跳至正文
OpenAgentSkill
浏览已开放 · 目录记录日期 2026-11-21 · 自动更新和账户服务暂时关闭。
其他技能

to-tickets

When asked to file tickets, break down approved work, or move a settled plan onto the board, turn the spec, closed decision map, or pressure-tested conversation into verifiable work items with real dependencies and non-ticket blockers.

timharris707 · 已有静态检查记录

安装与使用

运行命令前先阅读来源与依赖说明。复制命令不会自动安装或执行。

npx skills add timharris707/skills --skill to-tickets

技能说明


name: to-tickets description: "When asked to file tickets, break down approved work, or move a settled plan onto the board, turn the spec, closed decision map, or pressure-tested conversation into verifiable work items with real dependencies and non-ticket blockers."

To Tickets

Read the Codex desktop binding when this workflow needs harness mechanics, model routing, or recovery.

The pack's bridge from decided to takeable. decision-map produces decisions; orchestrate routes tracked items into lanes. This skill is what stands between them: it turns settled intent into work items a lane can claim without asking anyone what was meant.

Read the team-workflow binding doc first; it names the tracker, the label vocabulary, the verify commands, and the decider. The tracker discipline governs everything filed here.

Tracer bullets

Slice by tracer bullet: a thin path that goes all the way through, end to end, rather than a horizontal layer that goes nowhere alone. "Wire the read path from route to store, one field" is a tracer bullet. "Build the data layer" is a layer: it cannot be verified, cannot be demoed, and its acceptance criteria are always someone else's.

Three tests, all of which must pass:

  1. Verifiable alone. The item names how you would know it worked, using the repo's verify commands.
  2. PR-sized. One session, one branch, one review. An item nobody can finish in a sitting is a plan wearing a ticket's clothes.
  3. Ordered by what it unblocks, not by architectural tidiness. The first bullet should retire the most risk.

The slicing binds downstream: implement mandates that a lane's first verified checkpoint prove the thinnest end-to-end slice (tickets sliced tracer-style are built tracer-style), and the seams an item's verification names at filing time are where implement's test-first bar lands.

The wide-refactor exception

One slice shape legitimately fails the tracer-bullet tests: a wide refactor: a single mechanical change (rename a column, retype a shared symbol) whose blast radius fans across the whole codebase, so no vertical slice can land green alone. Sequence it as expand–contract instead of forcing it into a bullet: an expand item adds the new form beside the old so nothing breaks; migrate items move the call sites over in batches sized by blast radius (per package, per directory), each blocked by the expand, CI staying green batch to batch because the old form still exists; a contract item deletes the old form once no caller remains, blocked by every migrate batch. When even the batches cannot stay green alone, keep the sequence but give them a shared integration branch (recorded per orchestrate's integration-branch reference, the decider's approval of those slices counting as the opt-in) and have them all block a final integrate-and-verify item; green is promised only there, and the contract item adds that item to its blockers: the old form never comes out before the combined verification completes.

The sign-off gate

Reuse an exact breakdown already approved in the current request or recorded decision. For a new or materially changed breakdown, before Pass 1 files anything, put the slice list in front of the decider: each item's title, what it delivers end to end, and what blocks it, in plain English, no bodies yet. Ask whether the granularity is right, whether every edge is a real gate, and what should merge or split; iterate until they approve. The approval is recorded, not remembered: post the exact approved list as a comment on the plan-source or driving item before Pass 1, so what gets filed can be checked against what was approved. Filing first and asking after is the wrong order: a batch on the board is already colliding with other lanes' frontier queries while it is being argued about.

The two passes

Items need ids before they can reference each other, so filing is always two passes. Doing it in one produces edges pointing at numbers that do not exist yet.

Pass 1: file the bodies (the approved list, nothing else). Each item's body IS the spec, per the work-item spec template that setup seeds: destination, plan source, acceptance criteria, verification, out of scope. The plan source line is not optional: it links the primary source (the decision-map entry, the grilling verdict, the research findings) so review never relitigates a settled question. When items derive from an ingest evidence packet, also run the ingest.py link command the recommendation handed you for each item as you file it; the packet's retention lifetime is measured by those links.

Pass 2: wire the edges. Add native dependency edges for every blocker that is itself a tracker item, using the recipe's database-id form:

# <owner>/<repo> is the bound tracker repo, written literally — {owner}/{repo} would
# resolve implicitly and mis-target in forks:
gh api repos/<owner>/<repo>/issues/<blocker-number> --jq .id
gh api repos/<owner>/<repo>/issues/<child-number>/dependencies/blocked_by -F issue_id=<that-id>

Edges are authoritative wherever the blocker is a tracker item, so the frontier unblocks itself when the blocker closes. The blocked label is only for non-ticket blockers: a vendor gate, a scheduling constraint, a pending adjudication. An item blocked purely by edges must not also carry the label, or a stale label holds it blocked after its edges clear. A mixed-blocked item carries both: closing a ticket removes only that dependency; clearing a non-ticket blocker removes the label only when no non-ticket blockers remain. Neither event alone makes an item with the other blocker outstanding a

阅读完整来源 ↗