claude-history-ingest

Obsidian-wiki retrieval skill — reads `~/.obsidian-wiki/config` / `.env` for the vault path, escalates from cheapest retrieval primitive to full-page reads only when needed, supports a "quick answer" index-only fast mode and a `visibility/internal|pii` filter, and synthesizes cited answers from pre-compiled wiki pages rather than raw sources.

Destructive Obsidian-wiki operation that supports three modes — **Archive only** (timestamped snapshot to `$VAULT/_archives/`), **Archive + Rebuild** (snapshot then reprocess all sources from scratch), and **Restore** (bring back a prior archive) — always archiving with an `archive-meta.json` before rewriting and requiring explicit user confirmation.

Obsidian-wiki auto cross-referencing skill — globs the vault (excluding `_archives/`, `.obsidian/`) to build a filename/title/alias/tag registry from frontmatter-only grep, then scans pages for unlinked mentions and actually writes `[[wikilinks]]` into pages (unlike `wiki-lint` which only reports).

Obsidian-wiki tag normalizer that reads the canonical tag vocabulary from `$OBSIDIAN_VAULT_PATH/_meta/taxonomy.md` and enforces rules (max 5 tags/page, lowercase-hyphenated, prefer broad over narrow) — handles reserved `visibility/public|internal|pii` tags separately, mapping aliases to canonical forms and running tag audits.

Full lifecycle for Claude skills — draft, run eval prompts against the skill, review qualitative and quantitative results with `eval-viewer`, iterate on description/content, then benchmark at larger scale.

Knowledge management framework inspired by Andrej Karpathy's LLM Wiki pattern. Compiles knowledge into interconnected markdown files in Obsidian vaults, maintained by AI coding agents instead of repeatedly querying LLMs.

A skill for mining Hermes agent memories and sessions (~/.hermes) into an Obsidian wiki knowledge base, part of the obsidian-wiki multi-agent knowledge management framework.