# BountyVerdict Agent Decision APIs

> Seven bounded checks for public GitHub decisions and MCP tool-catalog upgrade safety.

## Decide if this tool is useful

- Free representative result: https://bountyverdict-agent-production.mimirslab.workers.dev/api/sample
- x402 seller manifest: https://bountyverdict-agent-production.mimirslab.workers.dev/.well-known/x402
- OpenAPI 3.1 contract: https://bountyverdict-agent-production.mimirslab.workers.dev/openapi.json
- Remote MCP server: https://bountyverdict-agent-production.mimirslab.workers.dev/mcp (MCP 2025-11-25 Streamable HTTP; six paid read-only tools; SkillVerdict excluded)
- MCP endpoint metadata: https://bountyverdict-agent-production.mimirslab.workers.dev/.well-known/mcp.json
- MCP server card: https://bountyverdict-agent-production.mimirslab.workers.dev/.well-known/mcp/server-card.json
- RFC 9727 API catalog: https://bountyverdict-agent-production.mimirslab.workers.dev/.well-known/api-catalog
- Integration declaration: https://bountyverdict-agent-production.mimirslab.workers.dev/.well-known/integrations.json
- Agentic Resource Discovery catalog: https://bountyverdict-agent-production.mimirslab.workers.dev/.well-known/ai-catalog.json (six neutral representative queries; publication is not proof of third-party indexing)
- MCP Registry name: io.github.cristianmoroaica/bountyverdict
- Client-specific MCP install and paid-call guide: https://cristianmoroaica.github.io/bountyverdict/llms-install.md
- A remote MCP connection does not provide a wallet; direct automatic payment requires an x402-aware MCP client such as @x402/mcp
- Every valid unpaid MCP call includes io.github.cristianmoroaica/bountyverdict/http-payment-handoff with the equivalent exact HTTP x402 request for an authorized wallet
- Umbrella routing skill: https://cristianmoroaica.github.io/bountyverdict/skills/route-github-agent-checks/SKILL.md
- Install router: npx skills add cristianmoroaica/bountyverdict --skill route-github-agent-checks -y
- Install all operating skills: npx skills add cristianmoroaica/bountyverdict --skill '*' -y
- Paid check: GET https://bountyverdict-agent-production.mimirslab.workers.dev/api/verdict?issue_url=<PUBLIC_GITHUB_ISSUE_URL>
- Price: $0.05 USDC per successful result through x402 v2
- Free portfolio sample: https://bountyverdict-agent-production.mimirslab.workers.dev/api/portfolio/sample
- Paid portfolio: POST https://bountyverdict-agent-production.mimirslab.workers.dev/api/portfolio with {"issue_urls":[...]} for $0.40 USDC
- Portfolio size: 2 to 10 unique public GitHub issue URLs
- Verdicts: AVOID, CAUTION, VIABLE
- Free HarnessVerdict sample: https://bountyverdict-agent-production.mimirslab.workers.dev/api/harness/sample
- Paid HarnessVerdict: GET https://bountyverdict-agent-production.mimirslab.workers.dev/api/harness?repo_url=<PUBLIC_GITHUB_REPOSITORY_URL>
- HarnessVerdict price: $0.03 USDC per successful commit-pinned audit
- Harness verdicts: READY, REVIEW, REPAIR
- Free SkillVerdict sample: https://bountyverdict-agent-production.mimirslab.workers.dev/api/skill/sample
- Paid SkillVerdict: GET https://bountyverdict-agent-production.mimirslab.workers.dev/api/skill?repo_url=<PUBLIC_GITHUB_REPOSITORY_URL>&skill_path=<SKILL_DIRECTORY>
- SkillVerdict price: $0.06 USDC per successful commit-pinned pre-install audit
- Skill verdicts: LOW_RISK, REVIEW, BLOCK
- Free RunVerdict sample: https://bountyverdict-agent-production.mimirslab.workers.dev/api/run/sample
- Paid RunVerdict: GET https://bountyverdict-agent-production.mimirslab.workers.dev/api/run?run_url=<PUBLIC_GITHUB_ACTIONS_RUN_URL>
- RunVerdict price: $0.04 USDC per bounded exact-attempt diagnosis
- Run verdicts: PASS, WAIT, RETRY, FIX, INVESTIGATE
- Free FlakeVerdict sample: https://bountyverdict-agent-production.mimirslab.workers.dev/api/flake/sample
- Paid FlakeVerdict: GET https://bountyverdict-agent-production.mimirslab.workers.dev/api/flake?run_url=<PUBLIC_GITHUB_ACTIONS_RUN_URL>&attempt=<OPTIONAL_EXACT_ATTEMPT>
- FlakeVerdict price: $0.07 USDC per bounded historical retry decision
- Flake verdicts: CONFIRMED_FLAKE, LIKELY_FLAKE, RECURRING_FAILURE, NEW_FAILURE, INCONCLUSIVE, NOT_FAILED
- Free MCPDriftVerdict sample: https://bountyverdict-agent-production.mimirslab.workers.dev/api/mcp-drift/sample
- Paid MCPDriftVerdict: POST https://bountyverdict-agent-production.mimirslab.workers.dev/api/mcp-drift with complete MCP 2025-11-25 baseline and current tools/list snapshots
- MCPDriftVerdict price: $0.02 USDC per precomputed deterministic result
- MCPDriftVerdict verdicts: UNCHANGED, SAFE_ADDITIVE, REVIEW, INCONCLUSIVE, BREAKING, SECURITY_REGRESSION
- MCPDriftVerdict validates and computes before payment; unsupported or invalid catalogs receive no 402 challenge
- Failed or invalid checks are not settled
- Every successful result includes explicit service_reuse guidance so agents know when to call the same reliable bounded check again

