claude-history-ingest

What it does

Ingest Claude Code conversation history into the Obsidian wiki. It extracts knowledge from past Claude conversations, imports the ~/.claude folder, and distills it into wiki pages. It supports pre-extraction to reduce noise, then parses extracted conversations or raw JSONL, and can integrate memory files and audit logs to build topic-based wiki content per project. It can be invoked directly or via the router /wiki-history-ingest claude.

How it works

• Resolve configuration to obtain OBSIDIAN_VAULT_PATH and CLAUDE_HISTORY_PATH, reading manifest and index to understand existing content.
• Determine projects to process and skip any whose names match WIKI_SKIP_PROJECTS. Apply exclusion uniformly across all steps.
• Ingest in modes (Append by default, Full if requested):
  • If Append: process files not in manifest or newer than ingested_at, using canonical paths and a manifest helper to avoid re-ingesting
  • Pre-extraction (recommended): run scripts/extract-jsonl.py to generate compact JSON in ~/.claude/extracted/, reducing token usage
  • Parse conversations from ~/.claude/projects/*/*.jsonl preferentially using extracted JSONs if present; otherwise fall back to raw JSONL
  • If using desktop data: parse audit.jsonl for detailed tool calls and actions, combined with session metadata in local_<uuid>.json to provide context
  • Use memory files in memory/*.md as pre-distilled signals and propagate their types (user, feedback, project, reference) into the wiki
• Step 3b (desktop only): integrate audit logs with session metadata to contextualize actions and identify patterns
• Step 4: Cluster knowledge by topic across conversations, not one page per conversation
• Step 5: Distill into wiki pages per Claude project, decoding path-encoded project names to clean project names and organize the vault accordingly

When to use it

Triggers when the user asks to process Claude history or mentions the .claude folder, Claude projects, session data, past conversation logs, or audit logs. Also usable via the wiki-history-ingest router.

What it can touch

• Reads: ~/.claude/, ~/.claude/projects/*/memory/*.md, ~/.claude/projects/*/*.jsonl, ~/.claude/extracted/*/*.json, session metadata files, audit.jsonl in desktop sessions
• Writes: wiki pages under the vault corresponding to each Claude project; updates and creates topic-based pages and references

Caveats

• Requires configuration resolution to locate vault and Claude history paths
• Relies on manifest to determine delta; may re-scan if paths are canonicalized differently
• Pre-extraction reduces token usage but may skip some raw data unless fallback is used
• The skill describes grouping by topic rather than one page per conversation; volume may affect page granularity