Task Context Decay in Terminal Workflows

Abstract

Developer task lists decay: items age into ambiguity because the context that made them actionable — branch, failing command, file, session state — is not captured with them. We formalize this as context decay and describe an experiment to measure it: instrumenting task capture with and without automatic context attachment, and measuring re-engagement time and task abandonment. This note pre-registers the design; no data has been collected yet.

What we investigated

Why do personal developer task lists rot? The common explanations are motivational (procrastination) or organizational (bad prioritization). We suspect a third mechanism: context decay — the task’s words survive, but the state that made it actionable does not, so re-engaging means re-deriving context from scratch. This note formalizes that hypothesis and specifies the experiment to test it.

Why it matters

If context decay is the dominant mechanism, the fix is not another prioritization scheme — it is capture-time context attachment, which is a tooling problem with a buildable answer. That is the design bet behind our Terminal Task Manager, and we would rather test the bet than assume it.

Method

Planned design, pre-registered here before any data collection:

Instrument an experimental CLI task tool to capture tasks in two modes, randomized per task: bare text, versus text plus automatic context (repository, branch, working directory, last command, recently touched files).
Measure for each task: time from re-open to first meaningful action (re-engagement time), completion versus abandonment at 14 and 30 days, and self-reported “did you remember what this meant?” on re-open.
Population: initially our own team (openly a convenience sample), then a small external cohort of volunteer developers if the effect looks real.

Findings

None yet. Data collection has not started. This section will report results when they exist — including a null result, which would be genuinely informative: it would suggest task decay is motivational after all, and that the tooling bet should be revisited.

Limitations

The initial population is our own team, who built the tool and want it to work; any internal effect must be treated as an upper bound. Randomizing context capture per task leaks (users know which mode they got). Re-engagement time is a proxy, not a direct measure of context loss. These constraints are accepted for a first pass whose job is to decide whether a proper external study is worth running.

Product implications

If the hypothesis holds, automatic context attachment becomes the core feature of the Terminal Task Manager rather than a nice-to-have, and the same principle — capture context at the moment of intent — feeds into how our agent systems record and resume work. If it fails, we will say so here.