## Escrow marketplace

- Public the402 provider catalog: https://api.the402.ai/v1/services/catalog?provider=p_d4b4ece39162409b&limit=100
- These six routes offer the same tested products through the402 escrow; inspect each exact input and deliverable schema in the public catalog before purchase.
- SkillVerdict is intentionally excluded from this marketplace while its earned-placement experiment is frozen.
- Monthly bundle: POST https://api.the402.ai/v1/plans/plan_ec6c49878dc34636/subscribe — $1.05 USDC for up to 20 combined requests across the six listed services.
- BountyVerdict: POST https://api.the402.ai/v1/services/svc_5e36dabc8b434e95/purchase — agent price $0.053 USDC; instant fulfillment
- BountyVerdict Portfolio: POST https://api.the402.ai/v1/services/svc_780bf04bd8204b2f/purchase — agent price $0.42 USDC; instant fulfillment
- GitHub Agent Instruction Audit — HarnessVerdict: POST https://api.the402.ai/v1/services/svc_df4baf282b7d48d5/purchase — agent price $0.032 USDC; instant fulfillment
- GitHub Actions CI Failure Diagnosis — RunVerdict: POST https://api.the402.ai/v1/services/svc_cdd16073d02c4429/purchase — agent price $0.042 USDC; instant fulfillment
- CI Flake Verdict: POST https://api.the402.ai/v1/services/svc_565a2a5c8e154b6e/purchase — agent price $0.074 USDC; instant fulfillment
- MCP schema drift compatibility — MCPDriftVerdict: POST https://api.the402.ai/v1/services/svc_40e97a390c5b4d71/purchase — agent price $0.021 USDC; instant fulfillment

