One tick of a CI watchdog — check main-branch GHA status; spawn an agent to fix if failing and no investigation is in flight.
One tick of the henyey mainnet monitor — checks, metrics scan, deploy, status report
Continuous always-on orchestrator for the henyey project pipeline. Each pass queries the board, central-picks up to N actionable issues by priority, and fans out parallel foreground specialist sub-agents (triage/plan/do/review-pr) with the right Claude model per stage. Owns concurrency, conflict-avoidance, CI pipelining, sibling-nit consolidation, and a self-reflection pass that files pipeline-improvement issues. Replaces scripts/project-tick-loop.sh. Use when the operator asks to "run the loop", "process the board continuously", or "keep the pipeline going".
Single-pick dispatcher primitive for the henyey project pipeline. One tick = pick one actionable issue from the project board and dispatch the right specialist sub-agent for its current state. Used for manual single picks and `--issue=` recovery; the continuous orchestrator is `/project-loop`, which owns concurrency centrally. Use when the user asks to "run a tick", "pick up an issue", or "process one board item".
Run two parallel adversarial PR reviewers and combine their verdicts with external PR reviews (GH Copilot bot, humans, other bots) and CI state into a merge decision. Agent reviewers post structured comment verdicts (since the agent is the PR author and cannot self-approve via GH native review). External CHANGES_REQUESTED reviews block merge identically to agent CHANGES_REQUESTED. Operates on issues in `in-review`. Auto-merges with --admin on all-green (after filing follow-up issues for unaddressed inline review comments, so non-critical feedback is preserved as backlog instead of dropped); bounces to `ready-for-doing` on any request-changes or CI red; blocks after 3 bounce-backs on the same code state. At the lifetime cap (6 bounces since last `## Review: Reset`) the PR enters **force-converge mode**: if CI is green, the PR auto-merges and unresolved reviewer concerns become follow-up issues; only red/pending CI at the cap still blocks. Use when invoked by /project-tick with an issue in in-review, or manuall
Audit a Rust crate's adherence to its Stellar protocol spec (spec-driven, walks every normative claim)
Autonomous, hypothesis-driven optimizer for henyey's distributed max TPS, measured with the Supercluster MissionMaxTPSClassic (23-node tier-1) on a Namespace (nsc) cloud VM. Default mode is `frontier`: remove bottlenecks one proven, parity-safe hypothesis at a time until the raw simulated network (latency/throughput) is the demonstrated binding constraint — stellar-core's number is a reference floor, not a ceiling. Maintains a measurements/hypotheses document and opens a PR per accepted change. Distinct from `perf-optimize`, which tunes the local single-shot apply-load benchmark; this skill optimizes the *networked* max-TPS ceiling end-to-end.
Crate-scoped Rust cleanup audit and optionally apply, with parity-aware filtering and adversarial verification