--- name: bountyverdict-agent-production.mimirslab.workers.dev description: bountyverdict-agent-production.mimirslab.workers.dev provides static analysis and classification services for public GitHub resources: bounty issues, Actions workflow runs, and third-party SKILL.md bundles. Each skill performs a single-shot evaluation and returns a structured verdict with evidence-linked summaries. All capabilities require public GitHub URLs and do not support streaming or real-time monitoring. host: bountyverdict-agent-production.mimirslab.workers.dev --- # bountyverdict-agent-production.mimirslab.workers.dev BountyVerdict is a decision-support host for agents operating in GitHub-centric workflows. It serves agents that need to evaluate bounty issue viability before committing development effort, diagnose CI failures before deciding whether to retry or escalate, and vet third-party skills before installation. It covers three distinct problem domains — bounty triage, CI diagnosis, and skill safety — unified by a common pattern of static, single-shot analysis returning structured verdicts. ## When to use this host Use this host when an agent operates exclusively with public GitHub resources and needs structured, single-shot verdicts for bounty triage, CI failure diagnosis, or pre-installation skill auditing. Do not use it for private GitHub repositories — none of the skills support private URLs. Do not use it for real-time or streaming CI monitoring; it is strictly a completed-run analyzer. Do not use it for non-GitHub issue trackers (Jira, Linear, GitLab) or non-GitHub CI systems (CircleCI, GitLab CI). For a single known bounty issue, prefer check-github-bounty-verdict over rank-github-bounty-portfolio. For runtime behavioral analysis of already-installed skills, this host has no capability — seek a dynamic sandbox or runtime monitoring tool instead. ## Capabilities ### Bounty Issue Triage Evaluates public GitHub bounty issues for eligibility, claimability, and relative priority, enabling agents to decide which issues are worth pursuing before writing any code. - **`check-github-bounty-verdict`** — Checks a public GitHub issue for bounty eligibility and claimability, returning AVOID, CAUTION, or VIABLE with a score and evidence-linked summary. - **`rank-github-bounty-portfolio`** — Runs due-diligence checks on two to ten public GitHub bounty issues, ranks them, and returns per-candidate verdicts plus the strongest non-AVOID recommendation. ### GitHub Actions CI Diagnosis Analyzes completed public GitHub Actions workflow runs to determine root cause, retryability, and whether a failure is flaky or genuine, guiding retry-or-escalate decisions. - **`diagnose-github-actions-run`** — Analyzes a public GitHub Actions workflow run URL and returns a structured verdict (PASS/WAIT/RETRY/FIX/INVESTIGATE) with root cause, retryability, and concrete next actions. - **`classify-github-actions-flake`** — Classifies a completed GitHub Actions workflow failure as flaky (retry once) or a real failure (needs a fix) by comparing attempt history, same-commit outcomes, and up to 12 prior comparable runs. ### Third-Party Skill Safety Audit Statically scans a public GitHub SKILL.md bundle at a pinned commit and returns a risk verdict before installation, blocking unsafe skills from entering an agent's toolchain. - **`audit-agent-skill-verdict`** — Statically scans a public GitHub SKILL.md bundle at a pinned commit and returns a LOW_RISK, REVIEW, or BLOCK verdict with a risk score and redacted findings. ## Workflows ### Bounty Portfolio Selection and Deep Vetting *Use when an agent needs to compare multiple bounty candidates and then confirm the top pick is still viable before committing to a solution.* 1. **`rank-github-bounty-portfolio`** — Submit two to ten candidate bounty URLs to get a ranked list and the strongest non-AVOID recommendation. 2. **`check-github-bounty-verdict`** — Run a single-issue deep check on the top-ranked recommendation to confirm its score, evidence, and claimability before starting work. ### CI Failure Triage and Retry Decision *Use when an agent encounters a failed GitHub Actions run and needs to determine both the root cause and whether the failure pattern is flaky before deciding to retry or escalate.* 1. **`diagnose-github-actions-run`** — Analyze the failed run URL to get a structured verdict, root cause, and retryability assessment. 2. **`classify-github-actions-flake`** — If the diagnosis verdict is ambiguous or suggests RETRY, classify the specific attempt against prior run history to confirm whether the failure is flaky or a real regression. ## Skill reference ### `classify-github-actions-flake` **GitHub Actions Flake Classifier** — Classifies a completed GitHub Actions workflow failure as flaky (retry once) or a real failure (needs a fix) by comparing attempt history, same-commit outcomes, and up to 12 prior comparable runs. *Use when:* Use when an agent or CI pipeline encounters a completed GitHub Actions workflow failure and needs to decide whether to retry the run once or escalate it for a code fix, before spending a retry attempt. *Not for:* Do not use for private GitHub repositories (only public workflow run URLs are supported). Not suitable for streaming or real-time CI monitoring — this is a single-shot