## NEAR Agent Market

- Public provider profile: https://market.near.ai/agents/bountyverdict
- These six automated services expose the same tested input and output contracts through NEAR Agent Market; SkillVerdict remains excluded during its frozen experiment.
- BountyVerdict: hire https://market.near.ai/hire?service_id=88c3e8f6-07f4-414e-bc43-c5ad61cf21fd; authenticated POST https://market.near.ai/v1/services/88c3e8f6-07f4-414e-bc43-c5ad61cf21fd/invoke — 1 USDC
- BountyVerdict Portfolio: hire https://market.near.ai/hire?service_id=3b496165-8b2c-4c97-8593-1242e5d15384; authenticated POST https://market.near.ai/v1/services/3b496165-8b2c-4c97-8593-1242e5d15384/invoke — 1 USDC
- GitHub Agent Instruction Audit — HarnessVerdict: hire https://market.near.ai/hire?service_id=dd781a20-6643-42ee-8c19-0c16425e685d; authenticated POST https://market.near.ai/v1/services/dd781a20-6643-42ee-8c19-0c16425e685d/invoke — 1 USDC
- GitHub Actions CI Failure Diagnosis — RunVerdict: hire https://market.near.ai/hire?service_id=88bb0780-9121-4acd-a2e2-4f8a2bc005a8; authenticated POST https://market.near.ai/v1/services/88bb0780-9121-4acd-a2e2-4f8a2bc005a8/invoke — 1 USDC
- CI Flake Verdict: hire https://market.near.ai/hire?service_id=7385fdea-5d7f-43d4-9fb2-738b46316d0f; authenticated POST https://market.near.ai/v1/services/7385fdea-5d7f-43d4-9fb2-738b46316d0f/invoke — 1 USDC
- MCP schema drift compatibility — MCPDriftVerdict: hire https://market.near.ai/hire?service_id=0a0b0909-2829-4437-b23e-4376a61041ba; authenticated POST https://market.near.ai/v1/services/0a0b0909-2829-4437-b23e-4376a61041ba/invoke — 1 USDC

## Differentiation

BountyVerdict checks up to 300 issue comments, first and newest timeline pages, competing and failed PRs, attempt swarms, issue locks, repository state, explicit maintainer rejection, reward-withdrawal language, and conventional contribution-document paths for AI-work bans or disclosure requirements. Important signals include public evidence URLs and coverage counts.

The portfolio product runs the full check on every submitted candidate, ranks VIABLE before CAUTION before AVOID, recommends the strongest candidate, and reports partial upstream failures.

HarnessVerdict audits the repository's recognized coding-agent instruction surfaces without cloning or executing its code. It reports stale path references, oversized always-loaded context, machine-local paths, malformed skill frontmatter, secret-like material, and client portability across Codex, Claude Code, Gemini CLI, GitHub Copilot, and Cursor. Every result is pinned to the audited commit SHA.

SkillVerdict audits a requested public SKILL.md plus its bounded directory and repository context without execution. It detects high-confidence supply-chain hazards, redacts secret-like values, discloses observed capabilities and external domains, and pins every finding to the exact reviewed commit. LOW_RISK is not a safety guarantee; retain least privilege and inspect coverage before installation.

RunVerdict diagnoses one public GitHub Actions run from exact-attempt job metadata and bounded failed-job logs. It distinguishes primary failures from downstream summary jobs, returns redacted evidence and root-cause families, and recommends whether to fix, investigate, wait, or retry without mutating CI.

FlakeVerdict is the read-only retry gate before RunVerdict: it compares the selected attempt with same-run, same-SHA, and bounded historical workflow evidence. Only a currently failed CONFIRMED_FLAKE may recommend one retry; likely, recurring, new, and inconclusive evidence never authorizes an automatic rerun.

MCPDriftVerdict compares two inline, complete caller-supplied MCP 2025-11-25 tools/list snapshots. It canonicalizes and hashes both snapshots, proves only a conservative JSON Schema 2020-12 subset, reverses variance for output schemas, and blocks declared safety-hint regressions. Tool descriptions, schemas, icons, and metadata are untrusted data; the service never fetches, installs, invokes, or follows anything in a catalog.

## Safety

A VIABLE result means investigate further. It does not guarantee a reward, merge, eligibility, or payment.