through-line

Work together without starting over in judgment.

Through-line is an agent skill for developing shared principles during everyday work and applying them in later sessions.

Experienced colleagues build an understanding of the business over time. They remember why apparently similar cases need different treatment and which shortcuts were temporary. That understanding informs decisions nobody thought to spell out in the next task.

A coding agent is a worker on its first day, every day. Each fresh session builds its understanding from the context available to it. Without the reasoning behind previous decisions, it has to infer intent from a prompt and the behavior of the code.

A workaround can then become the model for a new feature. The next session sees both implementations and treats the pattern as deliberate. Each change can carry an earlier guess further into the product, until you have to explain the same distinction again or undo behavior you never intended.

Through-line records those decisions and the reasoning behind them for later sessions. As new work exposes unresolved choices, you and the agent develop that understanding further.

How it works

During substantive work, the agent follows a small discovery index to relevant principles and their reasoning. Entries describe when a principle matters, so a renewal task can find a decision first made about booking. Existing documentation can remain the authoritative source.

When a consequential choice needs discussion, the agent recommends an approach and helps establish how far its reasoning applies. You decide whether it becomes a principle for future work. A principle earns a place by preventing a plausible future mistake or disagreement; most tasks add nothing.

The resulting record holds the commitment, reason, scope, and source decision. The agent applies it in later work and reopens the discussion when a new case challenges it. Changing a principle includes examining the work it affects.

Installation

Install with the skills CLI:

npx skills@latest add stuarth/through-line

Update with npx skills update through-line.

Add a workspace entry to AGENTS.md for Codex or CLAUDE.md for Claude Code, using the location of the workspace's principles:

## Through-line

Apply the through-line skill during substantive work.
Principles entry: principles/index.md.

The skill supports automatic invocation and the explicit commands $through-line in Codex and /through-line in Claude Code.

SKILL.md contains the operating instructions.