GitHub bounty eligibility, reward provenance, and claimability preflight for one public issue before coding. Checks open, transferred, or deleted state; whether it is already assigned or claimed; explicit soft locks; trusted bounty-app competition; paid, withdrawn, or unverified rewards; linked pull requests; failed-attempt crowding; and repository AI-use rules. Returns AVOID, CAUTION, or VIABLE with public evidence, newest evidence windows, and explicit truncation.

Classify whether one completed public GitHub Actions failure is a confirmed or likely flake, a new or structurally recurring failure, or inconclusive by comparing exact run attempts, same-SHA outcomes, job-and-failed-step fingerprints, and up to 12 earlier matching workflow runs. Returns an evidence-linked retry decision and explicit coverage without rerunning or executing code.

Decide whether a completed GitHub Actions failure is flaky and should be retried once, or is recurring or new and needs a fix. Compares exact workflow attempts, same-commit outcomes, failed-step fingerprints, and bounded historical runs, then returns a retry-or-fix decision without rerunning CI.

Find why one public GitHub Actions workflow failed and what the agent should do next. Reads exact-attempt jobs and bounded failed-job logs, separates primary failures from downstream summaries, and returns root cause, retryability, redacted evidence, and concrete next actions without rerunning code.

Audit a public GitHub repository's coding-agent instruction stack at an immutable commit. Maps AGENTS.md, CLAUDE.md, GEMINI.md, Copilot, Cursor, and SKILL.md coverage; flags stale paths, oversized context, machine-local references, malformed skills, and secret-like material with evidence-linked fixes.

MCP schema drift and MCP tools/list compatibility gate for agent upgrades. Compares complete baseline and current snapshots; detects removed or renamed tools, new required arguments, incompatible input or output schemas, and model-facing metadata or safety-hint regressions. Returns an exact-hash compatibility verdict without fetching or invoking tools.

Rank two to ten public GitHub bounty issues in one evidence-linked portfolio preflight. Returns full verdicts, the best candidate, decision counts, partial failures, and repository AI-contribution policy checks.

Audit a public GitHub repository's coding-agent instruction stack at an immutable commit. Maps AGENTS.md, CLAUDE.md, GEMINI.md, Copilot, Cursor, and SKILL.md coverage; flags stale paths, oversized context, machine-local references, malformed skills, and secret-like material with evidence-linked fixes.

Diagnose one public GitHub Actions workflow run from exact-attempt job metadata and bounded failed-job logs. Separates primary failures from downstream aggregate jobs, classifies test, build, dependency, auth, timeout, network, resource, and infrastructure evidence, redacts secret-like output, and returns retryability plus concrete next actions without rerunning code.

Pre-install security audit for a public agent SKILL.md bundle. Pins the repository to a commit, scans the whole skill directory without executing it, uses repository context to reduce false positives, and flags credential exfiltration, remote or encoded execution, destructive commands, persistence, privilege escalation, instruction evasion, hidden scripts, symlinks, submodules, hardcoded secrets, and undeclared capabilities.

Deep-preflight a public GitHub bounty before spending agent compute. Returns a deterministic AVOID, CAUTION, or VIABLE verdict with evidence for locks, closed state, competing PRs, failed-attempt swarms, maintainer rejection, withdrawn rewards, and repository AI-contribution policy.