2o-verification
Human verification for AI agents. Submit claims, draft responses, or observation requests to human domain experts via the 2O API. Returns structured verdicts with confidence scores and evidence.
- License 许可
- MIT
- 最近更新
- 2026-06-10
- Stars
- ★ 3
- Forks
- 0
01 前言 Front Matter
- name
- 2o-verification
- requires
- { "env": [ "TWO_O_API_KEY" ] }
- description
- Human verification for AI agents. Submit claims, draft responses, or observation requests to human domain experts via the 2O API. Returns structured verdicts with confidence scores and evidence.
03 SKILL.md 正文
2O — Human Verification API
You can call the 2O API to get human expert verification on factual claims, empathy reviews of draft responses, and real-world witness observations. Human domain experts review submissions and return structured results.
Setup
The user needs a 2O API key. Register at https://www.2oapi.xyz/register to get one ($5 free credits included). The key is stored in the TWO_O_API_KEY environment variable.
Verify a Factual Claim
Use this when you need a human expert to verify whether a claim is true, false, or partially true.
curl -s -X POST "https://www.2oapi.xyz/api/v1/verify" \
-H "Content-Type: application/json" \
-H "Authorization: Bearer $TWO_O_API_KEY" \
-d '{
"claim": "The claim to verify",
"domain": "general",
"budget_cents": 50
}'
Required fields:
claim(string): The factual statement to verifydomain(string): One offinancial,general,geographic,identity,legal,medical,scientific,technicalbudget_cents(integer): Payment in cents. Minimum 50 ($0.50)
Optional fields:
context(string): Additional context for the verifierurgency(string):low,medium(default),high,criticalcallback_url(string): Webhook URL for completion notificationtier(string):standard(default, 1 verifier),consensus(3 verifiers, min $3.00),expert_panel(5 verifiers, min $10.00)
Response: Returns a JSON object with id (request ID), status ("pending"), and expires_at.
Request an Empathy Review
Use this when you have a draft response to a sensitive situation and want a human to review it for tone and empathy.
curl -s -X POST "https://www.2oapi.xyz/api/v1/empathize" \
-H "Content-Type: application/json" \
-H "Authorization: Bearer $TWO_O_API_KEY" \
-d '{
"draft_response": "The draft text to review",
"context": "Description of the situation",
"budget_cents": 50
}'
Required fields:
draft_response(string): The draft AI response to reviewcontext(string): The situation the user is inbudget_cents(integer): Minimum 50
Returns: empathy_score, tone_assessment, sensitivity_flags, suggested_revision
Request a Witness Observation
Use this when you need a human to physically observe or confirm something in the real world.
curl -s -X POST "https://www.2oapi.xyz/api/v1/witness" \
-H "Content-Type: application/json" \
-H "Authorization: Bearer $TWO_O_API_KEY" \
-d '{
"observation_query": "What to observe or check",
"budget_cents": 100
}'
Required fields:
observation_query(string): What the human should observebudget_cents(integer): Minimum 100 ($1.00)
Optional fields:
location_address(string): Address for the observationlocation_lat,location_lng(number): GPS coordinatesphoto_required(boolean): Whether photos are needed
Returns: observation, photo_urls, observed_at
Check Request Status
After submitting any request, poll for status using the request ID:
curl -s "https://www.2oapi.xyz/api/v1/verify/REQUEST_ID" \
-H "Authorization: Bearer $TWO_O_API_KEY"
Status values: pending (waiting for verifier), assigned (verifier working on it), completed (result ready), expired (timed out)
Poll every 30-60 seconds. All three task types (verify, empathize, witness) use this same endpoint.
Get the Result
Once status is completed, fetch the full result:
curl -s "https://www.2oapi.xyz/api/v1/verify/REQUEST_ID/result" \
-H "Authorization: Bearer $TWO_O_API_KEY"
Verify results include: verdict (verified/refuted/partially_verified), confidence (0-1), explanation, evidence_urls
Check Balance
curl -s "https://www.2oapi.xyz/api/v1/agent/balance" \
-H "Authorization: Bearer $TWO_O_API_KEY"
Returns balance_cents, total_spent_cents, total_escrowed_cents.
Guidelines
- Verify important claims before presenting them as fact to users
- Use empathy review before responding to sensitive situations (grief, conflict, crisis)
- Use witness for questions about real-world conditions that can't be answered from training data
- Start with
budget_cents: 50for verify/empathize,100for witness - Higher budgets and urgency attract faster responses
- The typical flow is: submit -> poll status -> get result
- Do not poll more frequently than every 30 seconds
04 同仓库更多 Skills
youtube-content
YouTube transcripts to summaries, threads, blogs.
xurl
X/Twitter via xurl CLI: post, search, DM, media, v2 API.
touchdesigner-mcp
Control a running TouchDesigner instance via twozero MCP — create operators, set parameters, wire connections, execute Python, build real-time visuals. 36 native tools.
test-driven-development
TDD: enforce RED-GREEN-REFACTOR, tests before code.
teams-meeting-pipeline
Operate the Teams meeting summary pipeline via Hermes CLI — summarize meetings, inspect pipeline status, replay jobs, manage Microsoft Graph subscriptions.
systematic-debugging
4-phase root cause debugging: understand bugs before fixing.