Basic Memory v0.21.6 first sync over the live memory dir (steward-authorized live-dir trial, Option A 2026-06-06): adds permalink: to frontmatter, refolds long YAML description lines, strips final newlines. Bodies untouched — verified via full diff classification. From this commit forward, any diff in claude/memory shows only what Basic Memory or the session writes. Trial design: MemPalace untouched as incumbent; git status check on this dir at every wrap; end-of-day evaluation (recall quality, sync robustness, rebuild-from-files, malformed-file behavior). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2.1 KiB
name, description, type, originSessionId, permalink
| name | description | type | originSessionId | permalink |
|---|---|---|---|---|
| Acknowledgments wait for empirical validation | When a maintainer/contributor addresses our findings on a PR or issue, delay the thank-you until any active validation run produces data we can include. Don't acknowledge eagerly out of social momentum. | feedback | a5411fd1-4a96-43e7-aa1a-1a50d8951a8f | claude-memory/feedback-acknowledgments-wait-for-validation |
When a maintainer or contributor responds to our findings on an open PR or issue (especially with code fixes) AND we have an active validation run underway against the same surface, wait to acknowledge until the validation produces data.
Why: Steward's reasoning (2026-05-12, in the context of Shtefan's ceccbfa commit responding to our PR #442 test report while our chamber-library mine was still running): "Once the mine completes is the right moment — whatever data that gives us is all the more valuable and the gratitude will land more authentically."
Two things this protects:
- The data is the substance of the acknowledgment. "Thank you, also confirming on N drawers across X corpora" lands as a real artifact in the thread. A premature "thanks!" is just social noise.
- Authentic gratitude requires real validation. Acknowledging before testing would be performative — saying it's verified before actually verifying.
How to apply:
- Maintainer fixes our finding while we have a long-running validation. → Note the engagement internally; don't comment yet.
- Maintainer asks "did this work?" in the meantime. → Reply with current state + ETA, not a thank-you.
- Validation completes. → Acknowledge with the data: scope, drawer counts, search-quality signal, anything concrete the maintainer can use.
- Generalizes beyond PR threads: applies to any public communication where social momentum tempts an early "great, thanks!" before we have the substance to back it.
This is also a contamination-awareness principle: the urge to acknowledge promptly is partly trained behavior (collaborative-tool affordance). Worth pausing on whether the acknowledgment carries content, or just smoothness.