Every tool in the kit.

Fetch PR feedback and failing CI, triage, fix, and push back.

Dispatches parallel read-only probes to triage each issue, fixes in an isolated managed worktree, verifies with project test gates, and pushes. Never force-pushes.

Release pipeline: pre-flight, tests, commit, push, PR.

Runs /ground-state, the project test suite, drafts a commit message, pushes, and opens a PR with a structured verification summary. Add --verify for adversarial checks.

Safe large-scale structural changes via DAG executor.

Enumerates affected sites, groups into dependency layers, applies changes in parallel per layer with worktree isolation, and verifies at each boundary. Exits on first regression.

Discover incidental complexity, duplication, and dead code.

Four parallel discovery lenses: clone detection, dead code, complexity hotspots, and wrong abstraction. Produces a ranked, behavior-preserving reduction plan.

Parallel root-cause analysis for bugs and failing tests.

Dispatches sub-agents to form and validate hypotheses in isolated worktrees. Each hypothesis tested independently, preventing cross-contamination.

Adversarial re-derivation of sub-agent claims before you act.

Parallel verifiers independently re-derive 2-3 key claims from scratch using tool calls only. Returns CONFIRMED / REFUTED / STALE / UNVERIFIABLE per claim.

Parallel reviewers across a diff or PR: security, correctness, tests, perf.

Dispatches dimension agents covering security, correctness, api-compat, test-coverage, and perf-observability. Synthesizes findings by severity into a merge recommendation.

Catches silent success before Done is declared.

Cross-references each completion claim against the tool-call ledger and a fresh postcondition read-back. Gates Done behind a BACKED / UNBACKED / UNVERIFIABLE verdict.

Ground capability claims with file:line citations.

Default mode answers meta-capability questions with path:line evidence. Runtime-wiring mode traces actual execution paths and returns CONFIRMED / REFUTED per claim.

Two parallel sub-agents: external context + local codebase.

Dispatches one agent for web/API research and another for local file and git exploration. Results merged into a grounded context brief.

Map a module or subsystem before editing.

Two parallel agents: one maps structure (exports, interfaces, dependencies) and one maps test coverage. Eliminates sequential 3+ file reads.

Transform a loose idea into a structured, actionable spec.

Takes an idea, feature request, or problem description and produces a scoped specification ready for /mint or direct implementation.

Transform a sequential plan into parallel wave execution.

A planning agent restructures the current approach into orchestrated waves of parallel sub-agents for maximum throughput.

Three parallel critics generate alternatives before you commit.

Pragmatist, paranoid, and architect lenses each invent one alternative. A synthesis step ranks all options. External work gets a steelman critic too.

End-of-session survey and priming prompt for the next run.

Read-only sub-agent surveys recent work and drafts a vaguely-specific orientation prompt. Candidate directions framed as questions, not decisions.

Recon wave before implementation: git, infra, and memory.

Triangulates git state, project infrastructure, and prior-session memory. Assembles a verified grounding preamble pasted into every subsequent sub-agent brief.

Evaluator-generator loop to iteratively refine any artifact.

Locks criteria in wave 1, isolates the evaluator from generation history to prevent sycophancy. Emits the best version with remaining gaps flagged at cap.

Set up a scheduled headless run with Telegram summaries.

Creates a recurring afk daemon task on a cron schedule with push-notified results to Telegram.