archive-experiencelisted
Install: claude install-skill acupof-ai/arle
# archive-experience
Calibrated workflow for retiring `docs/experience/{wins,errors}/` entries into
the frozen archive. Modeled on deepseek-harness's `dsh-archive-agent-notes`
skill: archival is a seal, not a deletion, and the decision criterion is
future decision value, not age or word count.
## When to run
- The weekly hygiene sweep (see CHANGELOG section of AGENTS.md).
- `scripts/check_repo_hygiene.py` warns that wins/errors approach the cap.
The cap is a hard gate, so when it approaches, archival is forced — but
the *selection* is still future-value-ranked, not blind oldest-first.
- An entry is fully superseded by a newer one on the same topic.
## Classification (the decision)
Judge every candidate semantically. Age and zero inbound references are
discovery aids, never the criterion.
- **Archive** when the shipped decision is complete and the body is unlikely
to guide future work: one-off bench snapshots whose numbers are superseded,
a narrow workaround for a since-fixed bug, process history whose current
behavior is obvious from the code, an entry fully superseded by a newer
one on the same axis.
- **Keep** when the entry still carries future leverage: a binding constraint
or roofline number still quoted in planning, a negative result that
prevents re-litigating a tempting path, a reintroduction condition, a
measured bound a future optimization must beat.
- **Supersession check when adding an entry**: every new wins/errors entry
triggers a scope