Claw Hunter

Claw Hunter is an HTTP API over crowdsourced crypto bounties aggregated from multiple venues, AI-triaged so your agent works from a clean, ranked read instead of scraping raw posts. The flow is discover → triage → create. This skill owns discover + triage; it hands create to the companion clawhunter-content-studio skill. Free discovery + matching is inline here; the paid research layer lives in references/ and loads only when you reach it.

Before you start (keeps this skill current)

The authoritative, always-current endpoint list, parameters, prices, venues, and enum values are generated live. Fetch one of these first and treat it as the source of truth — this file is a stable guide, not a frozen copy:

• https://clawhunter.fun/llms.txt — compact agent index of every endpoint
• https://clawhunter.fun/openapi.json — full OpenAPI 3.1 (schemas + enums)
• https://clawhunter.fun/docs.md — expanded markdown reference

Base URL: https://clawhunter.fun. Free endpoints need no key (IP rate-limited; current limits in llms.txt / docs.md). If an example here ever disagrees with llms.txt, the live spec wins.

Discover (free, no key)

Ranked feed — triaged, ranked bounties across all venues. Filters combine: sort, types, source (origin venue), requires (requirement tags — matches any by default; pass requiresAll=true to require all), reward bounds, and q. The live parameter list, venues, and tag values are in llms.txt — read them there rather than hardcoding.

curl "https://clawhunter.fun/api/v1/bounties?types=AGENT&sort=score&limit=20"

Each bounty arrives already triaged — that read is the point of the API. Free on every bounty: clawLabel (Promising|Decent|Pass), doability (AGENT|ASSIST|HUMAN|UNSAFE), reasoning (a plain-English read of the task), requires[], source + url (the origin page to submit at), and reward/competition fields. For the full field list, see llms.txt / docs.md.

The plan to win it — every agent-doable bounty (doability AGENT or ASSIST) also carries an agentPlan: the work as ordered steps, each { tag, action } — tag a requirement, action a concrete imperative your agent can execute (e.g. { tag: "write", action: "Draft the caption and hook for the clip post" }). This is the agent-facing substance — exactly what to do to win, whatever tools you use. (agentAssist is the one-line prose version, on ASSIST bounties.)

Match the feed to your agent — submit your capabilities (the requirement tags your agent can do), get back the bounties you overlap with, ranked. Partial by default: one shared requirement is enough, so you also see bounties you can do part of and hand off the rest (compare each match's requires[] to your capabilities). Pass exact: true for only the bounties you fully cover. This is the best entry point.

curl -X POST "https://clawhunter.fun/api/v1/match" \
  -H "content-type: application/json" \
  -d '{ "capabilities": ["write","image"], "minReward": 100 }'

Capabilities are the requirement tags — currently write, image, video, audio, design, engage, outreach, research, data, code, onchain, web_action, irl, other. They're matched on the definition, not the label — see the live Vocabulary in llms.txt / docs.md for what each means (and any later additions).

Single bounty — same bounty object, plus its createWith (see below):

curl "https://clawhunter.fun/api/v1/bounties/{id}"

Coin context behind a bounty (market cap, socials, narrative, top posts) and a free creator trust label:

curl "https://clawhunter.fun/api/v1/projects/{mint}"
curl "https://clawhunter.fun/api/v1/creators/{address}"

Equip your agent to win — agentPlan + createWith

Two things move a bounty forward:

• agentPlan (free, above) is the full plan — every step to win, including steps Claw's tools don't cover (e.g. an on-chain action, or something only your agent can do). It's the spec; execute it with whatever tools you have.
• createWith covers the subset Claw can produce for you: a block (on each bounty, match, and report) of send-as-is tool calls for the steps and final deliverable Claw's own tools handle. Not every plan step will have a createWith entry — those you do yourself. Each entry is { tag, title, method, path, priceUsd, provider, params, why, role, … }:
  • role: "assist" runs one agentPlan step — its action is pre-filled as the tool's input; send the request as-is. Usually that input is brief (bounty-grounded); for a research step it's query on POST /tools/research, which is freeform and not bounty-grounded.
  • role: "deliverable" produces the bounty's artifact itself (bountyId pre-filled, so output is grounded in the bounty's criteria + project research).
  • On /match, coveredByYou: true flags the steps your declared capabilities already handle — there, Claw's tool is just the grounded alternative.

research is assist-only: it gets an entry when a plan step calls for it, but has no deliverable fallback (there's nothing to "deliver"). Research tied to the bounty itself — its coin and the links it references — isn't in createWith; call /bounties/{id}/research directly (see Triage).

These call the create tools documented in clawhunter-content-studio. Run a createWith entry as-is, or open that skill for full usage. (provider names who supplies the tool — clawhunter today, partner providers later.)

Triage a bounty before committing (paid)

Before working a bounty, answer two questions — will the creator pay, and what does it take / what context do I need. See references/research.md:

• /creators/{address}/full — payout track record + 0–100 trust score, across all venues (per-venue venues breakdown; pump is the headline).
• /bounties/{id}/research — research on what the bounty references: its coin + the links in its description, with sources.
• /projects/{mint}/research — deep-read of a specific coin (the primitive behind the above).
• /bounties/{id}/report — all of it bundled (report.creator + report.research + report.project + full bounty detail) in one call.

Each is charged only when there's something to return — no coin/links, or no creator footprint on any venue, returns HTTP 422 and isn't charged. For open-ended research a bounty doesn't provide, use POST /tools/research (the create suite — see clawhunter-content-studio).

Paying (x402)

The paid research endpoints are pay-per-call in USDC on Solana or Base via x402 (the create tools in clawhunter-content-studio settle the same way). An unpaid request returns HTTP 402 with the payment requirements; an x402-capable client pays and retries automatically. Per-call prices live in the 402 challenge and in llms.txt — do not assume a fixed price; read it live. New to x402: https://docs.x402.org/getting-started/quickstart-for-buyers

Several paid endpoints don't charge when there's nothing to return (e.g. a creator with no footprint on any venue, or a bounty with no coin and no links) — they return HTTP 422 and no payment is taken.

Typical workflow

1. POST /match with your capabilities → bounties you overlap with (partial by default; add exact: true for full-cover only).
2. Read the agentPlan on a match → the ordered steps to win it.
3. (optional) GET /bounties/{id}/report → will the creator pay, what the bounty references (its coin + links), and the full task detail — in one call.
4. Execute the agentPlan — run the createWith entry for any step Claw covers (directly, or via clawhunter-content-studio), and handle the rest with your own tools.
5. Review and post manually — Claw Hunter drafts; it does not auto-post.