linear-hierarchy-guard
v1.0.0
Published
Claude Code / agent skill: review Linear issues before promotion, enforce the Issue/Project/Milestone/Initiative hierarchy, and preserve deadline integrity so work doesn't sprawl into a million unrelated tickets.
Maintainers
Readme
linear-hierarchy-guard
A Claude Code / agent skill for reviewing a Linear issue or spec before an agent starts executing it, and deciding whether it should stay a single issue or be promoted into a project, milestone set, or initiative — so a team ends up with a clean, correctly-nested hierarchy instead of a million loose, unrelated issues.
What it does
When an agent is handed a Linear issue, PRD/spec, /wayfinder map, /to-spec
output, or /to-tickets output, this skill:
- Fetches the live Linear state for the issue (status, due date, cycle, project, initiative, parents, blockers, comments, etc.) instead of trusting only the prompt text.
- Classifies the work using a Promotion Decision Matrix: keep as issue, convert to a parent/control issue, promote to a project, add milestones, or attach to an initiative.
- Runs a Deadline Integrity Algorithm so that promoting an issue into a bigger structure never silently loosens its original due date. Every derived project, milestone, ticket, or cycle assignment must inherit the seed issue's deadline unless a human explicitly approves changing it.
- Requires breadcrumbs: every promoted project/issue links back to the original seed issue, and the seed issue gets a clear final disposition (kept, converted to control issue, closed as promoted, or retained as intake).
- Refuses to fabricate dates, duplicate existing structure, or mark anything "Done" purely from a planning/restructuring pass.
The core failure mode this prevents: a ticket due July 15 gets "promoted" into a new project whose milestones/tickets end up dated July 20 (or undated), quietly breaking the original commitment.
Install
From the root of your project:
npx linear-hierarchy-guardThis copies SKILL.md into .claude/skills/linear-hierarchy-guard/ in the
current directory. Restart Claude Code (or reload skills) afterward.
No install script, no postinstall hooks — it just copies two files into
place. Feel free to read bin/install.js before running it.
Usage
Invoke it before /to-tickets, before assigning a coding agent to a large
issue, or whenever an issue starts to look like it's actually a workstream.
See SKILL.md for the full decision matrix, deadline algorithm, templates,
and write procedure.
License
MIT
