participation as separate events.