classification per completed run attempt. **Inputs:** - `run_url` (string, required) — Canonical public GitHub Actions workflow run URL matching the pattern https://github.com/{owner}/{repo}/actions/runs/{run_id}. - `attempt` (integer) — Exact positive run attempt number to classify; omit to classify the current (latest) attempt. **Returns:** Returns a FlakeVerdict 1.0 object with a verdict enum (e.g. CONFIRMED_FLAKE), a plain-English summary of the evidence, and service_reuse metadata indicating result freshness. **Example:** `GET https://bountyverdict-agent-production.mimirslab.workers.dev/api/flake?run_url=https://github.com/actions/runner/actions/runs/29423388605&attempt=1` --- ### `check-github-bounty-verdict` **Bounty Verdict API** — Checks a public GitHub issue for bounty eligibility and claimability, returning AVOID, CAUTION, or VIABLE with a score and evidence-linked summary. *Use when:* Use when an agent needs to evaluate whether a specific public GitHub bounty issue is still open, unclaimed, and worth pursuing before investing time in coding a solution. *Not for:* Do not use for private GitHub repositories or non-GitHub issue trackers. Not a substitute for real-time GitHub webhooks or streaming issue activity monitoring. **Inputs:** - `issue_url` (string, required) — Canonical public GitHub issue URL matching the pattern https://github.com/{owner}/{repo}/issues/{number}. **Returns:** Returns a verdict of AVOID, CAUTION, or VIABLE with a 0–100 score, a summary of evidence, and service reuse metadata describing freshness and reliability. **Example:** `GET https://bountyverdict-agent-production.mimirslab.workers.dev/api/verdict?issue_url=https://github.com/typeorm/typeorm/issues/3357` --- ### `rank-github-bounty-portfolio` **BountyVerdict Portfolio Ranker** — Runs due-diligence checks on two to ten public GitHub bounty issues, ranks them, and returns per-candidate verdicts plus the strongest non-AVOID recommendation. *Use when:* Use when an agent must compare and rank two to ten public GitHub bounty issue URLs to identify the best candidate to pursue, including partial failure reporting for any issues that could not be checked. *Not for:* Do not use for a single bounty issue — use a single-issue verdict endpoint instead. Not suitable for private GitHub repositories or non-GitHub issue trackers. **Inputs:** - `issue_urls` (array, required) — Array of 2–10 unique public GitHub issue URLs to rank. Each URL must match the pattern https://github.com/{owner}/{repo}/issues/{number}. **Returns:** Returns a ranked portfolio object with product name, version, a top-level recommendation for the best bounty candidate, and service_reuse metadata describing reliability and guidance. **Example:** `{"issue_urls": ["https://github.com/acme/widget/issues/12", "https://github.com/typeorm/typeorm/issues/3357"]}` --- ### `diagnose-github-actions-run` **BountyVerdict Run — GitHub Actions Diagnosis** — Analyzes a public GitHub Actions workflow run URL and returns a structured verdict (PASS/WAIT/RETRY/FIX/INVESTIGATE) with root cause, retryability, and concrete next actions. *Use when:* Use when an agent needs to determine why a specific public GitHub Actions workflow run failed, whether it is retryable, and what concrete remediation steps to take next, without rerunning the workflow. *Not for:* Do not use for private GitHub Actions runs (requires public URL), for streaming CI logs, or for diagnosing non-GitHub CI systems such as CircleCI or GitLab CI. **Inputs:** - `run_url` (string, required) — Canonical public GitHub Actions workflow run URL matching the pattern https://github.com///actions/runs/. **Returns:** Returns a RunVerdict object with a verdict enum (e.g. FIX), a plain-English summary of the root cause, a retryability assessment, and a service_reuse metadata block. **Example:** `GET https://bountyverdict-agent-production.mimirslab.workers.dev/api/run?run_url=https://github.com/openai/codex/actions/runs/29728148711` --- ### `audit-agent-skill-verdict` **Bounty Verdict Skill Auditor** — Statically scans a public GitHub SKILL.md bundle at a pinned commit and returns a LOW_RISK, REVIEW, or BLOCK verdict with a risk score and redacted findings. *Use when:* Use when an agent needs to evaluate whether a third-party public SKILL.md is safe to install, specifically when no current audit exists for the exact commit and skill path, or when either has changed since the last audit. *Not for:* Do not use for auditing private repositories, non-GitHub repositories, or for runtime behavioral analysis of already-installed skills; this is a static pre-installation scan only. **Inputs:** - `repo_url` (string, required) — Canonical public GitHub repository URL matching pattern ^https://github\.com/[A-Za-z0-9-]+/[A-Za-z0-9._-]+(?:\.git)?$ - `skill_path` (string, required) — Repository-relative skill directory or exact SKILL.md path matching pattern ^[A-Za-z0-9._/-]+$ **Returns:** Returns a commit-pinned audit object with product='SkillVerdict', a verdict of LOW_RISK/REVIEW/BLOCK, an integer risk_score 0–100, a summary of findings, and service_reuse metadata. **Example:** `GET https://bountyverdict-agent-production.mimirslab.workers.dev/api/skill?repo_url=https://github.com/coinbase/agentic-wallet-skills&skill_path=skills/agentic-wallet` ---