twjohnwu
UserOrchestration framework for Claude Code — 9 Middle-earth agent roles with pinned model/effort/tools, institution dispatch rules, a 3-lens adversarial review panel, opt-in guard hooks, and an opt-in STDD spec-to-code pipeline.
Categories
Indexed Skills (13)
tlor-init
Installer ritual — sets up or upgrades the tlor-orchestration framework (agents, rules, CLAUDE.md/AGENTS.md routing, optional hooks). Run explicitly via `/tlor-init`.
tlor-restore
Restore-from-backup ritual — rolls back the tlor-orchestration installation to a backup created by `/tlor-init` or `install.sh`. Run explicitly via `/tlor-restore`.
erebor-ledger
Token/cost-savings ledger for tlor-orchestration dispatching — reports how much dispatching to tlor roles saved versus running the same work inline on the orchestrator model, split by Fable-5-orchestrator sessions vs Opus-orchestrator sessions. Triggers: "usage report", "省了多少 token/成本", "dispatch savings". Not for live cost estimation of a single in-progress dispatch — this is a retrospective report over existing Claude Code transcripts.
minas-tirith-archivist
Read-only search over archived decision records (general and project-scoped) — use when the user asks about past decisions, why a convention exists, "為什麼當初這樣做", "以前是怎麼決定的", "這題以前決過嗎", "查決策記錄", "search decision records", "past decisions", or explicit `/minas-tirith-archivist`. Read-only query counterpart to `/westmarch-scribe`; answers with citations, never writes or edits records.
rivendell-council
Multi-lens adversarial review — for major conclusions, architecture decisions, root-cause verdicts, or security judgments, dispatch the three opposition lenses (elf-archer, orc-saboteur, hobbit-gardener) in parallel to independently attack the conclusion; adopt it only if a majority survives. Triggers: "adversarial review", "多方抗辯", "抗辯", or any high-stakes judgment — irreversible operations, contract/schema changes, anything touching money or precision, architecture decisions, root-cause verdicts, production-affecting conclusions. Not for trivial edits or pure Q&A.
westmarch-scribe
Archives a DECIDED compact-MADR record (Decision / Decision Drivers / Considered Options / Pros and Cons / Decision Outcome) to the right durable location — the rules customize layer for a cross-project decision, or a project decision log / instruction file for a project-scoped one. Use when a decision just got finalized or reversed in conversation and needs to be archived — the user says "就這樣定了", "決定用 X", "把這個決策存起來", "記錄這個決策", "archive this decision", "record this decision", "we've decided", explicit `/westmarch-scribe`, or a stdd phase skill's closing advisory suggests archiving.
stdd-execute
STDD execute phase. Runs the per-task RED → GREEN → REFACTOR TDD loop against an approved STDD tasks.md, using a two-dispatch model (builder-RED, builder-GREEN+REFACTOR) with an independent verifier and a test-file fingerprint passed through the dispatch prompt. Triggers: "stdd-execute", "run the TDD loop task by task", "run RED GREEN REFACTOR", or any request to implement STDD tasks one at a time. Requires an approved spec.md and an existing tasks.md; refuses otherwise.
stdd-explore
STDD explore phase. A thinking-partner mode that clarifies a vague feature idea BEFORE any spec gets written: challenges assumptions with a six-phase first-principles method, asks batched confirm-then-ask questions, proposes 2-3 tradeoff options, and hands off to stdd-uiux or stdd-spec. Writes no code or files except a rejected-options note. Triggers: "/stdd-explore", starting a new STDD change from a rough idea.
stdd-lint
STDD mechanical checker. A pure rule-based (non-model-judgment) checker that runs the 13 checks catalogued in references/checklist.md (placeholder/coverage/fingerprint/cross-artifact checks) against a single STDD change. Triggers: "/stdd-lint", or any stdd-* skill's boundary/coverage call.
stdd-plan
Generate condition-based design artifacts (design-be.md / design-fe.md / api.yml) and a scenario-covered tasks.md from an approved STDD spec.md. Triggers: "stdd-plan", "generate design and task list", "produce a plan from spec", or any request to turn an approved GWT spec into a design + task list. Requires spec.md status: approved; refuses to run otherwise.
stdd-spec
STDD spec phase. Writes a GWT-format spec.md with stable REQ/S IDs, test-mapping and verification-command fields per scenario, runs a mechanical /stdd-lint self-review plus an adversarial-panel approval gate, and is the hard gate that blocks stdd-plan/stdd-execute until the spec (and design-ux.md, if any) are approved and their content fingerprints match. Triggers: "/stdd-spec", writing or approving a spec for an STDD change.
stdd-uiux
STDD UI/UX design phase. Conditional stage inserted between stdd-explore and stdd-spec only when a change has a user-facing UI surface: gathers design references, generates design-ux.md (MASTER + per-page/flow override structure), runs a mechanical anti-pattern self-review, and owns the reflow protocol when a later stage finds a UX defect or conflict. Triggers: "/stdd-uiux", any STDD change with a screen/interaction surface.
stdd
STDD status phase. A read-only status dashboard for an STDD change: reads existing artifacts (spec.md, design-ux.md, tasks.md), reports which stage the change is in, re-verifies the two-file fingerprint on a read-only basis, and suggests the next command. Never invokes any other stdd-* skill and never writes to any file. Triggers: "/stdd", "what stage is this change in", checking STDD progress.
Bio shown is the top-scored skill's repo description as a fallback — real GitHub bios land in a future update.