skanehira1126

refinery-maintenance

Knowledge Refinery vaultの定期棚卸し、品質監査、指定したナレッジ・metadata・taxonomyの修復に使う。通常の開発レビューや単発の保存確認、repo設定診断は対象外。

skanehira1126 0 Updated 1mo ago

Resources

1
GitHub

Install

npx skillscat add skanehira1126/knowledge-refinery/refinery-maintenance

Install via the SkillsCat registry.

SKILL.md

Refinery maintenance

Read the shared operating rules for authorization,
blocked operations, and completion. Input is the requested vault, project set, or records and
any reported defects. An audit-only or proposal-only request returns findings and repair candidates;
a repair request also authorizes supported factual fixes that preserve conclusions.

Choose the review scope

  • Vault audit or periodic reconciliation: start with refinery_validate and review project
    metadata, evidence, knowledge quality, and taxonomy across the requested scope.
  • Targeted review or repair: inspect the specified records and the sources needed to assess
    them. Choose only the relevant checks below. A factual metadata or tag-description correction
    does not require a survey of unrelated projects, memories, or tags.
  • Schema or provenance defects: use refinery_validate to establish the errors. This tool always
    scans the active vault; it has no project or record filter. Report unrelated errors separately
    without treating them as permission for additional repairs.

Vault-wide administrative validation and project listing can run independently of a repository.
Repo-scoped knowledge or tag access requires the shared access check; never switch repositories
to bypass an opt-out.

Check the affected knowledge

  • For project metadata, inspect current values and propose partial factual updates. Keep
    purpose/domain tags lowercase kebab-case, technology names in technologies, and local paths
    or unsupported guesses out of metadata. Preserve the project ID.
  • For knowledge quality, search current project memory together with shared memory, then current
    project experiences. Choose cross-project IDs with refinery_list_projects and pass project_ids;
    use all_projects: true only for a necessary vault-wide review, never together with project_ids.
    Inspect relevant low-confidence records, duplicates, conflicts, stale principles, and source
    experiences. A moved or missing evidence reference is a limitation, not grounds to delete a conclusion.
  • For taxonomy, read knowledge-tags.md and inspect the affected branches.

Repair and finish

Apply authorized factual metadata or taxonomy fixes with the current revision and read them back.
Use refinery-experience for experience content updates and refinery-memory for memory creation
or revision; those skills own record format, evidence, conflict handling, and promotion requirements.
A general maintenance request does not approve a specific shared principle, deletion, or material rewrite.

After schema or provenance repair, or writes made during a vault audit, run refinery_validate
once for the batch. A targeted factual update normally finishes after readback; validate more broadly
only if a concrete integrity concern remains. For a read-only audit, the initial validation is
sufficient unless the state changed or an unresolved concern requires another check.

Review the affected central-vault Git diff, including new files, after writes. Without vault Git,
review the changed files and report that limitation. Report unresolved paths and reasons; unrelated
pre-existing errors do not block completion of an otherwise verified targeted repair. Make no
product-repository changes as part of vault maintenance.

Categories