| name | memory-bank-writer |
| description | Use when updating files in memory-bank/ — adding new technical docs, refining domain notes, or maintaining tasks/work/ and tasks/log/ |
memory-bank-writer
Routing skill for writes to memory-bank/. Loads the inventory and the RAM-layer process so the agent knows what exists and how to maintain it.
Load:
memory-bank/project/doc-guide.md — required. Style, length, and hygiene rules for every memory-bank file.memory-bank/_index.md (the inventory — needed when authoring writes)memory-bank/tasks/process.mddocs/what-goes-where.md (if this project ships the standard's docs alongside)
Rules of thumb:
- Follow
doc-guide.md on every write. Length budgets, no duplication, bullets over paragraphs, reference don't copy.
- Decision rationale lives in two places: inline
**Why:** in technical/<topic>.md (standing rule) and chronologically in tasks/log/pr-*.md (the moment). If a log entry's rationale recurs or becomes a coding rule, follow tasks/process.md → promote durable knowledge to technical/. Don't create a decisions/ directory.
- Knowledge layer files (
project/, technical/) change slowly. Update only when the underlying truth has shifted.
- When you add or rename a memory-bank file, update
_index.md first, then update any integration-layer artefact (skill, rule, hook, AGENTS.md) that should reference the new file.
tasks/work/<slug>.md: update frontmatter (status, branch, updated) as state changes. Don't fabricate scratchpads that won't be useful.tasks/log/pr-<num>-<slug>.md: typically written by automation on PR merge. Scale length to complexity (3–6 / 8–12 / 15–20 lines). Don't document routine quality checks as achievements.
No knowledge in this file — see the listed files.
