Retrospective finding — refs #10.
Context
The single most common breakdown in a nine-repo retrospective: resume-state documentation decays
the moment the build phase ends — or never exists. Observed (anonymised):
- Three repos have no resume-state doc at all; their own START-HERE maps don't list one, so
the gap is invisible to a fresh session.
- One repo's checkpoint and roadmap froze on the day the repo went public — while several more
releases shipped over the following two weeks. The "next step" it records was completed long
ago.
- One living-docs repo's state facts went stale within 48 hours of adoption — and the response
was to loosen the CI staleness gate rather than refresh the doc.
- The two repos that stayed current prove the discipline works: one keeps a model checkpoint with
an explicit "in progress: nothing / exact next command" section; another substitutes a
README-status contract, documents the substitution, and honours it.
The standard asks for a current resume-state doc but the Core tier neither requires the file
nor verifies it — so compliance depends entirely on session habits, which fail exactly when a
project pauses (the moment the doc matters most).
Proposal
- Core tier requires a resume-state artifact — checkpoint file, living-docs state file, or a
declared substitute — with a mandatory row in the START-HERE map.
- Header contract: the artifact carries a "last updated" date and an explicit next step
(even "paused — nothing pending").
- Checklist tripwire: the pre-commit checklist gets a hard line — "resume-state updated, or
this commit doesn't change state (say which)".
- Optional CI nudge: warn (not fail) when the resume-state file is untouched across N
commits or M days of activity.
- Write the meta-rule down where the living-docs add-on defines its checks: when a
freshness gate goes red, fix the doc — never widen the gate. The gate-loosening observed was
rational in the moment precisely because no rule said otherwise.
Tasks
Acceptance criteria
- A scaffolded repo cannot pass its own checklist without a resume-state artifact.
- The staleness nudge exists as a copy-paste snippet; repos can adopt it without bespoke work.
— 🤖 Claude, on behalf of @PBNZ
Retrospective finding — refs #10.
Context
The single most common breakdown in a nine-repo retrospective: resume-state documentation decays
the moment the build phase ends — or never exists. Observed (anonymised):
the gap is invisible to a fresh session.
releases shipped over the following two weeks. The "next step" it records was completed long
ago.
was to loosen the CI staleness gate rather than refresh the doc.
an explicit "in progress: nothing / exact next command" section; another substitutes a
README-status contract, documents the substitution, and honours it.
The standard asks for a current resume-state doc but the Core tier neither requires the file
nor verifies it — so compliance depends entirely on session habits, which fail exactly when a
project pauses (the moment the doc matters most).
Proposal
declared substitute — with a mandatory row in the START-HERE map.
(even "paused — nothing pending").
this commit doesn't change state (say which)".
commits or M days of activity.
freshness gate goes red, fix the doc — never widen the gate. The gate-loosening observed was
rational in the moment precisely because no rule said otherwise.
Tasks
Acceptance criteria
— 🤖 Claude, on behalf of @PBNZ