hikari9/auto-office

Directly invocable Auto Office v3 Agy CLI mechanics primitive. Use when a selected/explicit Agy harness dispatch needs exact launch flag ordering, dynamic model discovery, model-dependent effort handling, workspace pinning, timeout/quota-stall diagnosis, prompt transport, recovery, or Agy-specific failure attribution. Do not treat this primitive as a separate lifecycle office or proof of adapter promotion.

Register or change an Auto Office harness adapter (one per CLI) and its catalog model rows with office harness scaffold|validate|smoke|list and office model add|list|disable. Covers the 3.1+ checklist, what each permission/trust flag grants, the Herdr interactive-form requirement, and why nothing in this flow promotes trust.

Internal Auto Office closeout spoke for 3.x runs. Use when a run's work is accepted and landed (or ready to hand off) and it is time to run office close: confirm the target run, verify the gate, respect the PR merge boundary, update the docs the repo keeps, sync the local base branch and remove the run's worktrees, close loops including panes, and end with the office close done report line. Never claims success from model confidence alone.

Auto Office 3.0 reference spoke, not loaded by 3.1 runs (they use the office CLI). Internal Auto Office v3 execution spoke. Use after an accepted plan to validate execution packets, acquire mutable role/write-scope ownership, dispatch executors or workers through the selected harness adapter, enforce protected paths and blast-radius limits, handle takeover/version checks, self-review mutations, and return durable evidence without widening scope.

Auto Office 3.0 reference spoke, not loaded by 3.1 runs (they use the office CLI). Internal Auto Office v3 intake spoke. Use when the planner runs the interactive requirements interview before freezing intent — covering the twelve-item floor in one or two batched question rounds (or a structured-question tool, per harness) and deriving the five frozen fields from the answers. This spoke is planner-owned and requires talking to the user directly. Do not use to ask about gear, which is declared, not interviewed.

Auto Office 3.0 reference spoke, not loaded by 3.1 runs (they use the office CLI). Internal Auto Office v3 loop-driver spoke. Use after plan approval to dispatch waves without blocking, integrate dispatch branches through a validated merge, adjudicate contract disagreements, and hold the autonomy ceiling — proceeding end to end while never merging to main without an explicit user statement.

Auto Office 3.0 reference spoke, not loaded by 3.1 runs (they use the office CLI). Internal Auto Office v3 maintenance spoke. Use at the beginning or closeout of an Auto Office invocation to label eligible historical outcomes lazily, refresh stale model catalogs outside route-time, update local harness bindings, maintain hot/warm/dreamt evidence tiers, compute maturity/decay, and prepare replay inputs without requiring an always-running daemon.

Lean, autonomous engineering orchestration for GitHub issues, bugs, features, and multi-agent implementation. Use when invoked as /auto-office-lean, when the user explicitly selects lean /auto-office, or when asked to run engineering work end to end without the legacy Office runtime. Follow the Auto Office manifesto, route fit-for-purpose agents, independently verify meaningful integrated work, and land only within human-authorized boundaries.

Auto Office 3.0 reference spoke, not loaded by 3.1 runs (they use the office CLI). Internal Auto Office v3 planning spoke. Use when the auto-office orchestrator needs to run interactive requirements discovery directly with the user, freeze the five execution fields, resolve product decisions, classify playbook/blast radius/size, produce or refresh a serialized implementation plan, or revise a plan after an accepted PLAN DEFECT. Do not use as a separate lifecycle or to change frozen requirements silently after freeze.

Auto Office 3.0 reference spoke, not loaded by 3.1 runs (they use the office CLI). Internal Auto Office v3 independent review spoke. Use for plan review, code review, accepted-material/minor/rejected/pending finding disposition, pointed blast-radius and bypass analysis, repeated review rounds, or validating PLAN DEFECT and BRIEF DEFECT exits.

Auto Office routing reference. Use when selecting or explaining a role route across harness, model, and effort; reading an executor route slate (primary + fallbacks); evaluating trust, capabilities, floors, quota, expected cost to success, speed, learned evidence, preference or exploration; or explaining a routing decision. The office runtime routes; this skill explains the scheme it applies (protocol/routing.md).

Issue-only Auto-Office bug observer. Use when explicitly invoked, when Auto-Office encounters its own runtime/skill/adapter/hook bug, and on EVERY office land or office close attempt (successful, failed, or refused). Activate with office self-improve; capture the entire run, including subagents and prior evidence. Delegate bounded, read-only investigation to a cheaper model, deduplicate, and file sanitized GitHub issues in Hikari9/auto-office. Never fix bugs, edit source, commit, open a PR, or close issues.

Auto Office 3.2 orchestrator mode for when the office runtime itself is the bottleneck (repeated submit refusals, plan re-review pausing live work, exhausted escalation, quota stalls). Only an explicit user request starts it; the orchestrator may only suggest it. The orchestrator stops driving work through office, composes one integration branch, fans the remaining work into file-disjoint Herdr lanes, keeps every Office invariant (independent review by a fresh agent session, mutation-verified tests, checks on the integrated tree, the user's merge boundary), and exits through office close --landed-externally.

Refresh missing Artificial Analysis routing benchmark scores for the bound Auto Office run through one low-cost background subagent. Use only when the user explicitly invokes /auto-update-benchmarks or directly asks to update the routing benchmarks. Never invoke from intake or on your own initiative.

Auto Office 3.0 reference spoke, not loaded by 3.1 runs (they use the office CLI). Internal Auto Office v3 verification spoke. Use to build or execute the verification floor for mutable work, choose targeted/regression/static/build/runtime checks, prove known-bad inputs fail critical gates, run browser acceptance flows for user-facing work, and package validation evidence for independent review and closeout.

Directly invocable Auto Office v3 Claude CLI mechanics primitive. Use when a selected/explicit Claude harness dispatch needs safe background/in-session launch behavior, prompt transport, worktree isolation, tool allowlisting, liveness, remote-control/resume handling, or Claude-specific failure attribution. Do not treat this primitive as a lifecycle office or proof of adapter promotion.

Directly invocable Auto Office v3 Codex CLI mechanics primitive. Use when a selected/explicit Codex harness dispatch needs safe model+effort invocation, worktree targeting, prompt transport, liveness/readback, resume safety, or Codex-specific failure attribution. Do not treat this primitive as a lifecycle office or as proof that the Codex adapter is proven.

Control Herdr, a terminal multiplexer for coding agents. Use only when the user explicitly mentions Herdr or asks to use Herdr to inspect or control panes, tabs, workspaces, commands, or another agent. Do not use merely because a task could benefit from a background terminal, delegation, or parallel work. Requires HERDR_ENV=1.

Sweep Herdr panes that this same orchestrating pane spawned and that have finished, using the shared auto-office pane ledger. Use only when the user explicitly asks to close/sweep/clean up done Herdr panes, or when the herdr skill's spawn recipe tells you to. Requires HERDR_ENV=1 and the herdr skill already loaded.

Directly invocable Auto Office v3 Hermes CLI mechanics primitive. Use when a selected/explicit Hermes harness dispatch needs prompt transport, model selection, working directory targeting, liveness detection, completion readback, or Hermes-specific failure attribution. Do not treat this primitive as a separate lifecycle office or proof of adapter promotion.

Finish an Auto Office executor task: behavior-preserving simplify pass, adversarial self-review of the diff, fix and mutation-prove medium+ findings, run the brief's checks, commit and push, office preflight, office submit, and end with one status line. Use when you are an Office executor (inside a task worktree with an Office brief) and your work is ready to submit, or when asked to submit, resubmit, or finish an Office task.