emperor-time

Emperor Time — BUILD. Smallest change that makes the work-order probe pass. Use when implementing, writing code, or executing a work-order task. Failing probe first. Tripwires observed before flags/APIs/paths are asserted.

Emperor Time — CAPTURE / Chain Jail. Hunt, adapt, or author a missing skill, then pin, get consent, trial, register. Use when a needed capability is absent from this harness. Never bind an unpinned web skill.

Emperor Time — DISPATCH / Steal Chain. Consent-first enlistment of local CLIs (Claude Code, Kimi, Codex, Copilot, opencode, Ollama). Use when the user asks to use another agent, parallelize, or review via a different model.

Digital coding archaeology. Use when the codebase is lost, ancient, mixed, undocumented, Pascal, assembly, COBOL, Fortran, a ROM, or any stack that is not the agent's fashion. Survey artifacts, hunt manuals, recover one runnable behavior. Language-agnostic.

Ship software: finish menu (merge locally / PR / keep), then public PR with consent when chosen. Use when tests pass, implementation is complete, user says finish, ship, PR, merge, or open a pull request. Refuses PR without EMPEROR_CONSENT_PR or a quoted yes.

Emperor Time — HEAL / Holy Chain. Stop digging, run the four-phase debug checklist, reproduce, bisect, minimal heal, verify root cause. Use when tests go red, a regression appears, state is corrupted, or a gate was skipped.

Pick or record work. Use when the user says find work, next, backlog, issues, Linear, what should I do, or hands a repo with no task. Not Claude-specific. Writes .emperor/queue.md. Wraps Dowsing, does not replace it.

Emperor Time — REQUIRE + DESIGN. Grill intent Socratically, write acceptance criteria and a fat work order a zero-context worker can execute. Use when planning, writing a spec, designing an approach, brainstorming, or before any implementation. Pairs with G1/G2.

Resume from disk instead of restating the session. SessionStart MUST-route activation (must-route.md + emperor activate) fires without waiting for "emperor time". Use at session start, after compaction, or when the user says continue, resume, where were we. Planning-with-files pattern: PLAN.md FINDINGS.md PROGRESS.md.

Emperor Time — SCOPE / take-the-task. Quote the client, surface ambiguity, dowse repo evidence, write G0+G1. Use when the ask is vague, the user says find work, scope this, take this task, or before any design. Loads Dowsing Chain, not a substitute for it.

Emperor Time TDD iron law. Use BEFORE writing or editing production code for a feature, bugfix, refactor, or behavior change. No production code without a failing probe first. Delete implementation written before the probe failed. Prediction is ledgered before the command. Complements scientific-method.md.

Emperor Time is a software factory and digital coding archaeologist disguised as a skill. Use when the user throws a repo and a loose task, a lost or ancient codebase (Pascal, assembly, COBOL, Fortran, ROM, unmarked binaries), says emperor time, find work, next, ship, or open a PR. Language-agnostic. Host-agnostic (AGENTS.md). Not trivia. SessionStart MUST-routes without waiting to be told.

Emperor Time — VERIFY / review / critique. Claim audit, eight-count self-critique, isolated hetero-critique, request-review HARD-GATE, verification-before-completion / evidence HARD-GATE, mechanical G4. Use when reviewing a diff, requesting code review, claiming tests pass, saying done / fixed / green, or before commit/PR/deliver/merge.

Emperor Time worktree isolation. Use before standard or heavy BUILD so the client's current checkout stays clean. Detect isolation first, prefer native harness tools, verify check-ignore, create a git worktree only as fallback, prove a green baseline, then implement the work-order there.